New AI framework ARMD reduces EV charging risk by 71% during disasters
Multi-agent reinforcement learning coordinates mobile charging trucks in hurricanes, cutting risk exposure dramatically.
During large-scale evacuations, electric vehicle (EV) charging demand can overwhelm fixed stations, leading to dangerous delays. To solve this, researchers introduced the Adaptive Risk-aware MCT Deployment (ARMD) framework, which dynamically routes mobile charging trucks (MCTs) to complement fixed stations. ARMD splits the problem into two parts: risk-aware allocation of MCTs among stations (solved as a decentralized partially observable Markov decision process) and dynamic routing. It uses multi-agent proximal policy optimization (MAPPO) to train a coordination policy offline in an evacuation simulator, then refines it online based on real-time conditions. A spatio-temporal travel time predictor enables rolling-horizon route updates.
The framework was tested in a simulated hurricane evacuation built with real-world data from Hillsborough County, Florida. ARMD consistently outperformed offline optimization, online heuristic dispatch, and rolling-horizon methods. Under demand perturbations, it reduced average risk exposure by up to 71.1% compared to baseline without MCTs. In scenarios with fixed charger failures or road link disruptions, ARMD achieved 39.3%–60.5% reduction, with greater advantages as disruption severity increased. These results highlight how AI-driven, adaptive mobile charging can significantly enhance evacuation safety for EV users.
- ARMD uses MAPPO to coordinate multiple mobile charging trucks via offline training and online refinement.
- Tested in a Hurricane evacuation simulation using real data from Hillsborough County, Florida.
- Reduces risk exposure by up to 71.1% for demand perturbations and 39–60% under infrastructure failures.
Why It Matters
As EV adoption grows, this AI solution ensures resilient charging during crises, potentially saving lives.