Coordinating heterogeneous energy storage units in islanded hybrid microgrids presents a critical control challenge, particularly when ensuring robust operation under severe renewable uncertainty and parameter mismatches. Existing decentralized schemes often fail to prevent persistent State-of-Charge (SoC) divergence among unequal battery banks, while conventional model-based optimization suffers from heavy computational burdens. This paper proposes a data-driven dynamic-equilibrium adaptive critic (DEAC) framework to solve the optimal energy management problem without explicit system dynamics. The core innovation lies in defining a capacity-weighted energy centroid, which constructs a dynamic equilibrium manifold for heterogeneous batteries. By embedding this manifold into a Hamilton–Jacobi-Bellman (HJB) formulation, we transform the multi-objective coordination problem into a tractable optimal tracking task that intrinsically synchronizes asynchronous charge/discharge cycles. A streamlined critic-only learning architecture is developed to approximate the value function, significantly reducing online computational complexity by eliminating redundant actor networks while theoretically guaranteeing uniformly ultimately bounded (UUB) stability of the closed-loop system. Validated against real-world load/generation profiles, the proposed strategy demonstrates superior adaptability compared to rule-based and standard actor–critic baselines. Results indicate a 12.55% reduction in fuel consumption and a 3.6% extension in battery cycle life proxies, with robustness stress tests confirming reliable performance under 10% forecast errors and communication delays. • Data-driven adaptive critic framework optimizes storage without explicit system models. • Capacity-weighted centroid mechanism coordinates heterogeneous battery banks effectively. • Streamlined critic-only architecture reduces computational complexity for online implementation. • Rigorous Lyapunov analysis ensures Uniformly Ultimately Bounded (UUB) weight convergence. • Systematic stress tests verify robustness against parameter mismatches and communication delays.
Wei et al. (Sun,) studied this question.