摘要
为了解决供水系统优化调度的管网模拟问题,提出了一种基于改进支持向量机(SVM)的宏观关系模型.利用支持向量机推求管网节点水头和供水泵站供水量、供水压力的非线性关系.考虑了样本数量的不平衡性,引入权重因子.在结构风险最小化准则的目标函数中适当加大峰值和谷值样本误差的权重.还考虑了模型参数选择问题,建立了能应用于实际预测的参数选择方法.经杭州市实例计算表明改进的支持向量机算法提高了测点在峰值压力和谷值压力时的拟合精度,模型95%以上测压点数据的相对误差都在±5%以内,4个测点的平均绝对百分比误差分别达到0.96%,0.99%,1.86%和2.06%.
For solving the problem of pipe network simulation of water distribution system optimization and control, a kind of macroscopic state model based on modified support vector machine (SVM) was established. The nonlinear relation between node pressure head of the pipe network and water capacity and pressure of water supply pump station was deduced by SVM. After weight factor was introduced by analyzing the imbalance of training samples, the weight to sample errors of peak and valley values was properly increased in the objective function of structural risk minimization. Considering the problem of model parameter selection, the method for parameter selection was constituted for practical forecasting. Application results in Hangzhou City show that the modified SVM algorithm improves the simulation accuracy of node pressure at both peak and valley points, the relative error of 95% simulating node pressures is smaller than ±5%, and the mean absolute percent errors (MAPE) of 4 simulating node pressures are 0.95%, 0.99%, 1.86% and 2.06 respectively.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2005年第6期858-862,共5页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(50078048).
关键词
支持向量机
供水管网
宏观状态模型
Algorithms
Electric power distribution
Optimization
Water distribution systems
