Load Frequency Control of Multi-Microgrids Based on Deep Deterministic Policy Gradient Integrated With Online Learning

The integration of wind power into microgrids significantly increases the complexity of the microgrid’s dynamical behavior and introduces higher levels of uncertainty. This paper addresses load frequency control (LFC) in wind-diesel multi-microgrids by exploiting a multi-agent deep deterministic policy gradient (MADDPG) method. The proposed method derives coordinated control strategies for multiple LFC controllers by combining offline centralized learning with online learning. During offline centralized learning, a neural network based on Long-Short-Term Memory (LSTM) is applied to learn the dynamical behavior of the multi-microgrids. The LSTM-based network updates actor networks to estimate the optimal control policy from the global action-value function. Online learning integrates with offline training, allows the LSTM-based network to adapt to dynamic changes of the microgrid and bridges the gap between training and real-time application. Simulations on a wind-diesel three-microgrid and the New England 10-generator 39 test bus validate the method’s effectiveness in improving the control performance of the microgrid.