PBI Based Multi-Objective Optimization via Deep Reinforcement Elite Learning Strategy for Micro-Grid Dispatch With Frequency Dynamics
Abstract
Micro-grid dispatch is facing a great challenge since the system security and economic cost are often time-varying with different requirements. In this paper, a novel multiple time-scale dispatch model is firstly constructed with considering load voltage and frequency dynamics of islanding awareness, which is limited with rate-of-change of frequency (RoCoF) of micro-grid. Based on this model, micro-grid power dispatch is converted into a multi-objective optimization problem (MOP) with taking economic cost, voltage deviation and frequency stability into consideration. To address this problem, a penalty-based boundary intersection (PBI) based multi-objective optimization approach is developed with an elite learning technique. In the above approach, a developed deep deterministic policy gradient (DDPG) is proposed to learn evolutionary parameters with grid based alternating weight vectors, which can lead to better convergence ability and diversity distribution. Simulation results show that the proposed optimization strategy can properly achieve minimum economic cost with simultaneously satisfying voltage and frequency stability, which provides a viable and promising way for tackling with micro-grid dispatch problem.