克隆选择粒子群优化BP神经网络电力需求预测  

Power Demand Forecasting Based on BP Neural Network Optimized by Clonal Selection Particle Swarm

作　　者：李翔[1] 崔吉峰[1] 熊军 杨淑霞[1] 杨尚东[3] 

机构地区：[1]华北电力大学商业管理学院,北京102206 [2]重庆三峡水利电力(集团)股份有限公司,重庆404000 [3]国网北京经济技术研究院,北京宣武区100761

出　　处：《湖南大学学报（自然科学版）》2008年第6期41-45,52,共6页Journal of Hunan University:Natural Sciences

基　　金：国家自然科学基金资助项目(70501010)

摘　　要：在普通BP算法基础上,引入克隆选择粒子群算法,建立电力需求预测模型.将当期国内生产总值、前期国内生产总值、人口、当期产业结构变化、前期产业结构变化等影响电力需求的因素作为网络输入,电力需求作为网络输出,同时选择合适的隐层节点数,确定模型的网络结构.利用克隆选择粒子群算法反复优化BP网络的权值组合,将优化后的权值作为BP神经网络的初始值,进行BP算法,直至网络达到训练指标.利用近几年相关输入输出变量年度数据,对建立的模型进行电力需求实证预测分析,并同普通BP神经网络预测结果进行对比.结果表明:基于克隆选择粒子群优化的BP神经网络不仅训练速度快,而且误差小,预测精度明显提高,说明该模型对于电力需求预测的有效性.Based on the ordinary BP algorithm, we first established a power demand forecasting model after the introduction of clonal selection particle swarm algorithm. Then, we identified the model＇s network structure by using the power demand＇s influential factors like the current GDP, the previous period GDP, population, the current changes of industrial structure, and the previous period changes of industrial structure as the input of the network. We used the power demand as the output of the network, and meanwhile chose the suitable number of hidden nodes. We repeated the optimization of the BP network＇s weight combination with the aid of a clonal selection particle swarm algorithm, and then adopted the weight optimized as the inital value ot the BP neural network. We carried on the BP algorithm until the network met the training requirement. Finally, we used the recent years＇ annual data of relevant input and output variables to empirically forecast the power demand with the established model, and then compared the forecasting result with the ordinary BP neural networks. The comparison has shown that BP neural network based on clonal selection particle swarm has both fast training speed and small number of errors. The forecast precision has also been significantly improved, thus proving the validity of this model for forecasting power demand.

关 键 词：BP神经网络 克隆选择算法 粒子群优化 电力需求 

分 类 号：TM714[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]