Water Consumption Prediction by Using SVM and Information Granular Time Series Regression

The accuracy of water consumption prediction has a direct effect on the practicality of optimal operation and control in a water distribution network system. In this paper, Support Vector Machine (SVM) was applied to conduct simulation training for hourly water consumption data (over a period of 11 months) which had been processed by fuzzy information granulation method, and water consumption for the following month was predicted. First, the maximum hourly water consumption was extracted from the previous 11 months' water consumption. Then every seven data (a week) were transformed to a triangle fuzzy particle. Three parameters of the triangle fuzzy particle, Low, R, and Up, were defined as the minimum change (Low), the average change (R) and the maximum change (Up) of the maximum water consumption during one week. Finally, SVM was used to predict these 3 parameters and the maximum hourly water consumption. In order to solve the SVM, problems occurred while adjusting related parameters, and optimization selection via genetic algorithm was proposed. The results in actual application indicated that this model had faster speed, higher precision and practicality.