Combined prediction of urban hourly water consumption using LSSVM based on multidimension embedding phase space

CHEN Lilin
(College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China)

Abstract: In order to improve the accuracy of chaotic hour consumption prediction in urban water supply system, a combined forecasting model for urban hourly water consumption using  least squares support vector machine(LSSVM)  was investigated. After the analysis of different effects on chaotic system forecast accuracy by various forecast methods and parameters of phase space reconstruction，a combined forecasting model for urban hourly water consumption using LSSVM based on multidimension embedding phase space was established. The different embedding dimensions were estimated by combining mutual information method and GP algorithm. Combined forecasting models were solved by LSSVM which can take advantage of all information in all dimension embeddings and forecast methods. The predictive bias under the single model was merged. In this way, the forecast accuracy was improved. The simulation results of hourly water consumption forecast in aplace shows that the forecast error is blow 2% and better than other forecasting results in single model. This proves the effectiveness and practicability of the approach.
Key words: urban hourly water consumption prediction; multidimension embedding; phase space reconstruction; least squares support vector machine(LSSVM)