City water demand forecast is an important step in water-supply monitor system. It's rather meaningful to discuss black-box forecast model's application in the complex water-supply system.This paper has focused upon the"balck-box"forecast modeling, briefly introduces water demand forecast research's present situation, and emphasizes on models such as Neural Networks, Support Vector Machine, and Time Series, both in algorithm principle and design scheme. Upon these bases, discusses the design methods of urban water demand forecast. In the modeling, we combine with the background of Shanghai Water Bureau, analyze historical data, and build the water supply system's forecast model (short-term, long-term). We also programmed the forecast software in MATLAB & LabVIEW, followed up by the conclusion and prospect of this paper.In the paper, we propose a new method"quantitative statistic of periodic water-demand", which improves the forecast accuracy. Meanwhile, using rolling-data-window method, build short-term forecast models which based on BP networks, RBF networks, and SVM algorithm respectively. The results show, their combined model can enhance forecast accuracy further. Besides, in order to help Water Bureau's construction, a long-term forecast model based on ARIMA time series analysis model has been discussed.The results show, off-line forecast models are accord with expected requirements. It's hopeful to be deployed in Shanghai Water Bureau's on-line monitor system.
|