ICPTT 2009 © 2009 ASCE 521

A Pipe Network Skeleton Method based on GIS Network Analysis Technologies 

Tao TAO1, Kunlun XIN 2, Suiqing LIU 3 and Xinyu ZHANG 4

1
TAO Tao, Tongji University, 1239 Siping Road, Shanghai, China; email:
taotao@tongji.edu.cn
2
XIN Kun-lun, Tongji University, 1239 Siping Road, Shanghai, China;
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

email:xkl@tongji.edu.cn
3
LIU Sui-qing, Tongji University, 1239 Siping Road, Shanghai, China;
email: liusuiqing@tongji.edu.cn
4
ZHANG Xin-yu Tongji University, 1239 Siping Road, Shanghai, China;
email: zxy@tongji.edu.cn

ABSTRACTS

During the process of urbanization, many of the existing urban water-supply systems
need to be expanded to meet the increasing urban water consumption. A precise
water network hydraulic model is necessary to perform water network expansion
plan. Building a hydraulic model is normally based on the established GIS of current
water network, which contains quite huge quantity of information about pipes with
various diameters and water facilities including pumps, valves and hydrants.
However, much of those data in water network GIS is not necessary for hydraulic
modeling and simulation. This paper proposed an auto skeleton method which based
on the ArcGIS utilities network analysis toolbox. Firstly the geometric network was
established in which the flow directions and the water sources are determined as well
as the topology of the water network. Secondly, the branches of network were
identified step by step and trimmed provided that the water demands in the branches
were accumulated to the root junction of the branches. Thirdly, the pipe loops which
can be deleted in the network were also indentified and deleted with demands
preserved to the corresponding junctions. Finally, the pipes with same diameters
which located at a non-branch route were merged providing the lengths of the
merged pipe were not beyond the present value. The proposed method can simplify a
complicated network to a hydraulic-equivalent network model in which about
50-60% of pipes were removed, and huge plenty of tedious manual work were thus
avoided.

KEYWORDS

Water distribution system; GIS; Skeleton; Modeling

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 522

INTRODUCTION
Job of pipe networks planning and designing are usually divided into two categories.
The first is for newly-built pipe networks. Designers first determine the layout of
main pipes, main loops and the connection status, which compose a whole network
model that can be directly used to carry out hydraulic simulation. The second
category is for pipe network rehabilitation and expansion. In this case, the city
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

usually has a well-established pipe networks, moreover, a GIS for pipe network
hydraulic model maybe already existed. Due to the requirement for urban
development, new pipes are needed to be laid to expand the existing pipeline
network. The established pipeline GIS contains the full details of the pipe network
like household pipes, various types of valves, water meters, as well as fire hydrants
and other ancillary facilities(Garaci, et al., 2002), (Umble, et al., 2003)( Figure 1). It
should not be directly used as a pipe network model to carry out hydraulic
calculation, otherwise too huge but unnecessary calculation loads will be brought. As
a result, the existing pipe network geographic information system or network model
should be simplified in advance before it is used for pipeline network simulation. So
in the case of the rehabilitation or expansion of the existing pipeline network, the
GIS-based network simplification strategy discussed in this paper is a more
reasonable solution to hydraulic modeling of pipe network.

Figure1. GIS OF WATER NETWORK

Figure 2. HYDRAULIC MODEL OF WATER NETWORK

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 523

FUNDAMENTAL TECHNIQUES OF GIS-BASED NETWORK ANALYSIS

Geodatabase is introduced as a new object-oriented spatial data model into ArcGIS 8

by ESRI. Geodatabase using an open-space data structure to manage spatial data
(include: vector, raster, image, three-dimensional terrain, etc.). Both the spatial data
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

and the associated attribute data are stored in a unified industry-standard database
management system (DBMS)(Jason, 2004). Apart from the storage of geometric
characteristics of space data, it also has the description of the data objects,
relationships between the objects and the operation rules. Therefore, the data for
users is no longer the abstract point, line, surface, but the familiar subjects in
practical application such as cables, pipes, valves, inspection wells. These elements
compose the corresponding tables in DBMS, while a specific element (Feature) is
related to a record in the table. Because Geodatabase is an inherent Data Model of
ArcGIS, the planning database in Geodatabase format can facilitate a variety of
spatial analysis work while performing the GIS-based network planning.

Geometric Network of Pipelines. In GeoDatabase, connecting table is used to

maintain the network topology among the elements (Figure 2). In the process of pipe
network simplification, the dimension of the node connection is a measurement to
distinguish the isolated node, end node or source node, shown as following:
1) 0-dimensional: that is, the non-pipeline connected nodes, or isolated nodes.
2) One-dimensional: only one pipeline is connected to the node, which means the
end node or source node.
3) More than 2-dimensional (including 2-D), non-isolated node;

Basic Network Analysis Techniques .

(1) Source, sink and flow direction
The source nodes in the geometric network those from which the fluid always flows
out to the other nodes in the network. On the contrary, the sink nodes are the nodes
that the fluid always flows into from the others. With the help of the sources and
sinks in the network, we can judge the flow direction of fluids according to the
relationships between network nodes.
(2) Weight of edges in the Geometric Network
Weights are generally used for searching the shortest path in the network. Weights of
edges in the geometric network are usually related to one or more values in data
table, and it is one of the important elements of network analysis especially in job
such as shortest path searching. For example, for water supply network, pipe
diameters are often treated as weigh for geometric network analysis. During the
process of pipe network simplification, through judging the weight corresponded to
the pipeline, we can decide the diameter boundary for pipe trimming, which means

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 524

pipes with diameter larger than the boundary shall be maintained, while those with
smaller diameters shall be trimmed.
SKELETON STRATEGY OF WATER SUPPLY NETWORK MODEL

Step 1- Preparation.
(1) Performing the flow direction analysis of network. Because in the process of
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

network skeleton, we need use some methods, such as “Trace Upstream” or “Trace
Downstream” and so on. Therefore, we should determine the flow direction of pipes
in the network in advance. To do this, for source nodes in the network, we set the
“Accilarryrole” property to be “Source”, while for others we set to be “Non-Source”.
(2) Performing the “SetFlow” operation on the geometric network.
(3) Setting the weights (this step should be done while setting up the geometric
network), including:
z Determining the weights of nodes according to the nodal flows in network
analysis in order to accumulate the nodal flows of the branched pipe network
to the root node.
z Determining the weights of pipes according to the diameters in network
analysis in order to determine the required diameter limits for pipe trimming.

Step 2- Trimming of branched pipe network.

(1) Method 1:
1) Traversing all the nodes that connected with three or more pipes (named as J0),
and begin to trace the end nodes and all pipes at the downstream of the J0.
2) Assessing the tracing results:
If the end node is connected with more than one pipe, which indicates that the node
is located on the looped network. The node in the looped network can not be
determined the upstream and downstream relationship, so the node and the upstream
pipes and nodes have to be retained temporarily.
If all end nodes are only connected with one pipe (recorded as J1), which means the
traced section of the network is a branched network. Then traversing the pipes of
branched network, if there is no pipe with diameter greater than the diameter limits
accumulate all the nodal demands to the root node of branch network, and trim the
whole branch network except for the root node.
However, it is concluded from the test results that the above method can not remove
all branched pipes at a time.
(2) Method 2:
1) Traversing all end nodes, if the diameter of pipes is not greater than the diameter
limits, delete all the pipes connected with it, and accumulate the nodal flows to the
root nodes.

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 525

2) Perform step (1) repeatedly until there is no node in the network connected with
only one pipe, which is in accordance with the standard of deleted end nodes.
With the second method we can cut away all the branched pipes perfectly. But
because it is a recursive procedure starting from the end of the network to trim the
pipes and nodes step by step, the second method is proved to be time-costing and low
efficient, especially for the complex network having a large number of branches.
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

With the first method we can cut off the vast majority of branched networks only in
one traversal, which means it is more efficient than the second method. However, it
will miss some branched pipelines out. The most efficient method is to trim the
majority of branched pipes firstly using the first method, and then trim the left
branches with the second method. The network after branches-trimming is shown in
Figure 4.

Figure 3. WATER NETWORK Figure 4. NETWORK AFTER

BEFORE SKELETON BRANCHES-TRIMMING

Step 3- Identifying and deleting of the isolated loops. Under normal

circumstances, in order to avoid the hydraulically deviating from the real network,
loops in the network are not permitted to be trimmed off. However, the isolated loop
in the branched network may be considered to be cut off, if the diameter of the pipes
in the loop meets the simplification requirements (loop L1).
To identify the loops qualified for being trimmed off, carry out the operation of
finding loops firstly according to the algebraic relations between pipes and nodes in
loop: Np = Nj + NL-1. In addition, it is needed to know whether there are common
pipes existing among the loops based on the algebraic relations of the number of
nodes, pipes and loops.
(1) First of all, traverse all the nodes that are connected with three or more pipes
as well as those have upstream nodes, then operate the recursive steps as follows:
Void FindAndCutLoops(Junc0) {

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 526

P=Junc0.UpstreamPipe // (upstream pipe of Junc0)

Net.SetFlag (Junc0) // (tag Junc0)
Net.SetBarrier (p) // (set pipe P as a trace barrier)
bHasLoops=Net.FindLoops (JuncIDs, PipeIDs) // (find a loop)
if( bHasLoops=False) return;
If(PipesIDs.Count-JuncIDs.Count = 0){/(Contain at least one of the
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

isolated loop)
For i=1 to juncIDs.count{
Junc1=JuncIDs(i) //(traverse all nodes in the searching
results)
If ( Junc1.ConnectedPipeCount >2 ){/(it means in the result
there are more than one isolated loops)
FindAndCutLoops(Junc1)//(Recursively find and cut off the
loop)
}
}
//(traverse and open all loop having Junc0)
While(Net.FindLoops(JuncIDs,PipeIDs){
Junc0.NextPipe.Break //(Cut off a non- upstream pipeline
connected to Junc0)
}}}
(2) After the isolated loops in the branched network has been identified and
broken, the original loop L1 changed to a branched network as T1. By using the
operation of trimming branched network again shown in step 2, the network was
further simplified (Figure 5).

Figure 5 . NETWORK AFTER Figure 6 . NETWORK AFTER

LOOP-TRIMMING PIPE-MERGING

Step 4 - Merging pipes with same diameters.

(1) In order to avoid the merged pipe is too long, set the upper limit of the
length of the merged pipeline as Restricted Length.

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 527

(2) Traverse all the node J0 connected two pipes (P1, P2). If the pipes have the
same diameter, divide the flow of J0 into two parts according to the pipe length and
assign separately to the adjacent nodes such as J1 and J2, then merge pipe P1and
pipe P2 into a new single pipe.
The procedure of merging pipes is shown as following:
for i=1 to Net.JuncCount{
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

J0=Net.Junction(i)
If( J0.ConnectedPipeCount=2) {
P1＝J0.Pipe(1)
P2 = J0.Pipe(2)
If ( P1.diameter＝P2.diameter and
P1.Length+P2.Length<RestrictedLength ) {
J1=P1.NextJunction
J2=P2.NextJunction
J1.Flow=J1.Flow+J0.Flow×(P1.Length/ (P1.Length + P2.Length)
J2.Flow=J2.Flow+J0.Flow×(P2.Length/ (P1.Length + P2.Length)
P1=Merge(P1,P2)
}}}
The network after the pipe merging is shown in Figure 6.

CASE STUDIES

The proposed method was then applied to a case studied network as shown in Figure
7. The network composed of 7378 pipes and 7292 junctions is the part of a real urban
water distribution system. As shown in Figure 8, after a series of operation, the pipe
network was finally simplified into a skeleton network which only contains 1247
pipes and 1235 junctions. The total CPU time for performing the skeleton operation
is less than 2 minutes on a computer with AMD Athlon64 X2 and 1GB RAM.

Figure. 7 NETWORK BEFORE Figure. 8 NETWORK AFTER

SKELETON SKELETON

ICPTT 2009
 ICPTT 2009 © 2009 ASCE 528

CONCLUSIONS

Through the simplification of the original complex GIS network, only the larger
diameter pipes and the main loop are retained, which greatly reducing the number of
nodes and pipes in the network, and increasing the efficiency of hydraulic
calculation. The network simplification strategy proposed in this paper is using
Downloaded from ascelibrary.org by East Carolina University on 09/16/15. Copyright ASCE. For personal use only; all rights reserved.

GIS-based network analysis technologies. Through the topology analysis of the

corresponding geometric network, the small pipes and loops can be quickly and
accurately trimmed and a simplify network model can be established for hydraulic
simulation, which provided a powerful tool for hydraulic modeling in rehabilitation
and expansion of water distribution network.

ACKNOWLEDGMENTS
This project was funded by National Key Technology R&D Program during the 11th
Five-Year Plan Period ,China (Project No. 2006BAJ08B03).

REFERENCES

Art K. Umble Melissa Moran, Tarlochan Bhullar and Michael C. Machlan.

(2003)“Utilizing GIS in Developing Realistic Demand Distributions to
Support Modeling in Water Supply Master Planning”. Pipelines.
Daniel F. Moore Curtis Thompson and R. Joseph Bergquist. (2002) “Geographic
Information System (GIS) Decision Modeling for Water Pipeline
Planning. ” . Pipelines.
ESRI. (2001). Exploring ArcObjects, ed. Z. M. , ESRI Press, California.
Jason Luettinger P.E. and Thayne Clark, P.E. “A GIS Based Pipeline Route Selection
Process”. World Water Congress. 2004.
Maling, D.H. (1991). “Coordinate systems and map projections for GIS.
Geographical Information Systems: principles and applications.” ed. M.F.G.
Maguire D.J., D.W. Rhind. Vol. 1., Longman Group UK Ltd. 135-146. M.
Garaci J. Sutherland, B.J. Mergelas. (2002). “Role of GIS and Data
Management Systems for Pipeline Integrity.”. Pipelines.
R. Burrows G.S. Crowder, and Zhang,J.(2000). “Utilisation of Network Modelling in
the Operational Management of Water Distribution Systems”. Urban Water,
2, 83-95.
Tomas M.Walski Chase Savic (2003). “Advanced Water distribution modeling.”
Waterbury, Haestad Press.
Zeilor Michael. (2001). Modeling Our World, ed. ESRI. ESRI Press, California.

ICPTT 2009