Deep entropy learning for multi-energy cooperation system with non-dispatchable generation and storage unit under load shedding

Abstract

This study explores the resilience of a multi-energy cooperation generation (MECG) system with electricity, natural gas networks, and sustainable energy units (SEUs). The wind farm generation (WFG), as the non-dispatchable unit, is adopted to supply the grid for the profit of customers. In particular, an innovative entropy learning-based soft-actor critic is introduced to assess system resilience against low-probability but high-destruction events. The suggested framework was validated through empirical analysis using the IEEE 24-bus network and the Belgian 20-node gas network equipped with wind turbine and battery unit. In deep entropy learning, the problem of the MECG system is modeled based on the Markovian decision process (MDP) to solve the optimization problem by interacting the agent with the complex network (environment). By maximizing a reward function, the deep neural network (DNN) of deep entropy learning is trained so that optimizes the complex MECG system with WFG and battery storage unit (BSU). Our findings illuminate the potential efficiency gains and the enhanced adaptive capacity achievable through strategic integration, providing actionable insights for policymakers, engineers, and researchers. By contributing to the discourse on resilient and sustainable energy systems, this study addresses the urgent need for robust energy infrastructures capable of withstanding today's dynamic environmental and operational landscapes.