EV Driver Behaviour Simulator 🚗⚡️
Agent-based modelling for EV charging patterns and flexibility insights
Project Details
Industry
Energy / Simulation / Data Science
Client
Energy Company (Technical Exercise)
Technologies
Project Overview
Developed a modular, agent-based simulator to model heterogeneous electric vehicle (EV) charging behaviours across diverse driver populations.
This system generates detailed individual agent traces and aggregate population statistics, culminating in an interactive Streamlit dashboard for real-time simulation execution and results exploration. The project delivers a flexible, reproducible environment focused on both individual-level traces (plug-in times, state-of-charge) and population-level insights (load flexibility, SoC distributions) – providing valuable data for energy companies and grid operators planning for EV integration.
Interactive Dashboard & Demo
Try it Live
Run simulations and explore results instantly in the web-based dashboard.
Run Locally
# Install dependencies
brew install uv && uv venv && source .venv/bin/activate && uv sync
# Launch the UI
streamlit run dashboard.pyConfigurable Parameters
- Adjust population size (10–1000 agents)
- Set simulation duration (1–30 days)
- Control random seed for reproducibility
- Select specific agent traces to inspect
- Override default configuration settings
Real-time Visualizations
- Population-level metrics and KPIs
- Individual agent state traces (SoC, plug status)
- Flexibility potential analysis
- Aggregate charging load patterns (Plotly)
- Event distributions (plug times, durations)
The Challenge
Modeling the complexity of EV charging behavior requires addressing several key challenges:
Driver Heterogeneity
Modeling diverse driver needs based on archetypes (commuters, high mileage, etc.) with varying schedules, driving patterns, battery sizes, and charger types.
Behavioural Stochasticity
Incorporating realistic randomness in daily driving distances, plug-in timing, and the probability of skipping charging sessions.
Realism vs. Simplicity
Balancing detailed behavioural modeling (like charging taper, target SoC variance) with a computationally efficient and understandable simulation core.
Insight Accessibility
Providing a non-technical, interactive way to run scenarios, visualize emergent population behaviour, and inspect individual agent states without needing to code.
Simulation Approach
Simulation Workflow
The simulation progresses step-by-step, updating each agent based on its models:
- Configuration & Initialization: Load parameters, create agents based on archetypes.
- Time-Stepped Processing: Advance simulation time incrementally.
- Agent Behavior Updates: Process scheduling, driving, charging models per agent.
- Event & Data Logging: Record state changes, events, stats for analysis.
- Results Generation: Compile outputs (load profiles, flexibility metrics).
Process Flow Diagram
Key Design Decisions
Daily Driving Energy Model
Calculates total daily energy need per agent (based on archetype, scaled, randomized) instead of individual trips, simplifying the model while retaining variability.
Stochastic Charging Behavior
Agents follow schedules but may skip charging probabilistically. SoC tapering and target SoC variance add realism.
Modularity Focus
Prioritized separation of concerns (scheduling, driving, charging) for clarity, testability, and easier modification.
Scope Prioritization
Focused on behavioral richness over grid physics (no network constraints/tariffs) for a pure Python implementation.
Key Outcomes
This project delivered several valuable outcomes for understanding EV charging behavior:
Interactive Scenario Testing
Enabled rapid exploration of how different population mixes and behavioural parameters impact aggregate charging load and flexibility via the web dashboard.
Granular Behavioural Insights
Generated detailed datasets for plug events, population statistics, and individual agent states, allowing deep dives into behaviours and model validation.
Quantified Flexibility Potential
Provided metrics on the average flexibility window per session and the percentage of sessions offering potential for demand shifting.
Balanced Realism & Usability
Achieved a balance between behavior realism, computational efficiency, and user-friendly operation valuable for diverse stakeholders.
Open Source Contribution
Released the codebase as open-source, allowing others to build upon the framework for research and applications.
Extensible Framework
Delivered a modular Python codebase suitable for further development (e.g., adding price responsivity, grid interactions).
Future Directions
Current Extensibility Points
- Add new driver archetypes via CSV data
- Integrate tariff optimization logic
- Scale simulation using Dask/Ray
- Modify behavioral models independently
- Adjust core parameters in
config.py
Potential Enhancements
- More sophisticated driving models (traffic, routing)
- Advanced charging physics (degradation, thermal effects)
- Grid interaction modeling (constraints, V2G)
- Dynamic price responsivity / Range anxiety logic
- Integration with real-world travel survey data