基于RBF神经网络的时间序列交叉供热负荷预报研究  

Study of Heat Load Forecasting Based on RBF Neural Network and Time Series Crossover

作　　者：陈烈[1] 张永明[1] 齐维贵[1] 邓盛川[1] 李娟[1] 

机构地区：[1]哈尔滨工业大学电气工程及自动化学院,黑龙江哈尔滨150001

出　　处：《电子学报》2009年第11期2444-2447,共4页Acta Electronica Sinica

基　　金：国家"十一五"重点科技攻关项目(No.2006BAJ03A05);哈尔滨市科技创新人才研究专项资金(No.Rc2006XK007001)

摘　　要：针对供热过程的特点及节能控制的需要,提出基于RBF神经网络的时间序列交叉供热负荷预报法.首先对现场实测的供热负荷数据进行预处理,取得建立预报模型所需的负荷样本阵列;随后,应用自相关法求取RBF神经网络的输入维数,并分别建立时间序列的横向及纵向预报模型;最后用最小二乘法求出横向与纵向负荷预报的交叉权系数,得到RBF神经网络的时间序列交叉预报模型.仿真结果表明,RBF神经网络交叉负荷预报的精度高于横向负荷预报及纵向负荷预报,其实时性要优于BP神经网络交叉负荷预报.According to the characteristics of heat supply and the demands of energy-saving control,heat load forecasting based on RBF neural network and time series crossover is proposed. Firstly,field measured data are pretreated to generate the load series which is used to found forecasting model. Then autocorrelation method is applied to determine the dimensions of the input vectors of the RBF neural network. Meanwhile, the horizontal and vertical forecasting models based on RBF neural network are established respectively. Finally, the crossover weight coefficients of the horizontal and vertical forecasting models are calculated by using the least-squares method. And the time series crossover forecasting model is obtained. Through corrtparing the simulation results, the accuracy of crossover forecasting is superior to horizontal and vertical forecasting, and the real-time ability of RBF neural network crossover forecasting is also better than BP neural network crossover forecasting.

关 键 词：供热过程 负荷预报 RBF神经网络 时间序列交叉 

分 类 号：TM921.2[电气工程—电力电子与电力传动]