SVM的信息粒化时序回归预测城市用水量

Prediction of urban water consumption by SVM information granulation within time sequence regression

利用基于信息粒化的支持向量机预测模型对某市11个月的时用水量数据进行模拟训练,对下一个月的每日最高时用水量进行预测。首先提取每日的最高时用水量,再将每7个数据变换为一个三角型模糊粒子,该模糊粒子中的三个参数Low、R和Up分别代表一周内最高时用水量变化的最小值、平均值和最大值,然后利用SVM对最高时用水量及Low、R和Up进行预测。针对SVM在预测时调整自身相关参数困难的问题,提出了运用网格法对模型中的参数进行优化选择。实例分析结果表明,该模型建模速度快,预测精度高,且实用性强。

The data of hourly water consumption of 11 months in one city were simulated and trained by support vector machine(SVM) prediction model based on information granulation,meanwhile the peak hourly water consumption each day in the next month was predicted.First the data of peak hourly water consumption each day were picked up,then each seven data were changed into one triangular fuzzy granule,the three parameters Low,R,Up in the fuzzy granule respectively represented the minimum,average and maximum values of the variations of the peak hourly water consumption in a week,and the peak hourly water consumption,Low,R and Up were predicted using SVM.Aiming at the issues of SVM in difficult regulation of itself relative parameters,parameters of the model were optimized and selected by the grid method.The result of the case analysis showed that the model had the features of quick establishment,high accuracy of prediction and practicality.