(EW-201349)

Central Plant Optimization for Waste Energy

Reduction (CPOWER)
December 2016
This document has been cleared for public release;
Distribution Statement A
 Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0188

The public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data
sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other
aspect of this collection of information, including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information
Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other
provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number.
PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To)
01/12/2016 Cost and Performance Report January 2014 - December 2016
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Central Plant Optimization for Waste Energy Reduction (CPOWER) W912HQ-13-C-0058
5b. GRANT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

Girija Parthasarathy EW-201349
Honeywell 5e. TASK NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

Honeywell - 1985 Douglas Drive North, Golden Valley, MN 55422 REPORT NUMBER

ERDC CERL - 3909 Halls Ferry Rd, Vicksburg, MS 39180

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR'S ACRONYM(S)

Environmental Security Technology Certification Program Program Office ESTCP
4800 Mark Center Drive
Suite 17D03 11. SPONSOR/MONITOR'S REPORT
Alexandria, VA 22350-3605 NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution is unlimited.

13. SUPPLEMENTARY NOTES

N/A

14. ABSTRACT
Central plants contain multiple chiller, boiler, and auxiliary equipment. Each piece of equipment operates on different
efficiency curves that vary with part load, ambient conditions, and other operating parameters. In addition, the site receives
real-time price signals for electricity, and operators must consider fluctuating fuel prices and other costs. The system-level,
dynamic optimization of central plants and distribution system implemented in this project has the potential to save energy
and cost. The objective of the project was to assess the energy and economic benefits of the real-time optimization
technology that commands all equipment in a central plant.
15. SUBJECT TERMS
Central Plant Optimization for Waste Energy Reduction (CPOWER), Building Automation System (BAS), Building Energy
Management System (BMS), real-time optimization technology, energy conservation, energy cost reduction.

16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF RESPONSIBLE PERSON
ABSTRACT OF
a. REPORT b. ABSTRACT c. THIS PAGE Girija Parthasarathy
PAGES
19b. TELEPHONE NUMBER (Include area code)
Unclassified Unclassified UU UL 57 763-954-6554
Standard Form 298 (Rev. 8/98)
Prescribed by ANSI Std. Z39.18
Page Intentionally Left Blank
This report was prepared under contract to the Department of Defense Environmental
Security Technology Certification Program (ESTCP). The publication of this report
does not indicate endorsement by the Department of Defense, nor should the contents
be construed as reflecting the official policy or position of the Department of Defense.
Reference herein to any specific commercial product, process, or service by trade
name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply
its endorsement, recommendation, or favoring by the Department of Defense.
Page Intentionally Left Blank
 COST & PERFORMANCE REPORT
Project: EW-201349

TABLE OF CONTENTS
Page

EXECUTIVE SUMMARY ...................................................................................................... ES-1

1.0 INTRODUCTION ................................................................................................................. 1
1.1 BACKGROUND .......................................................................................................... 1
1.2 OBJECTIVES OF THE DEMONSTRATION ............................................................. 1
1.3 REGULATORY DRIVERS ......................................................................................... 2
2.0 TECHNOLOGY DESCRIPTION ......................................................................................... 3
2.1 TECHNOLOGY/METHODOLOGY OVERVIEW..................................................... 3
2.1.1 Optimization Solution .................................................................................................. 3
2.1.2 Model Library .............................................................................................................. 4
2.1.3 Problem Formulation and Solver ................................................................................. 5
2.1.4 Optimization Hierarchy ................................................................................................ 5
2.1.5 Solution Architecture ................................................................................................... 5
2.1.6 Inputs and Outputs ....................................................................................................... 7
2.1.7 Inputs ............................................................................................................................ 7
2.1.8 Outputs ......................................................................................................................... 7
2.1.9 Chronological Summary............................................................................................... 8
2.2 ADVANTAGES AND LIMITATIONS OF THE TECHNOLOGY/
METHODOLOGY ....................................................................................................... 9
3.0 PERFORMANCE OBJECTIVES ....................................................................................... 11
4.0 SITE DESCRIPTION .......................................................................................................... 13
4.1 FACILITY/SITE LOCATION AND OPERATIONS ................................................ 13
4.2 FACILITY/SITE CONDITIONS ............................................................................... 14
5.0 TEST DESIGN .................................................................................................................... 15
5.1 CONCEPTUAL TEST DESIGN ................................................................................ 15
5.2 BASELINE CHARACTERIZATION ........................................................................ 16
5.3 DESIGN AND LAYOUT OF TECHNOLOGY COMPONENTS ............................ 20
5.4 OPERATIONAL TESTING ....................................................................................... 21
5.5 SAMPLING PROTOCOL .......................................................................................... 21
5.6 SAMPLING RESULTS .............................................................................................. 22
6.0 PERFORMANCE ASSESSMENT ..................................................................................... 25
PO1: SIMULATED OPTIMIZER SOFTWARE PERFORMANCE .................................. 25
PO2: OPTIMIZER SOFTWARE INTERCONNECTION WITH CONTROL SYSTEM.. 25
PO3 ENERGY SAVINGS AND PO5 ECONOMIC PERFORMANCE ............................ 25
PO4: COMFORT CONDITIONS IN BUILDINGS ............................................................ 27

i
 TABLE OF CONTENTS (Continued)

Page

PO6: EQUIPMENT SHORT CYCLING ............................................................................ 27

PO7: EFFECTIVENESS OF UI (QUALITATIVE)............................................................ 28
7.0 COST ASSESSMENT ......................................................................................................... 29
7.1 COST MODEL ........................................................................................................... 29
7.2 COST DRIVERS ........................................................................................................ 30
7.3 COST ANALYSIS AND COMPARISON ................................................................. 30
8.0 IMPLEMENTATION ISSUES ........................................................................................... 33
9.0 REFERENCES .................................................................................................................... 37
APPENDIX A POINTS OF CONTACT............................................................................... A-1

ii
 LIST OF FIGURES
Page

Figure 1. Technology Overview ..................................................................................................... 3

Figure 2. Optimization Implementation.......................................................................................... 4
Figure 3. System Architecture ........................................................................................................ 6
Figure 4. Central Plant Optimization Modules ............................................................................... 6
Figure 5. Location of Cooling and Heating Plants for the Demonstration ................................... 13
Figure 6. Location of Plants and Areas Served............................................................................. 14
Figure 7. Energy Consumption (24-hr periods) ............................................................................ 17
Figure 8. Evaluation of Models and Inputs (2015-16 data) – Color Map Representation of RMSEs
....................................................................................................................................................... 19
Figure 9. Actual and Model Comparison ...................................................................................... 19
Figure 10. Chiller Plant Schematic ............................................................................................... 20
Figure 11. System Integration and Controls ................................................................................. 20
Figure 12. Heating Plant Operational Data ................................................................................... 22
Figure 13. Supply Temperature Changing .................................................................................... 23
Figure 14. Chiller Plant Optimizer Enabled Periods .................................................................... 23
Figure 15. Chiller Plant Supply and Return Temperatures ........................................................... 24
Figure 16. Total Power Consumed by the Chiller Plant ............................................................... 24

iii
 LIST OF TABLES
Page

Table 1. Performance Objectives .................................................................................................. 11

Table 2. Regression Models and Variables for All Data (July 2015 – May 2016)....................... 18
Table 3. Testing Chronology ........................................................................................................ 21
Table 4. ON Duration Statistics .................................................................................................... 27
Table 5. OFF Duration Statistics .................................................................................................. 28
Table 6. Cost Model...................................................................................................................... 29
Table 7. Summary Cost Analysis for a Chiller Plant .................................................................... 31

iv
 ACRONYMS AND ABBREVIATIONS
ACS Automation and Control Solutions (A business unit of Honeywell)
ASHRAE American Society of Heating, Refrigeration and Air Conditioning Engineers

BACnet® A Data Communication Protocol for Building Automation and Control

Networks
BAS Building Automation System
BMS Building Energy Management System
BLCC Building Life-Cycle Cost

CHP Combined Heat and Power

CPOWER Central Plant Optimization for Waster Energy Reduction (title of this project)

DCS Distributed Control System

DoD U.S. Department of Defense
DPW Directorate of Public Works

EBI Enterprise Buildings Integrator (a Honeywell BMS)

EO Executive Order
ERDC-CERL U.S. Army Engineer Research and Development Center’s Construction
Engineering Research Laboratory
ESTCP Environmental Security Technology Certification Program

FEMP Federal Energy Management Program

FY fiscal year

GHG Greenhouse Gas

GPM or gpm Gallons per minute

HBS Honeywell Building Solutions (business unit in Honeywell ACS)

hr hour(s)
HTS Honeywell Technology Solutions

IAE Integral Average Error

I/O Input/Output
IPMVP International Performance Measurement and Verification Protocol

LonWorks® A networking protocol and platform for control applications

Mcf 1000 cubic feet

MMBtu 1,000,000 British thermal unit(s)
MMBH MMBtu/hr

PO Performance Objective

v
RMSE Root Mean Squared Error

UI User Interface

vi
 ACKNOWLEDGEMENTS

The following individuals’ contributions to the demonstration project are gratefully

acknowledged: From Honeywell: Keith Johnson, John Schlesinger, Bruce Skubon, Bill
Klingenschmidt, Benson Wei, Nick Tong, Andrew Chen, Benny Dong, Pedro Davalos, Rebecca
Kemp, and Janet Myers; from University of California, Berkeley: Sergey Vichik, Francisco
Borelli, and Jason Kong; from U. S. Army Engineer Research and Development Center’s
Construction Engineering Research Laboratory (ERDC-CERL): Matt Swanson and Noah
Garfinkle. We would like to thank Mr. Coby Jones, formerly the Energy Manager at Fort Bragg
Directorate of Public Works (DPW), for his support of the project. We thank the Environmental
Security Technology Certification Program (ESTCP) for the funding provided, and program
guidance and support; we would like to especially thank the current and former program managers,
technical advisors, and support staff at ESTCP: Mr. Scott Clark, Mr. Glen DeWillie, Mr. Peter
Knowles, Ms. Sarah Medepalli, Mr. Colin Dunn, Dr. Jim Galvin, and Mr. Tim Tetreault.

vii
Page Intentionally Left Blank

viii
 EXECUTIVE SUMMARY
OBJECTIVES OF THE DEMONSTRATION

Central plants contain multiple chiller, boiler, and auxiliary equipment. Each piece of equipment
operates on different efficiency curves that vary with part load, ambient conditions, and other
operating parameters. In addition, the site receives real-time price signals for electricity, and
operators must consider fluctuating fuel prices and other costs. The system-level, dynamic
optimization of central plants and distribution system implemented in this project has the potential
to save energy and cost. The objective of the project was to assess the energy and economic
benefits of the real-time optimization technology that commands all equipment in a central plant.
The performance objectives were: (1) correct optimizer performance in simulation (objective met),
(2) optimizer software interconnection with plant control system (objective met), (3) 10% energy
savings (objective not met), (4) preserving comfort conditions in buildings (objective met), (5)
economic performance (objective not met), (6) low short cycling of equipment (objective met),
and (7) effective user interface (UI) (qualitative) (objective partially met).

TECHNOLOGY DESCRIPTION

The deployed technology is a model-predictive, run-time optimization technology used to operate

the generation, storage, and distribution of cooling and heating energy while maintaining building
comfort. Based on the inputs of upcoming loads, price signals, central plant performance models,
and building response, a mixed-integer evolutionary optimizer algorithm solves the schedules and
setpoints for the major and auxiliary equipment in the central plant. The central plant model is
configured from a library containing models of chillers, boilers, cooling towers, pumps, and
thermal storage system. A dynamic building model mathematically represents the changes in
comfort conditions in the building in response to changes in energy supplied with the distributed
chilled or hot water. The models are set up based on historical data and updated as new data become
available. The optimal control commands are communicated to lower level controllers that operate
the equipment in the central plant. Feedback from the buildings provides corrections to the long-
term forecast load; this feedback is used to adjust supplied energy.

The demonstration was designed to collect data about the original control and the optimized
control, alternately. A software switch incorporated in the optimizer enabled the optimizer to run
the plant or switch to the original control operation to ensure similar occupancy and functions for
the buildings served by the central plant. Operational electricity and gas consumption data from
all equipment were collected in the optimizer’s database for analysis.

DEMONSTRATION RESULTS

A simulation system and models of the central plants and loads to test the optimizer software in
simulation. The simulation software was connected with the optimizer software using OPC server
protocol. We ran several simulations with the setup to test and fix optimizer software bugs and
validate its performance before deploying on site.

The optimization solution was integrated with the chiller plant control system. A systematic and
thorough testing and commissioning process was followed to bring the optimizer online.

ES-1
Observations and later analysis showed that the optimizer’s outputs were appropriate, as is
expected for energy use minimizing actions.

After training the operators, site resource manager, and other site personnel, and providing the
appropriate user manuals, the optimizer was handed over to site staff. Honeywell Laboratories in
Minneapolis, MN, continued providing remote phone and onsite support for running the plant
under optimizer control. The optimizer software was available and connected at the chiller plant
from April 2015, to May 2016, and was enabled to operate the plant for some periods during that
time. Data from the chiller plant is available for July 2015—May 2016. After removing invalid
and shorter duration data, the data analysis shows the optimizer operated onsite for 39 days (24-
hour [hr] periods) in several continuous periods. During the same period, the data shows 164
periods of original control days.

A rigorous baseline characterization methodology was developed to compare the actual energy
consumption during optimization with the expected energy consumption under original control
operation. Using all the data available, it was found that the optimized control of the plant did not
reduce the energy consumption in the plant, and in most cases it is within one standard deviation
error of the expected usage with original control. This unexpected result led to further analysis to
diagnose the problems. The analysis showed a number of discrepancies in the input data to the
optimizer software, which are explained in detail in the performance assessment section.

The optimizer works on real-time-sensed data to know the state of the plant, forecast loads, and
calculate optimal operating commands. Poor quality or incorrect sensed data will not result in
optimal outputs. The analysis of the data showed that there were no adverse effects to comfort
conditions in buildings. The analysis also showed that equipment short-cycling, although more
frequent than in original control, was still within guidelines provided by the site and able to be
adjusted with user provided parameters. The effectiveness of the UI and the optimizer software
architecture had mixed results.

IMPLEMENTATION ISSUES

Running the optimizer during the demonstration period depended on the chiller plant equipment
being in good operating condition (e.g., not experiencing maintenance issues forcing manual
operation) and the availability of site staff to monitor the operation periodically since optimizer-
controlled operation is a large departure from current practice. Several troubleshooting periods
took place in which the software was updated to manage site expectations and the difficult
transition from Research and Development (R&D) to production prototype.

Failure to achieve energy and cost savings during the demonstration period stemmed from the
following causes:

• incorrect inputs to the optimizer that were caused by communication disruptions or

incorrect configuration changes;
• the complexity and prototype nature of the software required monitoring and support from
skilled application engineers, but U.S. Department of Defense (DoD) site restrictions
prevented remote access to the workstation;

ES-2
 • data-driven plant equipment models were potentially not well learned due to varied
problems experienced by the optimizer preventing it from operating stably for longer
periods with all equipment components;
• the transition of complex software from R&D to production prototype required the team to
develop additional software tools and training of staff;
• the software’s architecture and the implementation scheme to control the full plant from
the supervisory layer caused two problems:
1. potential network communication problems required development of additional
optimizer software safety measures to prevent unsafe operation, and
2. site staff were uncomfortable with a supervisory-level algorithm controlling lower level
components in real time.

The results and lessons learned are planned to be published as part of a book (on intelligent control
systems) as well as in a white paper.

ES-3
Page Intentionally Left Blank

ES-4
1.0 INTRODUCTION
1.1 BACKGROUND

Many of DoD fixed installations receive usable energy in the form of heating and cooling via
central plants. These plants are excellent candidates for improvements in operational efficiencies
because of their aggregation of energy production and distribution and their impact on the energy
use profile of a military installation. Honeywell’s predictive, automated optimization for central
plants has significant potential to cost-effectively reduce energy consumption and costs by
choosing the right operating points for all equipment, considering pricing and several other factors,
in real time.

DoD central plants currently do not use automated optimization. Discussions with experienced
central plant operators and energy managers about current operations make it clear that an
opportunity exists for capturing efficiency savings from operational optimization. Central plants
are currently operated to reliably meet all demands, and not necessarily for fuel economy or energy
efficiency. Plant operators run the equipment according to a pre-set, fixed strategy. However, plant
equipment efficiencies vary with load and external conditions such as ambient temperature. In
addition, central plants have multiple chillers, boilers, and power generation equipment, which
may differ from each other in capacities and performance curves. The ability to select the most
efficient equipment for a load would offer great benefits.

The technology deployed is a model-predictive, run-time optimization technology to operate the

generation, storage, and distribution of cooling and heating energy, while maintaining building
comfort. Based on the inputs of upcoming loads, price signals, central plant performance models,
and building response, a mixed-integer evolutionary optimizer algorithm solves the schedules and
setpoints for the major and auxiliary equipment in the central plant. The central plant model is
configured from a library containing the models of chillers, boilers, cooling towers, pumps, and a
thermal storage system. A dynamic building model mathematically represents the changes in
comfort conditions in the building in response to changes in energy supplied with the distributed
chilled or hot water. The models are set up based on historical data and updated as new data become
available. The optimal control commands are communicated to lower level controllers that operate
the equipment in the central plant. Feedback from the buildings provides corrections to the long-
term forecast load that are used to adjust the energy supplied.

This demonstration was designed to collect data about the original control and the optimized
control, alternately. A software switch was incorporated in the optimizer, which enabled the
optimizer to run the plant or switch to the original control operation. This ensured similar
occupancy and functions for the buildings served by the central plants. All operational data
including electricity and gas consumption from all equipment were collected in the optimizer’s
database for analysis.

1.2 OBJECTIVES OF THE DEMONSTRATION

The performance objectives of the demonstration were:

(1) Correct optimizer performance in simulation: objective met

1
 (2) Optimizer software interconnection with plant control system: objective met

(3) 10% energy savings: objective not met

(4) Preserving comfort conditions in buildings: objective met

(5) Economic performance (net present value >=0): objective not met

(6) Low short cycling of equipment: objective met

(7) Effective user interface (UI) (qualitative): objective partially met.

1.3 REGULATORY DRIVERS

Executive Order (EO) 13514 (now replaced by EO 13693): EO 13514 set requirements for
improving federal government efficiency by decreasing fossil fuel dependence. EO 13693
provides goals to maintain Federal leadership in sustainability and greenhouse gas (GHG)
emissions reductions; specifically the goal to promote building energy conservation, efficiency
and management by reducing building energy intensity by 2.5% annually through end of fiscal
year (FY) 2025.

EO 13423: Section 2. (a) requires improved energy efficiency and reduced GHG emissions of the
agency, through reduction of energy intensity by (i) 3% annually through the end of FY 2015, or
(ii) 30% by the end of FY 2015, relative to the baseline of the agency’s energy use in FY 2003.

DoD Policy: DoD’s Strategic Sustainability Performance Plan [2] sets out DoD’s priority to invest
in reducing energy from traditional sources (Energy Management in Fixed Installations), sets a
target to reduce Scope 1 and 2 GHG emissions by 34% between FY 2008 and FY 2020.

EO 13327: Section 3.b.ii. prioritizes actions to be taken to improve the operations and financial
management of the agency’s real property inventory.

2
2.0 TECHNOLOGY DESCRIPTION
2.1 TECHNOLOGY/METHODOLOGY OVERVIEW

The Central Plant Optimization for Waste Energy Reduction (CPOWER) central plant
optimization solution, illustrated in Figure 1, provides optimal schedules and operating points for
all equipment in the plant. It relies on equipment performance models, forecasted load, a building
model, and energy price information. The equipment and building models are set up based on
historical data and updated as new data become available. The optimization is based on minimizing
energy costs or maximizing efficiency, and uses an evolutionary algorithm.

Figure 1. Technology Overview

2.1.1 Optimization Solution

The optimization solution in this project dynamically generates schedules and setpoints for plant
equipment that minimize operating cost over a specific period. The solution concept is illustrated
in Figure 1. The dynamic optimizer block shown in the center of the figure interacts with the
equipment performance models, the specific central plant layout, building model, forecasted load,
and external inputs such as electricity pricing. The optimal schedule and setpoints are
communicated to the controllers.

3
The online information flow is conceptualized in Figure 2. A demand forecaster predicts loads for
the next 24-hour (hr) period of optimization based on the current weather, load history data, and
occupancy criteria. The central plant model is configured from a library containing the models of
chillers, boilers, cooling towers, pumps, and thermal storage system. A dynamic building model
mathematically represents the changes in comfort conditions in the building in response to changes
in energy supplied with the distributed chilled or hot water. Based on the inputs of upcoming
demand loads, central plant performance, and building response, the optimizer solves the schedules
and setpoints for the major equipment in the supply and distribution of chilled and hot water. The
optimal schedules and setpoints are used by the plant controller to operate the central plant.
Feedback from the buildings provides corrections to the long-term forecast load that are used to
adjust energy supplied and the setpoints.

Figure 2. Optimization Implementation

2.1.2 Model Library

The model library is an integral part of the optimization solution that contains models to simulate
the performance of the central plant and the building response under given conditions. These
models are developed for a specific plant and building based on the data the optimizer collects
when connected to the Building Energy Management System (BMS). Most of the models are either
regression trees or a collection of regression trees. They are learned using historical data and are
periodically updated. The solver can determine the optimal solution from various candidate
solutions based on the plant performance. Since the optimizer models are based on data, they are
continuously updated and, therefore, do not lose their efficacy when the equipment deteriorates.
The models also provide the basis for performance monitoring of the plant. Separate models for
each type of equipment are built based on regression tree principles and using several influencing
factors as inputs, such as weather conditions, flow rates, and temperatures.

4
2.1.3 Problem Formulation and Solver

To search for the optimal schedule, the optimization problem is formulated with the following
objective function and multiple constraints over an optimization horizon of h time steps:
h
Min∑ (Cost t +αPenalty t )
t =1

subject to several constraints of equipment capacities, minimum outputs, ramp rates, interval
between startup and shutdown, and others.

Cost t is the total energy cost of the central plant during the time interval t and is the sum of

energy costs of all central plant equipment, determined from their models.
Penalty t represents
shortage of supply versus demand. α is a weight specified according to user preference for energy
saving ( α takes a bigger value) and comfort of occupants ( α takes a smaller value).

The above optimization problem is further parametrized and solved to find an optimal solution for
both discrete (i.e., ON/OFF) and continuous (i.e., setpoints) variables. This culmination of the
modeling and optimization results in the entire system working in the most efficient manner.

2.1.4 Optimization Hierarchy

The optimization problem is solved in two levels: the energy source dispatch between the thermal
energy storage and the chillers occurs first; the run-time optimization of the chillers, associated
pumps, and cooling towers occurs in the next level.

2.1.5 Solution Architecture

Figure 3 shows the system architecture, illustrating the interaction of the optimization layer with
respect to the central plant control system. Sensors and controllers are usually linked to
Input/Output (I/O) modules to send and receive data in a uniform format through standard
communication protocols such as LonWorks® or BACNet®. The data interface of the
optimization module can communicate with these I/O modules, controllers, or building automation
systems (BAS) using standard protocols. In the case of CPOWER at Ft. Bragg, the optimization
software interfaces only with the existing BAS for ease of implementation and to standardize on
one type of interface. The optimization module directly controls plant equipment.

5
 Figure 3. System Architecture

Figure 4 shows the software modules in CPOWER. The UI accepts user inputs and displays
relevant information. A data interface reads data (temperature, flow rate, power, etc.) and sends
control commands and settings (ON/OFF, temperature setpoint, pump speed, etc.) to all relevant
devices. A database saves data that needs to be archived and shared. The model library contains
simulation models of plant, building, and load forecast. The solver module solves for the optimum
schedules and setpoints based on the problem formulated. The fault detector monitors for alarms
or availability of chiller plant devices.

Figure 4. Central Plant Optimization Modules

6
2.1.6 Inputs and Outputs

System inputs can be categorized into five types:

1. Device Information
2. Connection Information
3. Ambient Conditions
4. Tariff Model
5. Running Settings

The outputs can be categorized as control commands, running settings, and supervisory
information about the chiller plant. Major and typical items are described below.

2.1.7 Inputs

Device Information
The device information includes all basic properties of chiller plant devices (chiller, boiler, cooling
tower, pump, etc.). Most of the design information is available from design documents or product
specifications. Most of the running data can be read from sensors already installed to the chillers
or the chiller plant.

Connection Information
The connection information describes how the water or piping system connects parts of a chiller
plant together. Multiple connection matrices are employed to indicate which primary pumps can
supply how much chilled water for a specified chiller, and which cooling water pumps can supply
how much cooling water for a specified chiller or cooling tower.

Ambient Conditions
The ambient conditions include representative indoor and outdoor air temperature and humidity,
which are averaged or given weighted averages from multiple sensors.

Tariff Data
The tariff data contains time-dependent price of electricity or fuel.

Running Settings
The system’s running settings include maintenance schedule (when a specified chiller or pump
will be offline in the near future for some maintenance work or overhaul), time settings of the
chiller plant (e.g., when building working hours, which days are working days), temperature
settings (the target indoor air temperature, allowed range of return/supply water temperature, etc.),
and user preference for energy saving or human comfort.

2.1.8 Outputs

The number of outputs is relatively small. For a chiller, the control commands are Open/Close chilled
water valve and cooling water valve (if applicable), chiller ON/OFF, and sometimes, the chiller working
mode (cooling or heating); the running settings may include chilled water temperature setpoint.

7
For a boiler, the control commands are hot water valve Open/Close and boiler ON/OFF; the running
setting is the hot water temperature setpoint. For a pump, the control command is ON/OFF and its
running setting is mainly the flow rate, or if it is a variable speed pump, the frequency. Although the
intelligent control system will monitor running status of the whole chiller plant, it will send commands
or settings only to devices that the user chooses for system control.

The inputs and outputs specific to the plants in this demonstration are shown in Figure 8 and Figure
9 in Section 5.3 (Design and Layout of Technology Components).

2.1.9 Chronological Summary

The central plant optimizer is the result of several years of investment by Honeywell. No DoD
funds were used in the development of the basic technology. Honeywell has been developing a
suite of optimization and control technologies that target the energy supply, distribution, and
demand. The first prototype was implemented at a Honeywell office building in Shanghai, China,
in 2010. Several other prototypes of the solution were implemented in China between 2010 and
2013, including a hotel and office building (40,000 square meters), NanJing subway station chiller
plant, and a chiller plant at an electronics manufacturing plant.

Summary of Development Under the Project

For the demonstration project, the optimizer was configured for specific conditions in the plants
and developed the approaches and features for the thermal energy storage tank and the heat
exchanger scheduling.

Heat Exchanger
The controls were developed in the software for automatically starting and stopping the heat
exchanger, based on the site protocol.

Thermal Energy Storage Tank

Site-specific optimal operating strategies were developed for the chilled water storage tank.

Simulation Models for Testing

Simulation models of the central plants and building loads were developed as part of the project.
The simulation models were used for testing the optimizer software prior to deployment to rectify
operational issues, parameter configuration issues, and other unforeseen conditions.

Expected Application
The technology is deployable at all central plants across DoD sites. As an example, there are 13
central plants in Ft. Bragg and 6 in Ft. Jackson, which indicate enormous energy and cost savings
potential. Information from Construction Engineering Research Laboratory (CERL) indicates that
there are 155 heating plants in the Army installations alone. The number of cooling plants,
combined heat and cooling (CHP), and heating plants at all DoD sites numbers in the hundreds.
The optimization technology has the potential to be applied to much of these central plants.

8
Although it was demonstrated at a central plant, the optimization technology is applicable to chiller
and boiler plants in buildings as well, and is therefore applicable to decentralized cooling and
heating plants at DoD sites. However, because of the challenges a prototype faces in field
installation, it is highly recommended that sites upgrade to the infrastructure and data capabilities
needed by a data intensive application.

2.2 ADVANTAGES AND LIMITATIONS OF THE TECHNOLOGY/

METHODOLOGY

Advantages
The central plant optimization offers automated energy and cost savings without human
intervention. It produces optimum operation outputs and directly supplies the control commands
to the plant thus ensuring that the commands are followed.

With the automated optimization technology, the plant operational parameters are continuously
calculated by using measured values; state-of-the-art alternatives use manufacturer-provided fixed
specifications. For optimization over a time horizon, a long-term load prediction is used, and short
term corrections are added for deviation from the forecast. This allows the operation to take
advantage of the thermal storage effect of buildings. The optimizer also considers real time pricing
for optimum scheduling; whereas for existing systems, the operational logic must be pre-
determined for real time pricing levels for the day.

Limitations
If the plant is not well instrumented and automated, additional sensors and meters and
communication must be added, which can increase the cost. Another limitation is that, if the
optimizer is not properly configured, there is risk of equipment switching frequently to save
energy, thus increasing maintenance costs. This limitation can be overcome by several means in
the software. In the current version of the prototype software architecture, the optimizer also
commands the sequence of low level control, which is not ideal for plant control. Delineating the
optimizer and local control functions can overcome this limitation. The operational logic is not
transparent to the operator, which can reduce acceptance, especially if the operator is not trained
to understand it, which may lead to the optimizer not being used at all. The recommendation for
future versions is to provide explanatory comments on the UI for optimizer actions.

9
Page Intentionally Left Blank

10
3.0 PERFORMANCE OBJECTIVES
Table 1 describes the project’s performance objectives (PO) and summarizes the results.

Table 1. Performance Objectives

Data
Metric Success Criteria Results
Requirements
Quantitative Performance Objectives
PO1: Simulated Optimizer software performance
Optimizer output of Simulated (not site Optimizer outputs are The software performance met the
plant operating data) optimizer within normal range of objectives in simulation.
schedules and outputs of equipment operation for equipment
setpoints (various schedules and >95% of the time
units) setpoints
PO2: Optimizer software interconnection with control system
Comparison of Optimizer outputs All required optimizer The software interconnection
optimizer output and and control outputs are transmitted objectives were met.
control system commands for the as control commands for
commands same period plant operation.
PO3: Energy savings
Difference in plant Electricity and gas >10% savings on The optimizer was commissioned
energy consumption consumption at the weather normalized successfully; however, post-data
between baseline and central plants, prices, energy consumption data analysis revealed incorrect inputs
demonstration plant outputs, into optimizer. Most of the demon-
periods in units of weather stration period was consumed by
kilowatt hour (kWh) troubleshooting configuration and
(cooling plant) and control interconnections; hence
one million British energy savings were not achieved
Thermal Units during the demonstration period.
(MMBtu) (heating
plant)
PO4: Comfort conditions in buildings
Deviation from Temperature and Integral average error The comfort conditions in buildings
minimum comfort humidity data from (IAE) from comfort was not adversely affected during
criteria in represen- representative conditions is within 10% optimized operation and the
tative buildings buildings of baseline period IAE. objective was met.
(degrees Farenheit)
PO5: Economic performance
Simple payback or Cost savings, initial Net Present Value of >=0 The main driver for cost savings is
life-cycle cost investment cost, and for a 10 year project the energy savings (PO 3). As
metrics produced by annual maintenance performance period explained above, energy savings
the Building Life- cost of the could not be demonstrated;
Cycle Cost tool technology therefore, the economic performance
criteria were not met during the
demonstration.

11
 Table 1. Performance Objectives (Continued)

Data
Metric Success Criteria Results
Requirements
Quantitative Performance Objectives continued
PO6: Equipment short cycling
Comparison of Equipment ON/OFF ON/OFF frequency Based on the analysis provided in
startup and shut- event data and times under optimized Section 6.6, this performance
down frequency and operation does not objective has been met.
duration between exceed manufacturer or
baseline and operator specifications
optimized operation
for chillers and
boilers
Qualitative Performance Objectives
PO7: Effectiveness of UI
Ability and comfort Feedback and A skilled DPW energy The site resource manager was able
of operators to assess questions from manager can effectively to effectively use the interface and
optimizer outputs for Directorate of Public use the interface and is was quite comfortable with the
operating the plant to Works (DPW) staff comfortable with the software. Some end-users expressed
meet all loads about the logic optimizer outputs concerns that will be considered in
behind optimizer providing a better user experience in
outputs, and actions the future.
taken

12
4.0 SITE DESCRIPTION
Fort Bragg, NC, was selected as the demonstration site. Within this site, the 82nd central cooling
plant and CMA heating plant were used as the demonstration plants.

4.1 FACILITY/SITE LOCATION AND OPERATIONS

Demonstration Site Description: Ft. Bragg, NC, is one of the largest U.S. Army installations,
served by six large central energy plants and a number of smaller plants. At Ft. Bragg, the 82nd
Cooling Plant and the CMA Heating Plant were selected for the demonstration. The 82nd Cooling
Plant consists of four large chillers (1000, 1200, 2000, and 2200 tons), four cooling towers,
associated pumps, and a chilled water storage tank of 2.5 million gallons’ capacity. This plant
provides cooling to approximately 70 major buildings. The location of the plant is shown in Figure
5. The CMA Heating Plant contains three large natural-gas-fired hot water boilers, each having a
heat input rating of 35 MMBH (million British thermal units per hour). Auxiliary equipment
includes primary and secondary hot water pumps and air separation and water treatment
equipment. This plant provides heating to approximately 100 major buildings.

The central chiller plant is monitored and controlled by Honeywell’s Enterprise Building
Integrator (EBI), and the heating plant is monitored by Honeywell EBI, but controlled manually
at the plant using the boilers’ Allen Bradley controls.

nd
82 Cooling Plant

CMA Heating Plant

Figure 5. Location of Cooling and Heating Plants for the Demonstration

13
 Figure 6. Location of Plants and Areas Served

4.2 FACILITY/SITE CONDITIONS

Plant Condition: Both the chiller and heating plants are overseen 24/7 by roving operators who
care for several plants onsite. Figure 6 shows the location of the plants and the areas served.
Honeywell’s automation software EBI monitors both plants and has limited access to control the
chiller plant. In the chiller plant, all control is automatic and has been programmed as different
sequences by a skilled control technician. The operators have been trained to monitor the operator
screens on EBI for this control. The control technician is also intimately involved with monitoring
the plant or taking calls from the operators. The site was able to provide us access to all chiller
plant controls including chiller starts and stops.
In the heating plant, the plant control—boiler start and stop and temperature setpoints—are all
manual. The boilers have Allen Bradley controllers.
The site could not provide automated on/off or temperature control for the boilers because of
warranty issues involving the boiler manufacturer (English Boiler) and the boiler control (Allen
Bradley). This situation meant that the optimizer outputs were provided only as recommendations
to the plant operators, who must then manually start or stop a boiler or change its supply
temperature setpoint. In working with the plant manager, operators, and control technician, a
process was developed so that the operators can follow the optimizer commands at the plant. Since
there is a long start up and shutdown period (the boiler should be well warmed before turning on
the gas to avoid thermal stress problems), the local control starts the primary pumps when
commanded by the optimizer. The operator sees the primary pump operation (from anywhere on
site, not just the specific plant) and is aware that the boiler should be turned on about 30 minutes
after the pumps are on. The supply temperature change is gradual enough for the operator to make
the change periodically at the plant.

14
5.0 TEST DESIGN
This section provides an overview of the system design and testing conducted during the
demonstration.

5.1 CONCEPTUAL TEST DESIGN

Independent At the top level, the presence or absence of the CPOWER optimization
Variable software that operates the central plants
• Total electricity consumed by the selected central plants
• Total gas consumed by the heating plant
Dependent • Total cost of electricity for the selected central plants
Variables • Total cost of gas for the heating plant
• Building temperature and humidity values (for occupant comfort)
• Runtime of the central plant equipment
Controlled • Central plant heating/cooling equipment
Variables • Buildings being served by the central plant
The hypothesis tested that the optimized operation reduces wasted energy
and energy costs by smart allocation of loads, by considering real-time
Hypothesis
price signals, and by operating at the temperatures, flows, and pump/fan
speeds to achieve maximum efficiency of the central plant energy system.
The baseline period ran concurrent with the demonstration period at times
that were convenient for the site personnel to monitor the optimizer
operation and when the plant equipment and control were not down. A
software switch was incorporated in the optimizer software and BAS that
could switch the system between the original automatic controls and
advanced optimization system. This switching could occur manually or at
Test Design set intervals. Because of operator preference and constraints, the interval of
optimized operation was for longer periods closer to a week. The original
control was in control most of the time, interspersed with a few days of
optimized operation. The data from the two operations was compared after
applying weather normalization and day-of-week normalization for the
operation with the existing control system (baseline) to enable fair
comparison for dissimilar weather and occupancy schedules.

15
 Phase I: Control assessment, upgrades and data collection
This phase consisted of surveying the plants to assess the existing control
and automation, upgrading the instrumentation and collecting plant
specifications and data for the modeling task. Based on the assessment, the
list of available points on the automation system is matched with the points
needed for optimization. The instrumentation and communication is then
upgraded to fill any unmet needs.
Phase II: Testing in simulation
The plant and load system are modeled in Simulink® with given plant
layout and specifications. The model is tuned with the data collected in
Test Phases Phase I. The optimizer software was integrated with the model and tested
in the simulation environment.
Phase III: Installation and commissioning
The CPOWER software was installed onsite and connected to the plant
automation system (Honeywell EBI) by mapping point in the appropriate
protocol. Commissioning tests will be performed and system brought on line
to control the plant.
Phase IV: Data collection and analysis
After commissioning, the software switch enabled the plants to run with
optimized control and the existing control. Data was collected during this
phase and analyzed.

5.2 BASELINE CHARACTERIZATION

Because only the central plants’ operation changed and no permanent hardware device was
installed for this demonstration, the project test design enables a baseline characterization period
that is concurrent with the demonstration period. The change between optimized control and
original control is accomplished with a software switch within the optimizer-BAS system. The
data for the baseline characterization was part of the optimizer database system, and was extracted
and transformed for analysis in MATLAB®. Data were used from July 2015, to May 2016, along
with data indicating original control operation.

The individual equipment power consumption data was summed at each time period to arrive at
the total power consumed at the plant. The individual equipment included chillers, primary,
secondary and condenser pumps and cooling tower fans. Other data used included: outdoor air
temperature and humidity, indoor air temperature at representative buildings, type of day (weekday
or weekend), and weather data such as wind speed. After analysis of the data, several anomalous
spikes and constant power values were removed before using the data for modeling the baseline
operation. Data was extracted for the baseline original control days using the
‘EnableClosedLoopControl’ point, which indicates if the plant was in optimized (value of 1) or
original control (value 0). The dataset was divided into optimized and non-optimized periods; these
periods were then sub-divided into 24-hr periods for energy analysis (after discarding any periods
shorter than 24 hrs.). The total energy in KWh, average weather quantities, and indoor air
temperatures for these 24-hr periods were calculated. The 24-hr period energy consumption is
plotted against date in Figure 7.

16
 Figure 7. Energy Consumption (24-hr periods)

To provide a fair comparison, factors that affect the energy consumed need to be normalized. The
approach was to develop a statistical model of the energy consumed during baseline operation,
which can then be used to calculate predictions of energy usage for original operation at the
conditions for optimized operation periods. The main factors affecting energy consumption are
weather, indoor air temperatures, and occupancy. Outdoor air temperature, humidity (and wet bulb
temperature as another measure of humidity), wind speed, heat index, averaged indoor
temperature, and day type of weekend or weekday (in lieu of actual occupancy), were considered
as factors in the regression models. The solar radiation data did not appear reliable in the weather
dataset for the location, and hence it was not used. The energy consumption data has a lot of
variability; to select a statistical model and regression variables that give the least prediction error,
the model based on an evaluation of a combination of regression model algorithm and the
regression variables was chosen. Baseline characterization was performed twice: first with
available data from July to December 2015, and later with all data from July 2015, through May
2016, when all such data became available. Results for the full 2015–2016 dataset are presented.
Table 2 shows the regression variables and regression models that were evaluated with the 2015–
2016 data, using a ‘leave-one-out’ approach (explained below). Each set of regression variables
was evaluated with each model type, for a total 24.

17
 Table 2. Regression Models and Variables for All Data (July 2015 – May 2016)

Model Type
Regression Variables Linear Interactions Pure quadratic Quadratic
Nova OAT X X X X
Nova OAT + wetbulb X X X X
Nova OAT + humidity X X X X
Nova OAT + humidity + windspeed X X X X
Nova OAT + humidity + windspeed + weekday X X X X
Heat Index X X X X

Key
Regression variables: Regression models:
Measured OAT: outdoor temperature measured on site Linear: model contains an intercept and linear terms for each
Novar OAT: outdoor temperature from external weather predictor.
source (Honeywell Novar) Interactions: Model contains an intercept, linear terms, and
Humidity: Relative humidity from external weather source all products of pairs of distinct predictors (no squared
Windspeed: Wind speed from external weather source terms).
HeatIndex: HeatIndex from external weather source Purequadratic: Model contains an intercept, linear terms, and
Indoortemp: Averaged (4 buildings) measured indoor squared terms.
temperature Quadratic: Model contains an intercept, linear terms,
Weekday: Weekday or weekend day type interactions, and squared terms.

Adding a weekday or weekend indicator or creating a separate weekday or weekend model did not
increase the model accuracy in the 2015 data analysis, so this variable was left out of the
evaluation.

Leave-one-out approach:
For each data set and each model type, leave one data row out of the training set and calculate the
prediction error; compute the root mean squared error (RMSE) from each prediction error by
leaving one row out at a time.

The RMSEs computed using the leave-one-out approach for all data using the models in Table 2 are
shown in a color map representation in Figure 8. For this dataset and models, the quadratic model
with outdoor temperature, humidity and wind speed as the regression variables provides the least
RMSE. This model is used as the baseline energy consumption model for the chiller plant. Figure 9
shows the comparison between the actual and expected energy consumption for this model. As
shown, even with the lowest RMSE model, the individual deviations are still significant.

18
Figure 8. Evaluation of Models and Inputs (2015-16 data) – Color Map Representation of
RMSEs

Figure 9. Actual and Model Comparison

19
5.3 DESIGN AND LAYOUT OF TECHNOLOGY COMPONENTS
The schematic in Figure 10 shows a chiller plant that is representative of the arrangement of the
82nd Cooling Plant at Ft. Bragg. A control system is usually installed to facilitate and simplify
automatic control of the plant so that chillers, pumps, and cooling towers can be started or shut
down automatically in a proper order, e.g., cooling water valve, cooling water pump, cooling fan,
chilled water valve, primary pump, and chiller. The optimization solution dynamically generates
optimal schedules and setpoints for plant equipment that will minimize overall operating cost over
a specific time period. Figure 1, above, illustrates the functional components of the optimization
system. The system architecture followed is the one shown in Figure 3 and the specific details are
in Figure 11

Figure 10. Chiller Plant Schematic

Figure 11. System Integration and Controls

20
 5.4 OPERATIONAL TESTING

During commissioning, system communication testing, point-to-point control testing, whole system
commissioning testing, and trial runs for performance were performed. During the demonstration
period, performance testing was executed by running the optimizer for extended periods ranging from
a day to a week. A chronology of all testing is shown in Table 3. The trial runs and performance tests
overlap, since during most performance testing periods configuration or software issues were found
that needed to be corrected. Nevertheless, because the project performance period has ended, results
are being provided based on the analysis of these testing periods.

Table 3. Testing Chronology

Year --> 2015 2016

Month --> 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8
Task 3: Site Implementation
Install with EBI and local controllers H C
Testing
Task 4: Measurement and Verification
Post install monitoring and support
Data aggregation
Analysis and reporting
Periods of optimizer operation - Heating plant (days) 2 5? ?
Periods of optimizer operation - chiller plant (days) 7 6 4 8 7? ? ? ?
Site Issues - heating X X
Site issues - chiller X X X X X X X X X

H Heating Plant
C Chiller Plant
X Issues with site or with software adaptation to site

5.5 SAMPLING PROTOCOL

Once commissioned, the software is set up to collect all control

input and output data points along with additional derived
quantities in its database at 1 minute intervals during optimized
operation as well as during operation with the original control
Data Description
system. The Honeywell BAS also collects data from the plant at
15 min intervals, including data needed for energy savings
calculations: energy consumption at the chiller and boiler plants,
cooling or heating outputs, and weather.
Data Collector(s) Honeywell staff onsite (Bruce Skubon, John Schlesinger)
Data recording was automatic, using the existing BAS
(Honeywell EBI) and Distributed Control System (DCS); the
Data Recording
newly added optimizer workstation connects to the BAS and
records the data in its database.
Data Storage and The existing BAS and DCS have redundancy and data backup
Backup built into the system.

21
 Data Collection A system diagram provided in Figure 11.
Diagram
We obtained electricity price information separately for the
Non-Standard Data demonstration period. This was input into the optimizer
software, and recorded in the database.
Survey Questionnaires No survey questionnaires were prepared or used.

5.6 SAMPLING RESULTS

Figure 12 is a summary plot of the raw heating plant data. It shows the supply, return temperatures,
zone supply flow rate, total heating supply, and the gas used by the boilers. The ‘Optimized’ plot
shows when the plant was under optimizer control and using operational recommendations
provided to the plant. However, it is clear from this plot that the data for the original control (or
non-optimized) period is not recorded, as seen by the constant value lines that correspond to the
value at the end of optimizer controlled operation. This situation may have occurred either because
the workstation was switched off between optimized controlled operation, or a duplicate set of
points was created for the optimizer to read from and write to. The duplicate points were probably
not written to the original local controller, which resulted in the optimizer not getting the correct
I/O data. However, the varying supply temperature (Figure 13) indicates that the optimizer is
working to command the hot water temperature setpoints for the boilers. In the original control,
these temperatures are seldom changed from a fixed setpoint of 220 degrees F.

Figure 12. Heating Plant Operational Data

22
 Figure 13. Supply Temperature Changing

Chiller Plant data:

The plots of optimized controlled periods, plant supply, and return temperatures and the total
instantaneous power consumed during optimized and non-optimized periods are shown in Figure
14 - Figure 16.

Figure 14. Chiller Plant Optimizer Enabled Periods

23
Figure 15. Chiller Plant Supply and Return Temperatures

Figure 16. Total Power Consumed by the Chiller Plant

24
6.0 PERFORMANCE ASSESSMENT
PO1: SIMULATED OPTIMIZER SOFTWARE PERFORMANCE
The purpose of the objective was to show that the optimizer software can control the chiller and
heating plants safely and within normal operating limits. The central plants and building loads
were modeled in the Mathworks® Simulink environment. The optimizer software was interfaced
with this model for testing. Several simulations of plant operation were performed and data was
gathered to check safe and correct optimizer performance in simulation. Several combinations of
activities were simulated covering the range of loads, weather and electricity prices, and their
transitions in the simulation framework. The data collected (optimizer outputs) was compared
graphically against known normal operating ranges for the equipment. Equipment run-times were
compared with minimum prescribed in the optimizer UI. Performance objective PO1 was
successful based on the analyses and metrics.

PO2: OPTIMIZER SOFTWARE INTERCONNECTION WITH CONTROL SYSTEM

The purpose of this performance objective was to test that the optimizer interface to the automation
and control system works correctly, the inputs and outputs have been mapped correctly, and an
overlooked local control doesn’t override the optimizer. Several tests were performed during
commissioning that included visual, quantitative and graphical analysis. With the successful
commissioning of the optimizer, it is concluded that this performance objective has been met.

PO3 ENERGY SAVINGS AND PO5 ECONOMIC PERFORMANCE

The purpose of this performance objective was to measure energy savings in cooling and heating
plants by operating them real time with CPOWER’s optimization solution. The data recorded in
the CPOWER database was used for analysis.

After commissioning, several issues were addressed to ensure that the optimizer ran as intended,
the plant was operated safely and reliably, and the plant personnel were comfortable with the
operation. The initial plan to switch the plant operation between the original control and optimizer
control on alternate days or weeks was modified to operating with the optimizer for several days
at a time, when site staff would be available to monitor, and no maintenance work was ongoing at
the plant. All data for points that were mapped for CPOWER operation and other calculated data
from the software were recorded in the CPOWER software database. MATLAB scripts read
Microsoft Excel files exported from the database and organized them into user-friendly structures.
Data extracts from different periods were merged to create .mat files with structures spanning the
period from July 2015, through May 2016.

Most of the data analysis described below uses this data, except when other data sources were
needed to corroborate or fill in gaps for periods of corrupted or unavailable data. Two other data
sources were used for this: Weather data from an outside source (Honeywell Novar weather data)
and the BAS EBI’s data. Honeywell’s Novar weather data is currently accessible from a
Hortonworks cluster; we query this dataset to obtain csv files for the periods and place of interest.
The plant EBI data is shipped as Excel file reports for each week, for several equipment points. A
separate set of MATLAB scripts was created to read these data into a streamlined usable format.

25
Data preprocessing involved removing anomalous spikes, periods of constant measurement
(indicating lost communication), and creating the scripts to demarcate optimized and non-
optimized periods and create analysis windows of 24-hr periods within those periods to calculate
total and average quantities. The total energy consumed in a 24-hr period during non-optimized
periods were used to characterize baseline operation by building a regression model using several
weather and occupancy factors. The baseline model was used to calculate predictions of total
energy usage for original operation at the conditions for optimized operation periods. We
compared the actual energy use during optimized periods with the expected energy use with
original control. The results show that in most cases, the optimized actual consumption is within
one standard deviation of the expected usage with baseline control. The unqualified overall usage
however, does indicate that optimized operation did not improve the energy consumption and
energy consumption increased by 5.84%.

Since the above results were completely unexpected, the data was analyzed to find out if the
optimizer had been functioning correctly, if other factors were affecting optimized operation, and
if input data into the optimizer during operation was correct.

1. Baseline model fit: Although a rigorous method was applied to model the baseline data, using
several factors and model types, the best baseline model has significant deviations from the
actual energy usage. It appears that several factors affect the total energy consumption of the
chiller plant and additional data and additional factors (e.g., solar insolation) may be needed to
obtain a better model.
2. Inputs to optimizer: During several periods of optimized operation the correct data was not
being transmitted to the optimizer. Without a continuous presence onsite or a remote
connection, it is impossible to know if the user provided parameters and real time inputs are
correct while the optimizer is in operation. The optimizer software is complex and does not yet
include standardized communication interfaces for controller or BAS integration, hence the
application engineering skills to transfer the technology to the field have not been fully
developed. The site staff includes mostly operations personnel. Software and communications
must be monitored when in operation to ensure not just that the plant operates correctly (the
site staff was qualified to do this), but the software is getting all its inputs and operating ideally
(needed Honeywell Laboratories personnel or optimizer software experts for this). The indoor
and outdoor temperature impacts how the optimizer forecasts load for starting and stopping
chillers and calculated corrections to the supplied energy in the short term. The outdoor
temperature had been wrong for certain periods. One of the first bits of anomalous data that
was noticed with the new set of data in 2016 was the big spike in Total Power calculated from
a summation of all equipment power data, which was traced to Condenser Pump #4. On being
apprised of this power spike, the site staff lead immediately said that was probably why Chiller
#4 was never being switched on, and would be switched off as soon as possible when the
optimizer was in control: ‘the optimizer hated Chiller #4.’
3. Learning plant equipment models: The sequence of issues faced during the demonstration
period meant that the optimizer software did not have long enough periods of stable operation
for learning equipment models, and sometimes was not recording the correct inputs for the
models.

26
 The main driver for the cost savings comes from energy savings in this project. From the
analysis provided for PO3 Energy Savings, it may be concluded that cost savings arising from
energy savings could not be achieved during the demonstration.

PO4: COMFORT CONDITIONS IN BUILDINGS

The purpose of the objective was to ensure cooling and heating comfort is not adversely affected
in the buildings during optimized operation. Space temperature data from representative buildings
was collected during the baseline and demonstration periods. The analytical approach (1)
compared the indoor temperature measurements with respect to the temperature limits specified in
the optimizer, and (2) used the data collected in 2014 as baseline for comparison with the indoor
temperatures during the demonstration period. Visual comparison of plots of baseline and
optimized operation showed no significant adverse difference. Average value of indoor
temperatures for cooling periods in 2014 (baseline control) was about 2 degrees higher than the
average value during optimized operation. The results show the performance objective was
achieved.

PO6: EQUIPMENT SHORT CYCLING

The purpose of this objective was to quantify short cycling of chillers and boilers. The plant
equipment ON/OFF data from the database was used for analysis and computed ON and OFF time
intervals for each chiller and boiler. The time intervals were compared with minimum ON and
OFF times for such equipment as gathered from manufacturer specifications and operator
interviews. The optimizer software allows the user to set up minimum and maximum run and rest
times for chillers and other equipment. These parameters are soft constraints, since they may be
overridden by other concerns such as safety or comfort. For example, the optimizer will respect a
minimum run time setting of 2 hours, unless a safety concern such as exceeding maximum
compressor current occurs, and then the chiller would be commanded OFF.

Quantitative information about the ON and OFF times for the four chillers are presented in Table
4 and Table 5. Columns 2 and 3 present the median duration of ON or OFF periods for optimizer
and original control periods. The last two columns present the number of periods when the
durations were shorter than the benchmark 2 hours (for ON), or 30 minutes (for OFF), versus the
total number of periods in the demonstration period.

Table 4. ON Duration Statistics

Median ON Median ON # Shorter than 2 # Shorter than 2

duration – duration - hours /total # hrs/total #
Chiller
optimization original durations- durations -
(minutes) (minutes) optimization original
#1 169.5 217.5 22/108 11/62
# 2* 50 773.5 2/5 0/4
#3 164 484 6/136 15/81
#4 428.5 536 3/52 12/57
* Chiller # 2 had problems and was not run much during the demonstration period.

27
 Table 5. OFF Duration Statistics

Median OFF Median OFF # Shorter than 30 # Shorter than

duration – duration - min/total # 30 min/total #
Chiller
optimization original durations- durations -
(minutes) (minutes) optimization original
#1 73 565 0/107 1/63
# 2* 228.5 29245 1/4 0/5
#3 45 376.5 14/135 5/82
#4 141 401 1/52 1/57
* Chiller # 2 had problems and was not run much during the demonstration period.

On average, the chiller ON/OFF durations are shorter for the optimized than for original operation.
However, that condition was expected, given the optimizer’s objectives. During the demonstration
period, the on and off times for both optimized and original control was analyzed. Apart from the
larger number of shorter cycles, it is not clear that the optimizer is exceeding a threshold very
frequently, even compared with the original control. The last two columns in Table 4 show that
the original control also had several instances of cycle durations shorter than the benchmark above.
Therefore, given that the optimizer software provides the flexibility to adjust the cycle times, we
consider this performance objective has been met.

PO7: EFFECTIVENESS OF UI (QUALITATIVE)

The purpose of this objective was to evaluate the need to improve the operator UI for future
widespread adoption. Feedback was obtained from interaction with the site staff during the
commissioning process and demonstration. With frequent use, the site lead became familiar with
the optimizer software and functionality compared to initial impressions. He was very comfortable
putting the optimizer in control and letting it operate without supervision overnight and
continuously over several days. The site lead liked that the optimizer changing the chilled water
and hot water supply setpoints continuously, within specified limits, because the current control
system is set up to operate at fixed setpoints.

The site personnel did not like the cycling of the equipment. The optimizer software was set up so
that chillers, which are large equipment, did not switch frequently; however the pumps and fans
were set up to give them flexibility in switching, within limits. The complexity and multitude of
parameters to be set on the software can be overwhelming to plant managers and operators.

The plant personnel also did not know why the optimizer would make a particular choice when
they would have intuitively made a different choice. The recommendation is to improve the
software by providing a concise quantitative reason that shows the comparison of energy cost
between a previous setting and current setting.

28
7.0 COST ASSESSMENT
7.1 COST MODEL

The costs given in Table 6 reflect an estimate based on experiences onsite and the vision for scaling
the demonstration for commercial use. The estimate reflects considerations of software
improvements to reduce site troubleshooting, changes in the software architecture, streamlined
interface for optimizer with local controller or automation system, training of application engineers
for installation.

Table 6. Cost Model

Cost Element Data Tracked During the Demonstration Estimated costs

Software license cost Software license $60,000 - $150,000

Software installation costs Estimate of labor required to install and configure $11,000
software
Training Software Training to operators and technicians $6,800

Hardware and installation costs Extra instrumentation on site – cost of hardware $10,000
and installation labor
Cost of PC workstation Cost of PC to host software $2,500
Maintenance Software maintenance updates and customizations $15,200 (recurring)

Software license fees: This is the estimated cost of the software license for small- to large-sized
complex chiller plants, ranging from 2 chillers and 1200 tons, to 5 chillers and 6500 tons.

Software installation cost: This cost includes labor to install and configure the software for a
specific site by connecting to the input and output points. It includes the labor for installing
appropriate compliant software on the workstation such as Army Gold Master OS and connecting
to the automation system.

Operator training: This cost includes the labor cost for an application engineer to train the
operators and facility manager.

Hardware and installation costs: It was assumed that a well-instrumented central plant would
have automation, but that not all required measurements and actuation for optimizer software
would be available. Typically, flow or BTU meters and power meters for pumps and cooling
towers may not be available. In addition, it is possible that an existing sensor, actuator, or controller
may have the requisite measurement but is not connected to the automation or control system.
Communication cards may be needed to bring in all the points the optimizer requires. The
installation costs include labor for installation of additional sensors, meter, communication cards,
and the labor to map these measurement points to the automation system.

29
Cost of PC workstation: Cost of the computer to host the software on site. This estimate may
change in the future as enhancements are addressed in the software architecture and automation
system architecture, such as Cloud hosted services.

Maintenance: This estimate provides the labor cost of software upgrades and customizations for
the site (after commissioning).

7.2 COST DRIVERS

Cost drivers that can affect the cost of implementing the technology include:

• Status of instrumentation and automation at the site: Several sensors and meters are needed
to gather all data inputs for the optimizer. If a site is already well-instrumented and
automated, the cost of upgrading to a supervisory level optimizer will be lower.

• Availability of skilled control technicians on site: The cost of implementation will decrease
as more support and knowledge from the site becomes available on mapping and
contextualizing control points.

7.3 COST ANALYSIS AND COMPARISON

The realistic cost estimates for the technology when implemented operationally are provided in
the previous section (Table 6) and described further in the same section. Table 7 illustrates a cost
analysis for a central chiller plant. The full comparative life cycle analysis and inputs are in
Appendix of the full Final Report.

Assumptions:

1. For the cost analysis, it was assumed that a site with a large plant, but without the complexity
of storage tank or free cooling that was encountered at the Ft. Bragg, NC site.
2. The site is well instrumented and the site has control technicians able to provide support for
integrating the software at the plant.
3. The plant is well maintained, with minimum downtime of plant equipment.
4. The site has modern communication and automation infrastructure that is well maintained.
5. The optimization software has been productized with a robust architecture and other
improvements, and application engineers and technicians trained in installation and
commissioning provide standardized support.

30
 Table 7. Summary Cost Analysis for a Chiller Plant

Inputs Outputs
Project Name: CPOWER Results 15-yr
Project Location: North Carolina  Energy Consumption Cost Savings $ 443,698.00
Analysis Type: FEMP PV of total savings $ 215,698.00
Base Date: April 1 2015 Net savings $ 85,398.00
Beneficial Occupancy Date: April 1 2015 Savings-to-investment ratio 1.66
Study Period (years): 15 Adjusted Internal Rate of Return 6.52%
Discount Rate: 3% (default Payback period (simple and discoun 7 years
Discounting Convention: End-of-year Electricity savings (kWh) 8,245,290.00
Electricity Savings Per Year (kWh) 549,761.29
Emissions reduction
CO2 reduction (kg) 9,761,923.21
Optimization Package Capital $130,300 SO2 reduction (kg) 32,358.95
Annual Maintenance, Updates $15,200 Nox reduction (kg) 14,606.06

31
Page Intentionally Left Blank

32
8.0 IMPLEMENTATION ISSUES
Three types of issues during the demonstration period were encountered:
1. Technical and personnel resource issues: The optimizer is complex software with advanced
algorithms. In addition, it performs the actions of a simple controller, commanding equipment
in real time. The optimizer needs to be integrated well with the existing automation, which
requires experience and skill in a succession of staff in the project sequence—algorithm
developer, software architect and developer, application engineer, control engineer and
technician, BAS programmer, plant supervisor, plant operator, and site technical manager. A
number of the issues occurred because the prototype software hadn’t yet been architected for
easy deployment, with appropriate tools, and this succession of staff weren’t always available.
A productized version of the software will not face the same issues and the mobilization of staff
would be automatic: software that is a current business offering has the backing of trained staff
to support the releases which is their job priority; a prototype version is still in the proof of
concept phase and staff has to be mobilized on a case-by-case basis.
2. End User concerns: The end users were not always comfortable with the software. Some of
the concerns have been documented in the performance objectives section. In summary, the
main points of user concern are:
a. Operating the plant with the optimizer is a very different from current practice. In current
practice, the controller operates the chillers in different fixed modes; e.g., fixed chiller
supply and condenser return temperatures. Under the optimizer operation, when the site
staff see changing supply temperatures, flows, and switch on/off of equipment, they cannot
understand the operation and motivation until they become more familiar with the software.
To improve and speed up site staff familiarity with the software, one recommendation is to
develop an improved UI that can explain automated system changes and the benefits to the
user, real-time.
b. The users felt that the optimizer cycled the equipment too much compared to the current
practice. This concern was handled to some extent by configuring user parameters in the
optimizer software as well as making changes in the software. This concern will have to
be addressed through software improvements that can assign a cost to cycling, training of
personnel, and data-driven explanations on the software front end to the user.
3. Site issues:
a. Data quality: A lesson learned during this demonstration is that the data quality needs
continuous monitoring. Although rigorous testing took place during commissioning and at
other visits, the following two assumptions were incorrect because the focus was on
correctly operating the optimizer: (1) that the data continues to be good if the optimizer can
operate reasonably within limits, and (2) the data recorded by the optimizer is the same as
that used by the local original control. From an operational perspective, it was found that
despite bad data, the optimizer continued to function reasonably smoothly, however, it did
not control optimally. It was discovered that a duplicate set of points were created for the
interface to the optimizer, which meant that the optimizer did not see all the same states
and commanded points that the original control used unless they were written to the
duplicate points by the original control.

33
 b. Remote monitoring and troubleshooting: Because of DoD site restrictions, no remote
access to the optimizer workstation was permissible. This severely restricted the speed and
quality of troubleshooting that could be provided without being on site. As stated
previously, the software is complex and in a prototype state; therefore, it is difficult to
manage and monitor continuously without the experts, since it works in real-time. The
software should ideally be provided as a cloud service and, at a minimum, with expert
remote support. Providing a process for secure remote access would have greatly increased
the effectiveness and the value of the project.
c. Information assurance: A DoD-wide smooth information assurance process would have
saved time and effort in this project. The information assurance pre-work was started in
early 2014. It was understood from the DPW Energy Manager that the CoN (Certificate of
Networthiness), and later the Interim Authority to Test (IATT), were the approval process
for implementing a software onsite. The network architecture was created and gathering
information on the process and information to be provided from the NEC as well as
NETCOM through the DPW Energy Manager was attempted. CERL colleagues assisted
in accessing the sites, as a Common Access Card was needed. This formal process was
finally not required, since the software was implemented on a test basis, on a VLAN that
is isolated from other site networks.

Procurement issues: All hardware required for implementation is standard commercial off-the-
shelf [COTS] and not expected to be a concern in the future.

The program resulted in the successful commissioning of a very complex supervisory level
optimization software that continuously receives real-time sensor data, computes optimal
operating points, and commands plant equipment in real time. The testing provided valuable
lessons for improvement of the software, user experience, and transitioning to DoD sites. Below
are some recommendations for improvement of the specific technology process, as well as the
project process.
(1) Re-architect the software to separate the supervisory and local control layers; the
supervisory layer providing high-level operating schedules and setpoints which are then
managed and controlled by the local control layer. This will not only improve the software
ease of implementation and performance, but eliminate safety concerns due to network
communication issues. It will also vastly improve the operational staff’s comfort with the
software.
(2) Phase in the commercial transition with less complex plants, e.g., chillers only without
additional energy sources.
(3) Develop standard implementation tools to quickly and reliably configure the software and
connect it to the local control on site.
(4) Improve user experience by providing explanations for the optimizer’s major actions.
(5) Improve cycling frequency by considering equipment cycling as a cost in the optimization
objective function.

34
(6) Data quality check process: Data quality checks were done at several points in the project,
which led to successful commissioning. However, for any control, software or data-
intensive applications that require continuous data streams, the data quality check and
cleaning should be inserted as an automated data anomaly detection software. This would
alert the field engineers if the data coming into the application is correct.
(7) For complex software that needs advanced development skills, it is difficult to develop
software that is simple for field engineers to understand or that has no field engineer
concerns. Securing remote access to the system would have made it possible for offsite
expert engineers to monitor the in-operation performance and would have flagged issues
early. Another approach may be to partner with advanced solution providers near the DoD
site (e.g., universities, national labs or industry partners) who could be embedded onsite
for closer monitoring of the system operation.

35
Page Intentionally Left Blank

36
9.0 REFERENCES
1. Department of Defense Strategic Sustainability Performance Plan, FY 2010, Public Version,
26 August 2010.

2. Eto, J, H., ‘On Using Degree-days to Account for the Effects of Weather on Annual Energy
Use in Office Buildings’, Energy and Buildings, 12/1988, 113-127.

3. Executive Order 13327, http://www.gpo.gov/fdsys/pkg/FR-2004-02-06/pdf/04-2773.pdf

4. Executive Order 13423, http://www.gpo.gov/fdsys/pkg/FR-2007-01-26/pdf/07-374.pdf

5. Executive Order (EO) 13514, https://www.fedcenter.gov/programs/eo13514/

6. Executive Order 13693, https://www.fedcenter.gov/programs/eo13693/

7. Hutchinson, M.W. Air Force Energy Implementation, SAME Texoma Regional Conference,
24 August 2011.

8. Makhmalbaf, A., Srivastava, V. and Wang, N., ‘Simulation Based Weather Normalization
Approach to Study the Impact of Weather on Energy Use of Buildings in the U.S.’, Proceedings
of BS2013, 13th Conference of International Building Performance Simulation Association,
Chambery, France, August 2013.

9. Nexant, Inc., ‘M&V Guidelines: Measurement and Verification for Federal Energy Projects’,
https://www1.eere.energy.gov/femp/pdfs/mv_guidelines.pdf, Version 3.0, April 2008.

10. S.T. Taylor. 2012. Optimizing Design and Control of Chilled Water Plants, ASHRAE
Journal, June, pp. 56-74.

11. www.ipmvp.org, ‘Concepts and Options for Determining Energy and Water Savings’, Volume
1, http://www.nrel.gov/docs/fy02osti/31505.pdf.

12. Ye, C.Z., X. Xiang, D.Q. Zhang, and J.W. He (Inventors). 2010. Dynamic Economic Load
Dispatch by Applying Dynamic Programming to a Genetic Algorithm, U.S. Patent No.
7,752,150 B2, July 6. Assignee: Honeywell International.

37
Page Intentionally Left Blank

38
APPENDIX A POINTS OF CONTACT

Point of
Organization Phone
Contact Role in Project
Email
Name
Girija Honeywell ACS (763) 954-6554 Principal Investigator,
Parthasarathy Labs girija.parthasarathy@honeywell.com Program Manager
Keith Johnson Honeywell ACS 763-954-4426 Data and specifications,
Labs Keith.Johnson4@Honeywell.com Software configuration,
implementation
Rebecca Kemp Honeywell Labs 763-954-2712 Contract Management
Rebecca.Kemp@Honeywell.com
Richard Honeywell Building PM for site support and
Arizmendi Solutions implementation
John Honeywell Building 910-391-8040 Ft. Bragg Energy team
Schlesinger Solutions John.schlesinger@honeywell.com member, site plant
technical advisor
Bruce Skubon Honeywell Building 910/436-5144 BAS expert, programming,
Solutions Bruce.skubon@honeywell.com data collection
Bill Honeywell Building (910) 436-0440 Control system technician
Klingenschmidt Solutions William.klingenschmidt@honeywell.com
Francesco University of 510-517-9203 Technical lead for
Borelli California, Berkeley fborrelli@berkeley.edu modeling and simulation
tool
Sergey Vichik University of 510-666-7162 Modeling and simulation
California, Berkeley sergv@berkeley.edu development
Jason Kong University of 650-898-7551 Modeling and simulation
California, Berkeley jasonjkong@berkeley.edu development
Matt Swanson U.S. Army ERDC- (217) 373-6788 CERL lead, Technical
CERL Matthew.M.Swanson@erdc.dren.mil advisor, and dissemination
of results
Noah Garfinkle U.S. Army ERDC- (217)373-4576 CERL Technical advisor
CERL noah.w.garfinkle@erdc.dren.mil
Laura Curvey U.S. Army ERDC- 217-352-6511 ext. 7338 Technical advisor,
CERL laura.curvey@usace.army.mil coordination, and
dissemination of results
Runqing Zhang Honeywell +86 021-2894 4100 Optimization solution
Technology Runqing.zhang@honeywell.com developer
Solutions, China
Qing Li Honeywell (21)2894-2557 Optimization solution
Technology Qing.li@honeywell.com developer
Solutions, China
Benny Dong Honeywell Benny.dong@honeywell.com Optimization solution
Technology application engineer
Solutions, China

A-1
 Point of
Organization Phone
Contact Role in Project
Email
Name
Benson Wei Honeywell Optimization solution
Technology developer
Solutions, China
Nick tong Honeywell Optimization solution
Technology developer
Solutions, China
Ft. Bragg DPW personnel (not formally performing the project)
Coby Jones Formerly Ft. Bragg 704-502-7575 DPW Energy Manager
DPW joseph.c.jones4.ctr@mail.mil
Jim Peedin Ft. Bragg DPW james.f.peedin.ctr@mail.mil Ft. Bragg Energy Team
consultant

A-2