An intelligent energy management strategy for a microgrid based on proximal policy optimization

An intelligent energy management system is crucial to the optimal and cost-effective operation of a modern microgrid. Generally, distributed renewable energy in a microgrid exhibits strong intermittency and uncertainty. We present a proximal policy optimization strategy enhanced by stochastic programming for a wind-solar-diesel-storage microgrid. Wind power is first predicted by the gated recurrent unit (GRU). Latin hypercube sampling is utilized to generate probable wind power errors, which are then synthesized with the GRU prediction to obtain scenario sets of wind power under different probabilities. The complex optimization problem is formulated as a Markov decision process with a high-dimensional action and state space, enabling proximal policy optimization to determine the optimal day-ahead power for each unit by minimizing the daily economic cost while satisfying the microgrid security indicators. Simulations demonstrate that the intelligent energy management system is efficient and robust for practical microgrid operation.