中文版本 ↑返回顶部
南京大学大学生创新计划项目，开发基于强化学习的无人机自主导航系统，用于：
- 追踪大气甲烷羽流至高斯中心
- 定位点源位置（误差<5米）
- 反演排放通量（误差<20%）
├── ppo1.0/
│ ├── ppo_basic/ # 标准PPO算法实现
│ ├── fixed_threshold/ # 经验浓度阈值（800-1200ppb停止）
│ └── gaussian_env/ # 高斯羽流仿真环境
- 特点：首次实现PPO与化学阈值停止的融合
├── ppo2.0/
│ ├── lstm_module/ # 浓度时间序列预测器
│ ├── dynamic_stop/ # 动态停止阈值（500-1500ppb）
│ └── nc_analyzer/ # 分析训练输出的阈值优化
- 改进：LSTM预测最优停止阈值（测试集R²=0.82）
├── ppo2.1/
│ ├── gradient_detec/ # 基于∇[CH₄]的源定位
│ └── trend_predict/ # 通过dC/dt模式确认源区
- 突破：完全摒弃固定阈值，采用微分趋势分析
| 组件 | 实现细节 |
|---|---|
| 羽流模型 | 高斯扩散模型（σ_y=0.3x^0.71） |
| 状态空间 | [CH₄]、风速矢量、无人机位置 |
| 奖励函数 | R = Δ[CH₄] - 0.2‖Δθ‖ |
| 训练硬件 | NVIDIA RTX 3090（3840 CUDA核心） |
English Version ↑Back to Top
Nanjing University Innovation Program developing reinforcement learning UAV system for:
- Tracing methane plumes to Gaussian centers
- Locating point sources (<5m error)
- Quantifying emission fluxes (<20% error)
├── ppo1.0/
│ ├── ppo_basic/ # Standard PPO implementation
│ ├── fixed_threshold/ # Empirical stop threshold (800-1200ppb)
│ └── gaussian_env/ # Gaussian plume simulation
- Key Feature: First integration of PPO with chemical threshold stopping
├── ppo2.0/
│ ├── lstm_module/ # Concentration time-series predictor
│ ├── dynamic_stop/ # Adaptive stopping threshold (500-1500ppb)
│ └── nc_analyzer/ # Threshold optimization from training outputs
- Improvement: LSTM predicts optimal stop threshold (R²=0.82)
├── ppo2.1/
│ ├── gradient_detec/ # Source localization via ∇[CH₄]
│ └── trend_predict/ # Source confirmation through dC/dt patterns
- Breakthrough: Eliminates fixed thresholds using derivative analysis
| Component | Implementation Details |
|---|---|
| Plume Model | Gaussian dispersion (σ_y=0.3x^0.71) |
| State Space | [CH₄], wind vector, UAV position |
| Reward Function | R = Δ[CH₄] - 0.2‖Δθ‖ |
| Training Hardware | NVIDIA RTX 3090 (3840 CUDA cores) |
📝 Citation 引用格式
Nanjing University CH₄ UAV Team. (2023). Autonomous Plume Tracing System. Student Innovation Program.
```
Nanjing University CH₄ UAV Team. (2023). Autonomous Plume Tracing System. Student Innovation Program.