基于季节分类和RBF自适应权重的并行组合电价预测  

Seasonal classification and RBF adaptive weight based parallel combined method for day-ahead electricity price forecasting

作　　者：林琳 刘譞[1] 康慧玲 Lin Lin;Liu Xuan;Kang Huiling(Baoding Electric Power Vocational and Technical College,State Grid Jibei Electric Power Company Limited Skills Training Center,Baoding 071051,China)

机构地区：[1]保定电力职业技术学院国网冀北电力有限公司技能培训中心,保定071051

出　　处：《电子测量技术》2020年第12期101-105,共5页Electronic Measurement Technology

摘　　要：电价预测在世界能源市场建设中具有重要意义,基于季节性分类,提出了一种由自回归移动平均模型(ARIMA)、多层前馈神经网络(BP神经网络)和支持向量回归模型(SVR)组成的并行组合电价预测方法。为了充分利用不同方法的优势,将ARIMA、BP、SVR分别应用于日前电价预测中,通过径向基神经网络(RBF)对4个不同季节的3个预测值进行非线性拟合,得到最终的预测结果。所提方法的创新点在于对于每个季节都有特定的预测模型,不同预测方法之间非线性权重值随时间变化而变化,与传统的回归组合预测方法和季节非分类情况相比,其仿真结果表明所提方法具有更好的适应性和更高的预测精度。Electricity price forecasting is of great significance especially in the construction of energy market all over the world.This paper proposed a parallel combined forecasting approach consisting of ARIMA,BP neural network and SVR on the basis of seasonal classification for day-ahead electricity price forecasting(DAEPF).ARIMA,BP and SVR were applied in DAEPF respectively and the final result was obtained by RBF neural network to integrate three predicted values in four different seasons by nonlinear fitting way in order to take the advantages of different methods.The main novelty of presented method is the specific model to every season and the nonlinear weights between various forecasting methods could vary along with time so as to follow the development of society.The simulation based on the data from PJM market showed better generalization ability and higher forecasting accuracy of proposed model than traditional regressive parallel combined method and non-classification situations.

关 键 词：日前电价预测 季节分类 自适应权重 并行组合法 RBF拟合 

分 类 号：TM93[电气工程—电力电子与电力传动]