Study On Urban Water Demand Forecasting Based On Hybrid Intelligent Algorithm

Being a vital part of urban infrastructure,water supply systems play a positive role in ensuring city's normal operation,promoting the social and economic development and guaranteeing the health of citizens.With the rapid development of China's economy and the improvement of people's living standards,per capita water use shows upward trend so urban water demand increases significantly;on the other hand,the improvement of urbanization level and the continuous perfecting of urban function contribute to higher complexity of water supply system.Water demand forecasting can provide data basis for scientific scheduling of water supply system and also raise utilization efficiency of water resource.Urban daily water demand forecasting plays a guiding role in water supply system scheduling to improve production efficiency and reduce production cost of water supply enterprises.Annual water demand forecasting,being the fundamental of water supply,water use and water conservation planning,guides the urban short term and long term water supply planning and construction so it plays a decisive role in reducing the total investment of water supply facilities.Therefore,the article focuses on daily water demand forecasting and annual water demand forecasting and adopts the observed water consumption data to verify the proposed forecasting models.Two kinds of urban daily water demand forecasting model are prosed in the article:?In view of shortages of BP neutral network in daily urban water supplies forecasting,forecasting model based on multi-resolution BP neural network is proposed.Complex characteristics of daily urban water supplies time series are transformed into single characteristics under different scales by discrete wavelet transform.Single characteristics are forecasted separately by BP neural network.Daily water supplies time series is analyzed by phase space reconstruction for its chaos features.Reconstructed time series is input variables of network.Application example indicates that compared with single BP neural network,multi-resolution BP neural network can reflect details and changing characteristics of time series better and has higher forecasting precision with mean absolute percentage error of 1.481%.It can be used in daily urban water supplies forecasting.?In order to improve forecasting precision of urban daily water consumption,a forecasting model based on support vector machine optimized by shuffled frog leaping algorithm is proposed.First daily water consumption time series is analyzed by phase space reconstruction and input of the model is determined.On this basis,employ shuffled frog leaping algorithm to optimize key parameters of support vector machine.Finally,use support vector machine after optimization to forecast.Application example indicates that compared with support vector machines optimized by genetic algorithm and particle swarm optimization algorithm,support vector machine optimized by shuffled frog leaping algorithm has higher forecasting precision,and it can be applied in urban daily water consumption forecasting.Generally speaking,urban annual water demand has less data sample,while grey system theory specializes in systems with little sample and incomplete information.So the article studied urban annual water demand forecasting based on GM(1,1)in grey system theory:?According to the characteristics of GM(1,1)model,this paper sums up the modeling steps of annual water demand forecasting based on GM(1,1)model and takes the water consumption of Chongqing city in 1997~2011 as a sample to illustrate how to set up a forecasting model.?As for water consumption system without impact disturbance,in order to improve forecasting accuracy the GM(1,1)model in annual water demand forecasting,the paper proposed mathematical model of the initial value and the background value optimization of GM(1,1)model and used SFLA to solve the optimization problem.And SFLA-GM(1,1)urban water demand forecasting model is constructed.An example is used to verify the prediction effect.?As for water consumption system with impact disturbance,buffer operator with variable weight is applied to data series with impact disturbance and SFLA is used to quantitatively optimize variable weight ?,so disturbance factors take effect before mutation point.Then optimized GM(1,1)forecasting model is set up.The example shows that SFLA-GM(1,1)based on buffer operator can greatly improve the prediction accuracy.