融合寻优算法的双馈风力机控制参数分步辨识方法
STEPWISE IDENTIFICATION METHOD OF CONTROL PARAMETERS FOR DFIG BASED ON COMPOSITE OPTIMIZATION ALGORITHM
徐恒山 1李颜汝 1李文昊 1薛飞 2王伟3
作者信息
- 1. 三峡大学电气与新能源学院,宜昌 443002
- 2. 国网宁夏电力有限公司电力科学研究院,银川 750001
- 3. 中国南方电网有限责任公司超高压输电公司贵阳局,贵阳 550081
- 折叠
摘要
为获得准确的双馈风力机(DFIG)控制参数以提高电力系统机电/电磁仿真分析和计算的准确性,将长短期记忆(LSTM)神经网络与改进粒子群(IPSO)算法相结合对DFIG的控制参数进行辨识.首先,利用RT-LAB平台通过硬件在环(HIL)实验获得真实DFIG控制器的响应数据集;其次,为避免无关特征干扰LSTM模型的预测结果,利用最大信息系数提取出DFIG中高相关性的观测量特征;在此基础上,为提高算法的寻优速度,利用LSTM初步寻优到DFIG控制参数的初始值与搜索范围;最后,通过IPSO算法精确辨识出DFIG的控制参数,提高了辨识算法的寻优效率和精度.HIL测试结果证实了LSTM-IPSO辨识方法在20%～80%低电压穿越工况下具有良好的适应性,并能有效提高DFIG控制参数的辨识精度.
Abstract
To precisely obtain the control parameters of doubly fed induction generator(DFIG)and improve the accuracy of electromechanical/electromagnetic simulation for the power system,a composite optimization algorithm with combining long short-term memory(LSTM)neutral network and improved particle swarm(IPSO)algorithm is used to identify the control parameters of DFIG.Firstly,the response data of real DFIG controller is obtained by hardware-in-loop(HIL)based on RT-LAB platform.Secondly,the observable quantities that have high relevance of DFIG are extracted through maximum information coefficient to avoid the effect of irrelevant features on the predicting results of LSTM model.On the above basis,in order to improve the optimization speed of the algorithm,LSTM is used to preliminary optimize the initial value and search range of DFIG control parameters.Finally,IPSO algorithm is used to precisely identify the control parameters of DFIG,which improves the searching efficiency and accuracy.The HIL testing results verify that LSTM-IPSO identification method has well adaptability and high identification accuracy to the control parameters under the 20%-80%low voltage ride through(LVRT)conditions.
关键词
风电机组/长短时记忆网络/粒子群算法/参数辨识/最大信息系数Key words
wind turbines/long short-term memory network/particle swarm optimization/parameter identification/maximum information coefficient引用本文复制引用
基金项目
国家重点研发计划(2017YFE0132100)
国家自然科学基金(52067001)
出版年
2024
NETL
NSTL
万方数据