Pressure Prediction of a Water Distribution Network Based on SVM

With the development of SCADA systems, the monitored real-time data of a water supply network is more complete than before. Taking full advantage of the data to forecast the pressure of the monitoring points is the basis of water pipe network schedule optimization. A support vector machine (SVM) model for pressure prediction was established to solve the problems by making use of the data. Kernel function and parameters were chosen by means of cross validation. The EPANet example 1 and the water distribution system of City SZ were used to test the method. According to the EPANet example, the SVM method has a higher prediction accuracy and more stable results than those of the BP neural network. The example of SZ water supply network shows that RBF-SVM can be applied to predict nodal pressure in a network. The largest relative error of total results is 8.72%, and 85% predicted errors are smaller than 5%.