An Integrated Generation-Compensation optimization Strategy for Enhanced Short-Term Voltage Security of Large-Scale Power Systems Using Multi-Objective Reinforcement Learning Method

摘要

High penetrations of industrial loads have placed significant pressures on short-term voltage security. This paper proposes an integrated generation-compensation optimization strategy, which coordinates the generators and the switchable capacitor banks to enhance short-term voltage security as a multi-objective dynamic optimization (MODO) model. This model is established containing dynamics, power flow balances, and security constraints to minimize the voltage deviation and the cost of control strategy. The differential-algebra equations are converted into algebra equations using the Radau collocation method. Furthermore, a novel multi-objective reinforcement learning (MORL) method is utilized to mitigate the computational burdens and to obtain Pareto optimal solutions by filtering the dominated solutions. Compared with conventional MORL methods, the proposed MORL method divides the full feasible region into several small independent regions to reduce and eliminate the searching for Pareto optimal solutions. Meanwhile, the state functions of MORL are redefined, and the state sensitivities are introduced to judge whether the trial and learning accumulate sufficient knowledge. Moreover, the Pareto optimal solutions are further improved by introducing several possible solutions. Finally, the tradeoff solution is obtained based on Fuzzy decision-making strategy. The effectiveness and efficiency of the MORL method are verified by numerical simulations on a provincial 748-bus power system.