基于k-means聚类与径向基神经(RBF)网络的电力系统日负荷预测

Daily Load Forecasting of Industrial Enterprise Power System Based on k-means Clustering and RBF Neural Network

对大型工业企业电力负荷震荡剧烈、幅度大等问题,提出了借助卡尔曼滤波对"失真数据"进行实时估计,弱化陈旧数据权值的k-means聚类和径向基网络的组合预测模型。依据最小逼近误差,利用最小二乘法对RBF网络的输出权值进行优化。文中方法对电力系统不同的工况背景,使用k-means聚类算法来确定隐藏节点的数据中心和扩展常数,改善了常规RBF网络随机选择网络中心的情况,同时使电力负荷精度得以有效提高。最后,依据文中的组合模型和其他4种常用算法分别对宝钢电力系统负荷数据进行分析预测,仿真结果表明:方法对负荷预测效果较好。

For high frequency and issues of large industrial enterprises power load vibration,presented the advanced combined the forecasting model based on k-means clustering and RBF Neural network with Kalman filter for ＂ false data＂ real time estimates,weaken the weight of stale data,preprocess,minimum based on approximation errors,using the least-squares optimization on RBF network output weights.Method on different working conditions of power system in the background of the text,using the k-means algorithm to determine the hidden node＇s data centre and the expand constant,improved General RBF network to randomize the center of the network,improved the load forecasting precision,using the model for Baosteel data analysis of electric power system load forecasting,simulation results indicated that the effectiveness of the model.