Water Demand Forecasting Method Based On Elman Neural Network

With the further development of industrialization and urbanization,the demand for water resources will continue to grow over a long period of time.In addition,the impact of global climate change will lead to more sharp contradiction between supply and demand of water resources,and the rational use of water resources is an important task.The rational utilization of urban water resources is mainly determined by the urban water supply dispatching system,and the accuracy of water demand prediction directly affects the rationality of investment,layout and operation of water supply system.Therefore,we must scientifically and reasonably predict the needs of urban water resources.The core problem of water demand prediction is the technical problem of prediction,or the mathematical model of prediction.Traditional mathematical models are described by mathematical expressions.They have the advantages of small computation and fast speed,but there are also many defects and limitations,such as lack of self-learning,adaptive ability,robustness of prediction system and so on.However,the variation of water consumption data is nonlinear,time-varying and uncertain.The traditional prediction method can no longer meet the accuracy requirement of water demand prediction.The use of neural networks provides a new way of thinking.In this paper,the combined forecasting method based on Elman neural network is used to predict the water consumption on campus.The main contents include the following parts.In the data processing part,it is very important to purify and effectively pre process raw data for accurate prediction.At the same time,the data of water use is a random nonstationary sequence,which adds a certain degree of difficulty to the prediction.Therefore,this paper analyzes and processes the water consumption data to get a better set of data to meet the input requirements of the prediction algorithm.First,the missing and abnormal data are complemented and replaced.The average number substitution method is used for the missing value,and the horizontal and vertical processing methods are used to deal with the abnormal value.Then,because the data has greater volatility,not smooth,so,using EEMD method to decompose the data,get the data component is relatively stable,due to the excessive number of components,so the grey correlation analysis,will be associated with a higher degreeof component to merge,reconstruction sequence,the final sequence quantity is less and relatively stable as a predictive input.The prediction of water data is carried out.The water consumption prediction model based on Elman neural network is proposed in this paper.Elman neural network has faster convergence speed and can be dynamically modeled.It has certain advantages in forecasting large fluctuation of water data.However,the Elman neural network still has the disadvantage of poor global search capability.Therefore,aiming at this defect,we use genetic algorithm to optimize Elman neural network,and select the best weights and thresholds predicted by Elman neural network,so that the network performance is optimal.In the simulation,the Elman neural network optimized by genetic algorithm to find the global optimal weight threshold,and then the optimal solution in the Elman neural network as the initial parameters of Elman neural network training,the trained network using the rolling prediction method,update the training data samples,in order to to improve the prediction accuracy.Finally,the experimental data verify the effectiveness of the water demand prediction method proposed in this paper.