Quantum-Enhanced DDQN for Hybrid Energy Storage Decision Optimization in Islanded Microgrids

This paper proposes a Quantum-Machine-Learning-enhanced Double Deep Q-Network (QML-DDQN) for the supervisory control of battery–supercapacitor hybrid energy storage systems in islanded microgrids. This method combines a variational quantum circuit as a nonlinear state encoder with a DDQN decision layer for safe discrete dispatch. Three representative islanded cases, Island 1, Island 2, and Island 3, were used to evaluate the robustness under different scales, renewable profiles, and reliability requirements. Compared with deterministic optimization, predictive control, metaheuristics, and classical reinforcement-learning baselines, the proposed controller delivers the best overall trade-off among operating cost, renewable utilization, diesel reduction, and loss-of-power-supply risk. On the three-case averages, QML-DDQN reduces daily cost and LPSP by 0.99% and 4.04% relative to DDQN, by 2.91% and 7.32% relative to DQN, and by 9.09% and 16.63% relative to MILP; it also lowers curtailment and diesel share by up to 13.02% and 9.09%, respectively, across the same benchmark sets. The largest gains appear under volatility-dominated and stress-scenario conditions, where the quantum encoder strengthens the state representation, and the DDQN backbone mitigates value overestimation. These results highlight the practical advantages of the QML-DDQN as a resilient and high-value supervisory strategy for islanded hybrid energy storage operations.