基于数据驱动的差分RBF神经网络的台区短期负荷预测方法  

Short-term Load Forecasting Method Based on Data-driven Difference RBF Neural Network

作　　者：段炼 乡立 吴琼 陆颢文 李茜莹 孙毅[2] DUAN Lian;XIANG Li;WU Qiong;LU Haowen;LI Qianying;SUN Yi(Guangzhou Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Guangzhou,Guangdong 510630,China;School of Electrical and Electronic Engineering,North China Electric Power University,Beijing,102206,China)

机构地区：[1]广东电网有限责任公司广州供电局,广东广州510630 [2]华北电力大学电气与电子工程学院,北京102206

出　　处：《广东电力》2020年第11期66-74,共9页Guangdong Electric Power

基　　金：广东电网有限责任公司广州供电局科技项目(GZHKJXM20180012)

摘　　要：为应对短期负荷预测中台区负荷数据流的动态性、不完备性和非线性等特征对负荷预测带来的挑战,针对性地构建了基于数据驱动的差分径向基函数(radial basis function,RBF)神经网络的台区短期负荷预测方法。首先采用基于集成分类器的数据流分类算法对多维台区负荷数据进行分类;其次采用灰色关联度分析确定指标参数关联度排序;借鉴增量编码的应用方式构建了负荷预测编码器和解码器,进而对负荷数据进行差分RBF预测分析,并进行了交叉验证。仿真结果表明,基于数据驱动的差分RBF神经网络的台区短期负荷预测方法可以在保持较低误差的同时,降低运算时间,压缩误差区间。In order to deal with the challenges brought by the dynamic,incomplete and nonlinear characteristics of the load data flow in the substation area in short-term load forecasting,this paper constructed a short-term load forecasting method based on the data-driven difference radial basis function(RBF)neural network.Firstly,it uses the data flow classification algorithm based on the integrated classifier to classify the multi-dimensional load data and grey relation analysis(GRA)to determine the rank of index parameter correlation degree.By using the application mode of incremental code,it constructs the encoder and decoder for load forecasting and makes difference RBF prediction of load data as well as carries out cross validation.The simulation results show that the short-term load forecasting method based on the data-driven differential RBF neural network can reduce operation time and compress the error range at the same of maintaining low errors.

关 键 词：负荷预测 短期 差分 日负荷 径向基函数神经网络 

分 类 号：TM715.1[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]