Post-Disaster Home Energy Management with Fuel Cell Vehicles: Evaluating Reinforcement Learning Against MILP Under Islanding Conditions
8th Global Power, Energy and Communication Conference, GPECOM 2026, Naples, Italy, 3 - 05 June 2026, pp.1058-1063, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/gpecom70462.2026.11578574
- City: Naples
- Country: Italy
- Page Numbers: pp.1058-1063
- Keywords: fuel cell electric vehicle, home energy management, mixed-integer linear programming, reinforcement learning
- Yıldız Technical University Affiliated: Yes
Abstract
Natural disasters and extreme weather events increasingly cause prolonged residential power outages, motivating resilient home energy systems that integrate photovoltaic (PV) panels and fuel cell electric vehicles (FCEVs) leveraging both the vehicle's onboard battery and hydrogen tank in vehicle-to-home (V2H) mode. This paper formulates both a mixedinteger linear programming (MILP) model and a proximal policy optimization (PPO)-based reinforcement learning (RL) controller for real-time energy dispatch in such systems and evaluates them under identical conditions. Results show that MILP achieves the global optimum (average load level 1.33, cumulative comfort score 3,360) in 47.3 s, while PPO-based RL achieves an average load level of 0.33 (cumulative score 1,920) with sub-millisecond dispatch. The comparison demonstrates the trade-off between optimality and real-time adaptability across offline and online control paradigms.