基于深度神经网络的DFIG低电压穿越技术研究  

Research on Low Voltage Ride Through Technology of DFIG Based on Deep Neural Networks

作　　者：余欣梅 陈豪君 王星华[2] YU Xinmei;CHEN Haojun;WANG Xinghua(China Energy Engineering Group Guangdong Electric Power Design Institute Co.,Ltd.,Guangzhou 510663,China;School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]中国能源建设集团广东省电力设计研究院有限公司,广州510663 [2]广东工业大学自动化学院,广州510006

出　　处：《南方能源建设》2021年第3期122-130,共9页Southern Energy Construction

基　　金：国家自然科学基金项目“基于张量技术的多视图特征选择方法研究”(61903091)。

摘　　要：[目的]双馈风机(DFIG)的低电压穿越(LVRT)性能在一定程度上依赖于控制参数的优化,而目前对控制参数的优化基本都是离线模式,原因在于优化算法难以满足实时控制对计算速度的要求。[方法]基于深度神经网络(DNN)原理,提出基于“离线训练、在线计算”思路的低电压穿越实时优化控制方法。首先针对含DFIG电网在不同运行方式下发生的大量不同故障进行控制参数的离线优化并形成相应的LVRT优化控制策略,将不同的故障状态进行归类并形成成对的故障模式和参数表,从而构成深度神经网络的训练样本。[结果]电网故障瞬间可以将输入参数直接通过训练好的DNN网络迅速实现控制方案和最优参数的优选。[结论]基于PSCAD和Matlab的联合仿真结果论证了所提思想在优化效果和优化速度方面的优势,并说明了其实用性。[Introduction]During the power grid fault,the low voltage ride through(LVRT)performance of the doubly-fed induction generator(DFIG)depends on the control parameters.At present,the optimization of control parameters is basically in the offline mode,which lies in the fact that it's hard for algorithm optimization to meet real-time control's requirement of the calculation speed.[Method]Therefore,the real-time optimization control method of LVRT following“offline training,online computation”was presented based on the principles of deep neural networks(DNN).Firstly,the appropriate LVRT strategy for optimization control was proposed for different fault severity levels.The parameters were clustered and optimized according to the respective objectives of each strategy,then the parameter list was formed.[Result]At the moment of power grid fault,the input parameters can be directly input into the trained DNN networks to quickly realize the optimization of control scheme and optimal parameters.[Conclusion]The joint simulation results based on PSCAD and MATLAB demonstrate the advantages of the proposed idea in optimization effect and optimization speed and the practicability is also illustrated.

关 键 词：双馈风机 低电压穿越 深度神经网络 

分 类 号：TM7[电气工程—电力系统及自动化] TM315