A hybrid neuro-fuzzy deep reinforcement learning framework for non-convex energy system optimization in modern electrified grids

The global electricity system is undergoing a rapid transformation driven by large-scale electrification, accelerated deployment of renewable energy, and the rapid expansion of artificial intelligence (AI) infrastructure. These developments introduce substantial operational complexity, including non-convex system dynamics, high variability in renewable energy sources, and significant uncertainty in electricity demand. To address these challenges, this study proposes a Hybrid Neuro-Fuzzy Deep Reinforcement Learning (HN-FDRL) framework for optimizing stochastic energy systems. The approach integrates an Adaptive Neuro-Fuzzy Inference System (ANFIS) with a Twin Delayed Deep Deterministic Policy Gradient (TD3) architecture, enabling interpretable and stability-aware policy learning for high-dimensional dispatch problems. Realistic operating conditions are captured using Latin Hypercube Sampling combined with Copula-based dependency modeling, representing correlated renewable variability and demand uncertainty, including extreme low-renewable “Dunkelflaute” events and AI-driven load surges. Simulation results show that HN-FDRL reduces expected unserved energy by 80.8%, decreases negative electricity prices by 42%, and achieves 30% faster learning convergence compared with deterministic optimization and standalone DRL methods, while maintaining stable storage dispatch policies. These findings demonstrate that hybrid neuro-fuzzy reinforcement learning provides a scalable, interpretable, and robust decision-support framework for managing next-generation electricity systems characterized by high renewable penetration and rapidly evolving demand.