摘要
为提升可控换相换流器在复杂电网环境中的控制性能,提出一种基于深度强化学习与神经模糊系统的换流器混合自适应控制框架。首先,设计基于深度Q网络算法的动态控制策略;其次,利用在线学习机制动态调整模糊规则,构建自适应神经模糊控制策略;最后,采用全局-局部动态调节与双层学习的协同工作机制,实现全局优化控制和局部精细调节的融合。仿真试验结果表明,相比比例-积分控制器、模糊控制器、自适应神经模糊控制器和深度强化学习控制器,所提混合自适应控制框架在所有测试场景中的响应速度、能量效率、总谐波失真以及稳态误差均表现最佳。
To improve the control performance of controllable commutation converters in complex power grid environments,a hybrid adaptive control framework for converters based on deep reinforcement learning and neuro-fuzzy systems was proposed.Firstly,a dynamic control strategy was designed based on the deep Q-network algorithm;secondly,online learning mechanism was utilized to dynamically adjust fuzzy rules and construct adaptive neuro-fuzzy control strategies;finally,a collaborative mechanism of global local dynamic adjustment and two-layer learning was adopted to achieve the integration of global optimization control and local fine adjustment.The simulation test results show that compared with proportional-integral controller,fuzzy controller,adaptive neuro-fuzzy controller,and deep reinforcement learning controller,the proposed hybrid adaptive control framework performs the best in response speed,energy efficiency,total harmonic distortion,and steady-state error in all test scenarios.
作者
周亮
任佳丽
张俊
查鲲鹏
刘虹
Zhou Liang;Ren Jiali;Zhang Jun;Zha Kunpeng;Liu Hong(College of Electrical Engineering,Zhejiang University,Hangzhou Zhejiang 310058,China;China Electric Purui Power Engineering Co.,Ltd.,Beijing 102200,China)
出处
《电气自动化》
2025年第5期47-49,53,共4页
Electrical Automation
关键词
可控换相换流器
控制策略
深度强化学习
自适应神经模糊控制
深度Q网络
在线学习机制
controllable commutation converter
control strategy
deep reinforcement learning
adaptive neuro-fuzzy control
deep Q-network
online learning mechanism
