Balancing Cost and Self-Consumption in Home Energy Management: A Deep Reinforcement Learning and Model-Based Perspective with EV and Storage Systems

This study presents a comparative analysis of two home energy management system (HEMS) strategies: a model-based mixed-integer linear programming (MILP) approach and a reinforcement learning-based Deep Q-Network (DQN), under EV-PV and EV-PV-Battery Energy Storage System (BESS) configurations. The MILP approach serves as the optimality benchmark to assess the effectiveness of the proposed DQN strategy. A multi-objective function is introduced to balance cost minimization and self-consumption, allowing user-defined prioritization. Results show that MILP consistently outperforms DQN in cost savings and self-consumption. In the EV-PV setup, both methods achieve around 20% cost savings, though weekday EV charging restrictions limit self-consumption gains. Adding a BESS improves performance, with MILP reaching 28.16% cost savings and 21.59% self-consumption, compared to DQN’s 19.77% and 10.05%. While MILP offers superior performance, DQN’s adaptability to uncertainty makes it well-suited for real-time, dynamic environments.