The Study And Application Of Kalman Filter And Hybrid Intelligent Algorithm In Daily Urban Water Consumption Prediction

Along with urban economy and population growth,the urban water demand in China increases rapidly.At the same time the water pollution become serious,and the conflict between water demand and water resources scarcity is more serious.Now the water distribution network in most of cities in China is scheduled by experience,and we can't make sure whether the network is scheduled under a good condition or not.So pipe burst and leakage occur frequently and plenty of valuable water is wasted.Scientific dispatch of water network can significantly improve the security,economy and reliability of water supply system.It will effectively reduce the urban water shortages.Predicting daily water demand accurately and quickly can provide an important basis for optimal dispatch of water distribution network,so it has theoretical and economic significance.In this paper,the research progress of urban daily water demand prediction was systematically summarized.Because whether the model parameters should be changed or not and how to change the parameters were seldom mentioned in traditional water demand prediction,Kalman filter and hybrid intelligent algorithm theory was introduced to solve the problem of dynamic parameters estimation of daily water demand prediction model.The main contents are as follows:1.After the analysis of the daily water consumption series was carried out,the conclusion that the high correlation between the daily water demand of the prediction day and the daily water demand 6 days before the prediction day was obtained.The daily water demand 6 days before the prediction day were used as model 1's inputs,and adaptive genetic algorithm(AGA)was introduced to optimize the parameters of least support squares vector machine(LSSVM)-based model to build the GA-LSSVM-based model 1.To prove the importance of main factors of daily water consumption for predicting the daily water demand,the main factors of daily water consumption and the daily water demand 6 days before the prediction day were used as model 2's inputs,and adaptive genetic algorithm(AGA)was introduced to optimize the parameters of least support squares vector machine(LSSVM)-based model to build the GA-LSSVM-based model 2.Case study shows that model 2 performed better than model 1,and the importance of main factors of daily water consumption for predicting the daily water demand was proved.2.Using GA,the parameters of historical daily water consumption model 1 before the prediction day were determined to obtain the series of parameters.With the series,Extended Kalman Filter(EKF)was applied to estimate the parameters of model 3.Using GA,the parameters of historical daily water consumption model 2 before the prediction day were determined to obtain the series of parameters.With the series,Extended Kalman Filter(EKF)was applied to estimate the parameters of model 4.Among four models,model 4 obtained the best performance.The validity of EKF and the importance of main factors of daily water consumption for predicting the daily water demand were proved.