基于强化学习算法的自适应直流附加阻尼控制器

提出了基于强化学习算法的直流附加阻尼控制器。控制器主体采用模糊神经网络,利用由系统性能指标生成的强化信号在线训练控制器参数。与传统的模糊控制器相比,由于该控制器采用自适应启发式评价算法,将系统输出性能指标转化为强化信号反馈给控制器,使其能够在线修改控制器参数,因此有效地克服了传统阻尼控制器的设计对系统精确数学模型的依赖。仿真结果表明,与传统的阻尼控制器相比,基于强化学习算法的直流附加阻尼控制器能够有效地抑制区域间的功率振荡,提高交直流系统的动态稳定性,并且对多种运行方式具有一定的鲁棒性。

Adaptive HVDC supplementary damping controller based on reinforcement learning

An adaptive supplementary damping controller of HVDC is presented based on reinforce-ment learning algorithm,which is mainly composed of neuro-fuzzy network using reinforcement signal to train the parameters of controller. Contrary to conventional fuzzy controller,the reinforcement signal,transformed from the output performance index of power system by the adaptive heuristic assessment algorithm,is fed back to the controller to update the key parameters, which effectively reduces the dependence of damping controller on accurate mathematical model. Simulation results show that this supplementary controller efficiently damps the power oscillation among areas,improves system stability,and has better robustness in various operation modes than the conventional supple-mentary damping controller.