基于相空间重构和BiLSTM的风电功率短期预测  

Short-term Forecasting of Wind Power Based on Phase Space Reconstruction and BiLSTM

作　　者：邓韦斯 卢斯煜 刘显茁 戴仲覆 田伟达 张旭东 杨秋玲 DENG Weisi;LU Siyu;LIU Xianzhuo;DAI Zhongfu;TIAN Weida;ZHANG Xudong;YANG Qiuling(CSG Power Dispatching Control Center,Guangzhou,Guangdong 510663,China;State Key Laboratory of HVDC,CSG Electric Power Research Institute Co.,Ltd.,Guangzhou,Guangdong 510663,China)

机构地区：[1]中国南方电网电力调度控制中心,广东广州510663 [2]直流输电技术国家重点实验室(南方电网科学研究院有限责任公司),广东广州510663

出　　处：《广东电力》2023年第7期22-30,共9页Guangdong Electric Power

基　　金：中国南方电网有限责任公司科技项目(ZDKJXM20210047)。

摘　　要：针对复杂多因素(气象信息、时间序列的混沌特性等)影响风电功率的短期预测,及风电时间序列的长期依赖问题,提出基于相空间重构和双向长短期记忆(bidirectional long short-term memories,BiLSTM)神经网络的风电功率短期预测方法。以全球能源预测竞赛的数据集为背景,基于嵌入定理从风电功率序列中重构出相空间,以展示其内在的混沌特性,其中相空间重构的参数依据C-C法确定;对选取的气象预测数据(未来风速、风向)进行归一化处理,并组合重构后的风电功率数据作为BiLSTM的输入量,重构前的功率数据作为输出量,训练预测模型。在全球能源预测竞赛2012提供的wf1数据集上进行日前预测实验,测试集前30 d的平均均方根误差为0.1194,测试集107 d的平均均方根误差为0.1409,相较于ANN、BiLSTM、RF和KNN,相空间重构-BiLSTM(Re-BiLSTM)的预测准确度和精度更高,验证了所提出的短期风电功率预测模型的有效性、适用性和泛化性。Aiming at the short-term forecasting of wind power affected by complex multiple factors such as meteorological information,chaotic characteristics of time series,and the long-term dependence of wind power time series,this paper proposes a short-term forecasting method of wind power based on phase space reconstruction and bidirectional long short-term memories(BiLSTM).According to the data set of the global energy forecasting competition,the phase space was reconstructed from the wind power series based on the embedding theorem to show its intrinsic chaotic characteristics.The parameters of the phase space reconstruction were determined by the C-C method.The selected meteorological forecast data including future wind speed and wind direction was normalized,the reconstructed wind power data was combined as the input of BiLSTM,and the power data before reconstruction as the output to train the forecasting model.The day ahead forecasting experiment was conducted on the wf1 dataset provided by the Global Energy Forecasting Competition 2012.The average root mean square deviation of the first 30 days of the test set was 0.1194,and the average root mean square deviation of the first 107 days of the test set was 0.1409.The results show compared with ANN,BiLSTM,RF and KNN,Re-BiLSTM has higher forecasting accuracy and precision,which verifies the effectiveness,applicability and generalization of the proposed short-term wind power forecasting model.

关 键 词：气象信息 混沌特性 短期预测 嵌入定理 双向长短期记忆 

分 类 号：TM614[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]