An LSTM-DDPG framework power management strategy for a heterogeneous energy storage system in a standalone DC microgrid

Abstract

A heterogeneous energy storage system (HESS) is implemented to combat the DC bus voltage instability and power allocation problem caused by high penetration of renewable energy sources (RESs) in a standalone DC microgrid. The HESS comprises a battery and supercapacitor aims to smooth DC bus voltage. Thus, a potent ongoing trend is the development of intelligent power management strategies (PMS) that boost the efficacy of RESs-based DC microgrids using deep reinforcement learning (DRL) techniques. In particular, long short-term memory (LSTM) is incorporated into a deep deterministic policy gradient (DDPG) framework to tackle real-world microgrid power management problems. This method uses DDPG for power allocation decisions, while LSTM is applied to extract environmental state variables from observations. Implementing an LSTM-DDPG PMS quantifies transient performance and minimises voltage deviations from the rated DC bus voltage while maintaining battery and supercapacitor state of charge (SOC) within defined limits to prevent overcharging and excessive discharging cycles. Furthermore, the HESS’s power flow is dynamically regulated by optimal current controllers, which facilitates efficient PMS and enhances overall system stability by precisely tracing respective currents. The effectiveness of the proposed PMS is assessed under different solar, wind, and load power situations. The simulation results demonstrate a substantial decrease in highest disparities of the DC bus voltage compared to traditional PMS approaches. The suggested technique is confirmed effective by real-time validation of the simulation outcomes with the OPAL-RT real-time simulator.