The Study And Application Of Grey Theory And Hybrid Modeling Algorithm In Urban Daily Water Consumption Prediction

With the fast development of urbanization in China, the scale of water distribution network becomes larger and water distribution network(WDN) becomes more complex. The traditional dispatch strategies of WDN based on the pressure of monitoring node can not meet the requirement of safe and economic operation of WDN. Optimal dispatch strategies of WDN are needed to reduce the energy consumption of pump and improve the safety and reliability of water supply system. Urban daily water consumption prediction is an important premise and foundation for optimal dispatch scheduling of WDN. The accuracy of urban daily water consumption prediction will determine whether the scheduling scheme can be applied to real WDN or not.In this paper, the research progress of urban daily water consumption prediction at home and abroad is systematically summarized. The main work of the paper is as below:(1) As the missing value in daily water consumption series is generally determined by subjective judgment, a combined interpolation method based on grey theory was proposed. An autocorrelation method was used to analyze the daily water consumption series to determine the series related to the missing value. With the daily water consumption series before and after the missing value, grey theory was applied to obtain backward and forward interpolation estimation, respectively. Two values were optimally combined to achieve the estimation of missing value. Case study shows that the proposed method has a higher prediction performance than the average value method.(2) As the parameters of least squares support vector machine(LSSVM) determined by cross-validation is time-consuming, a city daily water consumption forecasting method based on particle swarm optimization(PSO) and LSSVM was proposed. The main influencing factors of daily water consumption in the prediction day and relevant seventh daily water consumption before the prediction day were used as the model input. The daily water consumption in the prediction day was used as the model output. PSO algorithm was introduced to optimize the parameters of LSSVM, and the individual fitness values in PSO algorithm were determined by cross-validation. The daily water consumption forecasting model was built. Case study shows that the proposed forecasting method based on PSO and LSSVM has higher computing speed and better estimating performance than traditional LSSVM-based method.