Application Of Neural Network In Groundwater Level Prediction

In recent years,our comprehensive strength is always in the front rank and the construction of our country has also rapidly accelerated.However,it is also accompanied with the shortage of many kinds of resources,such as the source of life—water.At present,the relationship between water supply and demand is very tense.As one of the urban water-supply sources,the utilization of surface water is limited because it is easy to be polluted by external conditions.But the exploitation quantity of groundwater has been increasing quickly year by year because of its good quality,convenient extraction and low cost.Unfortunately,the excessive exploitation of groundwater resources has brought many hidden dangers to cities such as the land subsidence,the descending funnel and the like.As we know that the groundwater level can be used as an important factor to estimate the total groundwater resources.Therefore,it is particularly important for the sustainable development of the city to establish an appropriate neural network model to predict the groundwater level and then estimate the total groundwater resources.Only in this way can the groundwater resources be supervised and exploited more economically and reasonably.In this dissertation,the daily groundwater level of shallow wells and three influencing factors,temperature,rainfall and humidity from January to November 2017 of Yangzhou are used as experimental parameters and they are all normalized before.And then it chooses BP neural network model,wavelet neural network model and NARX neural network model respectively to train,verify and test by using the software-MATLAB R2012 a.To begin with,it establishes a three-layer BP neural network model.The range of the number of hidden neurons is decided according to the relevant formulas.And then the errors are calculated and compared one by one to choose the optimal network structure.After that we predict the groundwater level by the trained model.What's more,it also establishes a three-layer wavelet neural network model with the Morlet mother wavelet basis function as the active function of the hidden layer.During the time of training the network,the relevant factors and the connection weights are corrected according to the error.After determining the best model,input the test data for prediction.Finally,it establishes a NARX neural network model.After many trials and comparisons,the optimal parameters are selected.Thus,use the model to get the predicted value of the groundwater level.Comparing and analyzing the predicted results and the actual values of the above experiments,we can find that the predicted results of NARX neural network model are obviously better than the other two models in all aspects.It provides a decision basis for the relevant water conservancy departments to scientifically supervise and control of the groundwater resources in the future.