Intelligent EV Charging Layout Optimization via SUMO Simulation and ML

This project aims to optimize electric vehicle (EV) charging station layouts using realistic traffic simulations and machine learning (ML) models. It is built upon SUMO (Simulation of Urban Mobility) and targets Glasgow’s real road network.

---

Project Goal

Build a digital pipeline that simulates EV usage in real traffic, evaluates charging station layouts, and trains an ML model to predict the effectiveness of future layouts — enabling faster iteration without rerunning SUMO each time.

---

Overall Workflow

┌────────────────────┐
│ scenario_matrix.csv│◄────── experiment configs (EV count, cs count, layout ID, etc.)
└────────┬───────────┘
         │
         ▼
  load_scenario.py   ←── parses parameters
         │
         ▼
 clean_net.py        ←── cleans raw net.xml (removes drone, rail, etc.)
         │
         ▼
generate_cs_candidates.py
         │
         ▼
generator_charging_site.py
         │
         ▼
generator_trip.py  ←──── generates routes
         │
         ▼
generate_sumocfg.py
         │
         ▼
run_simulation.py  ←──── runs SUMO via TraCI, collects outputs
         │
         ▼
extract_layout_features.py
         │
         ▼
build_training_dataset.py  ←── combines simulation + layout for ML training

---

## File Structure
.
├── config/
│
├── data/
│   └── map/glasgow_clean.net.xml     # Cleaned map for Glasgow
|   |__ _dataset_1                      # One dataset of the experiment, like fixd EV number = 2000
|       |
│       └── scenario_matrix.csv       # Defines dataset1 configurations
|       |__ layout_registry.json      # Define  charging station layout of datatset1
│
├── sumo/
│   └── dataset_1
|       |
|       |__S001/                        # Example scenario output
│          ├── routes/                  # .rou.xml files
│          └── cs/                      # charging_stations.xml and layout_features_sample.csv
|       |__ output/                  # traci_data.csv
│
├── scripts/
│   ├── clean_net.py
│   ├── generate_cs_candidates.py
│   ├── generator_charging_site.py
│   ├── generator_trip.py
│   ├── generate_sumocfg.py
│   ├── run_simulation.py
│   ├── extract_layout_features.py
│   ├── build_training_dataset.py
│   └── load_scenario.py

---

Output Dataset Format (for ML)

After simulation + feature extraction, the ML-ready training dataset will include:

scenario_id	cs_layout_id	num_cs	layout_features (e.g. coords, spread)	avg_wait_time	avg_charge_time
S001	cs_group_001	13	[x1,y1,x2,y2,...]	142.5 s	370.2 s

---

Potential Use Cases

• Predict best-performing charging layouts without rerunning SUMO
• Use ML model to explore large layout design space
• Fine-tune layout strategies for large-scale deployment cities

---

Requirements

• Python ≥ 3.10
• SUMO (with TraCI)
• pandas, numpy, scikit-learn
• Optional: torch, xgboost for modeling

---

Author

Tingting Yang, MSc Computing Science (University of Glasgow) Special focus: intelligent systems, mobility simulation, AI companionship research

For contributions, issues or guidance, feel free to contact or open a pull request.

---