Active Voltage Control and Demand Peak-Clipping Using Soft Actor-Critic Method

Abstract

This study presents a proprietary Soft Actor-Critic (SAC) algorithm for power distribution network active voltage control and demand peak-clipping, including on-load-tap-changing and direct load control. Voltage violation and transformer loads, determine the system's state, while actions comprise transformer tap changing, heat pump setting temperature, and Vehicle-to-Grid at 30% demand flexibility and rewards are designed to prevent voltage violations and transformer overload. After testing with various load levels, the trained agent can optimally select actions and minimise violations, however the model will be tuned with different parameters and demand shift and energy storage are suggested to be added in future work.