Prediction Model Research About Daily Water Demand Based On Optimized Neural Network

At present, the water resources shortage problem is increasingly serious, to have the reasonable deployment of water resources in time and space structure is becoming more and more important.Urban water supply system is a very important part of the rational utilization of water resources,scientific prediction of urban daily water can provide powerful guarantee for urban water supply enterprises production and optimal scheduling.Urban water consumption is subject to the factors such as people’s lifestyle and climate,the relationship between the factors is complex and has strong nonlinear characteristics.Artificial neural network,as a kind of nonlinear parallel computing model, has strong adaptive and fault-tolerant ability and is able to get the potential relationship between input and output by learning.BP neural network is chosen as the reference model of daily water prediction in Urumqi,through adjusting and optimizing the network parameters, realize its high precision prediction of daily water.The pros and cons of the parameters decide whether the model can achieve the target prediction.According to the actual daily water situation of Hongyanchi water plant area,three-layer BP neural network is established.Mutual information theory is used to select effective predictor,Golden section method is used to expand its hidden layer interval to obtain the appropriate hidden layer units,The particle swarm into harmony search algorithm is used to adjust the connection weights of BP neural network.The optimized BP neural network model is applied to predict daily water supply region of Hongyanchi water plant in Urumqi,the results shows that it sped up the convergence of the network, and obtained the higher prediction precision,the mean absolute percentage error met the application requirements,which provide ideas and solutions for water resources scheduling in Urumqi on the space-time structure.