Optimal Frequency Regulation in Multi-Microgrid Systems using Federated Learning

Abstract

This paper proposes a Federated Learning-based Zeroing Neural Network (FL-ZNN) tuned optimal proportional-integral-derivative (PID) control strategy for frequency control of Multi-Microgrid (MMG) system. The proposed FL-ZNN technique employs a distributed learning approach that allows each neuron to train the network based on its own local data. The local models are then aggregated into a global model, which is used to update the neurons of the network to auto-tune the PID controller's parameters in each microgrid. The proposed FL-ZNN-based PID controller is able to provide robust and efficient frequency control in MMG under different operating conditions, including successive load variations and communication delay. Simulation results demonstrate the effectiveness and superiority of the proposed FL-ZNN-based control strategy over the ZNN PID, and conventional ZNN controller in terms of response time, overshoot, and settling time. Further, the proposed controller has been validated using Hardware-in-the-Loop (HIL) in OPAL-RT.