关键词: 改进深度强化学习模型, 火电厂, 静止无功补偿器, 动态无功补偿, 优化方法

Abstract:

When optimizing reactive power compensation of static var compensator (SVC) in thermal power plants, taking the internal active power demand of SVC as the threshold will cause deviation of voltage safety margin and aggravate voltage amplitude fluctuation, thereby deteriorating the compensation performance. Accordingly, an optimization method for dynamic reactive power compensation of SVC in thermal power plants based on an improved deep reinforcement learning (DRL) model is proposed. The DRL model is improved with the Dueling architecture, and the Dueling network structure is adopted to evaluate state values more accurately so as to raise the safety margin of the access voltage of SVC. The minimization of voltage deviation is taken as the objective function to calculate the optimal voltage amplitude for dynamic reactive power compensation of SVC in thermal power plants and realize optimized dynamic reactive power regulation. Experimental results demonstrate that the optimized voltage amplitude fluctuates between 0.97~1.00 p.u. with slight fluctuation and favorable compensation performance, which is of great significance for guaranteeing the stable operation of thermal power plant equipment.

Key words: improved deep reinforcement learning(DRL) model, thermal power plants, static var compensator(SVC), dynamic reactive power compensation, optimization methods