EV Charging Network Resilience Under Power Outages

Analysis of electric vehicle (EV) charging infrastructure vulnerability to power outages and coordinated infrastructure attacks across the United States.

Overview

This project investigates resilience of EV charging networks through:

• Network analysis on AFDC charging station data (2018-2026, 49 states)
• Bayesian power outage modeling using 10 years of historical EAGLEI data (2014-2023)
• Attack simulations combining targeted node removal with realistic outage scenarios

Project Structure

analysis/
├── run_charging_network.py              # Main network construction & attack simulation
├── plot_charging_network_results.py     # Results visualization
├── build_network_pickle_only.py         # Network serialization
│
├── attack_under_PO/                     # Integrated attack + outage analysis
│   ├── monte_carlo_outage_attack.py     # Monte Carlo simulations
│   ├── realistic_outage_simulator_improved.py
│   ├── compute_impact_radius.py
│   └── module_3_03/                     # Advanced simulation module
│
└── power_outage/
    ├── PO frequency modeling/           # Bayesian hierarchical model for outage frequency
    └── PO severity modeling/            # Bootstrap resampling for outage severity

notebooks/
└── PO_data_cleaning/                    # Data preprocessing & EDA

Quick Start

1. Build charging networks

# Single year
python analysis/run_charging_network.py "data/raw/alt_fuel_stations_historical_day (Jan 1 2025).csv"

# Batch process all years (2018-2026)
python analysis/run_charging_network.py --batch-from-raw

Method: 200m spatial clustering → Voronoi adjacency → weighted network graph

2. Simulate attacks under power outages

cd analysis/attack_under_PO/
python monte_carlo_outage_attack.py --state CA --n-sims 1000

3. Visualize results

python analysis/plot_charging_network_results.py --run-dir outputs/charging_network/run__*

Requirements

pip install numpy scipy pandas networkx scikit-learn pyproj matplotlib seaborn arviz pymc3

• Python 3.7+
• Core: numpy, scipy, pandas, networkx
• Network: scikit-learn, pyproj
• Bayesian: arviz, pymc3
• Visualization: matplotlib, seaborn

Key Findings

Network Characteristics (2025):

• Analyzed 49 US states with 10,000+ charging stations
• Built weighted graphs with Voronoi adjacency
• Computed network efficiency, clustering, degree distribution

Attack Resilience:

• Targeted attacks (degree/betweenness) cause 3-5× more damage than random failures
• Small networks (MT, WY, VT) show <3 alternative paths per node
• Large networks (CA, TX, NY) demonstrate higher redundancy

Outputs:

• network_characteristics_ALL.csv: Per-state metrics (nodes, edges, efficiency, clustering, density)
• attack_results_ALL.csv: Efficiency loss by strategy, removal %, and node removal strategy
• State-level vulnerability rankings and geospatial impact maps

Algorithm Details

Network Construction

1. Spatial Clustering: 200m greedy aggregation of charging station locations
2. Graph Construction: Voronoi diagram adjacency (Delaunay fallback)
3. Edge Weighting:

   w_ij = (norm_distance + ε) / √(capacity_i × capacity_j)
   

Attack Simulation

• Strategies: Random, Degree, Betweenness, Capacity, Closeness centrality
• Metrics: Global efficiency loss at removal percentages [5%, 10%, ..., 85%]
• Reproducibility: Random attacks run 10 times with fixed seed

Power Outage Modeling

• Frequency: Bayesian hierarchical GLM (ADVI) for county-level event rates
• Severity: Bootstrap resampling from historical EAGLEI database
• Integration: Monte Carlo simulation of outage impact on network resilience

License

Research use - Available for academic and educational purposes

References

• AFDC (Alternative Fuels Data Center): www.afdc.energy.gov
• EAGLEI Database: Historical power outage records
• NetworkX documentation for graph algorithms