基于聚类LSTM深度学习模型的主动配电网电能质量预测  

Power quality prediction of active distribution network based on clustering LSTM deep learning model

作　　者：翁国庆[1] 龚阳光 舒俊鹏 黄飞腾[1] Weng Guoqing;Gong Yangguang;Shu Junpeng;Huang Feiteng(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023)

机构地区：[1]浙江工业大学信息工程学院,杭州310023

出　　处：《高技术通讯》2020年第7期687-697,共11页Chinese High Technology Letters

基　　金：国家自然科学基金(51777193);浙江省自然科学基金(LY17E070005)资助项目。

摘　　要：针对较长时间跨度上电能质量(PQ)数据的时序性和非线性特点,提出一种基于K-means聚类和长短期记忆(LSTM)网络的主动配电网(ADN)电能质量预测方法。在构建LSTM深度学习模型的基础上,将大量的电能质量历史数据、环境因素及负荷数据以多维向量的形式进行K-means聚类,并针对每一类数据集分别使用LSTM模型进行网络的训练和性能评估,然后利用完成训练和评估的聚类LSTM网络模型进行主动配电网电能质量稳态指标项的预测。最后,通过IEEE-13节点含分布式电源的主动配电网仿真算例,分析验证了所提聚类LSTM网络法比时间序列预测法、反向传播(BP)神经网络法和标准LSTM网络法具有更优的预测性能。Aiming at the time series and nonlinear characteristics of power quality(PQ)data over a long time span,an active distribution network(ADN)power quality prediction method based on K-means clustering and long-term short-term memory(LSTM)network is proposed.Based on the LSTM deep learning model,a large number of power quality historical data,environmental factor data,and load data are clustered by K-means in the form of multi-dimensional vectors,next using the LSTM deep learning models to train and evaluate for each category of data,and then using the clustered LSTM network model of the completed training and evaluating to predict the data of the steady-state indicators of the power quality of active distribution network in the future.Finally,through the IEEE-13 node active distribution network simulation example with distributed power supply,the simulation results show that the proposed clustering LSTM network method has more significant prediction performance than the time series prediction method,back propagation(BP)neural network method and standard LSTM network method.

关 键 词：电能质量(PQ)预测 深度学习 长短期记忆网络(LSTM) K-MEANS聚类 主动配电网(ADN) 

分 类 号：TM711[电气工程—电力系统及自动化]