Research On Time Series Deep Learning Prediction Method For Non-point Source Pollution In Watershed

With the extensive application of the deep Learning technology in the field of time series prediction,more and more scientific researchers utilize deep learning method in the environmental field.in which,the water environment problem of non-point source pollution is becoming increasingly prominent,turning the characteristic and trend analysis and prediction of water quality into a key research content in order to better protect the quality of the water environment.it is of certain difficulties to realize mechanistic models based on physical processes owing to multitudinous parameters and the complicated process in determining parameters.In the wake of the development of computer technology,deep learning has gradually become the mainstream algorithm of time series prediction,which can effectively overcome the problems of traditional prediction methods.Nevertheless,there are also key practical problems existing in deep learning models.For example,in the case of data extremum,the model cannot predict well,or even produce seriously deviant results.In addition,in the water quality prediction of non-point source pollution,spatial factors such as land vegetation should also be taken into consideration.Therefore,how to solve the above problems of deep learning model construction for non-point source pollution is studied in this thesis,generating the following research results:(1)In the first place,in order to address the effects of spatial factors such as land use,vegetation on non-point source pollution,this thesis introduces the VGG method and constructs a spatial feature extraction model based on deep learning.The VGG model is used to extract the eigenvalues of the spatial image of the study area,since the dimension of the spatial feature output is too high,the principal component analysis method is used to reduce the dimension of the high-dimensional feature,and the Euclidean distance is calculated to obtain the optimal feature extraction effect,the dimensional features are converted into multi-dimensional time series data,which is used as feature expression vectors of the image data of the subsequent deep learning model.(2)Second,this thesis studies the migration and diffusion process of non-point source pollutants,calculates the analog value by virtue of the physical model,constructs the simulation-observation(SOD)module to calculate the time series error of target pollutants,combined with hydrological and meteorological data,other pollutant data,and spatial feature time series data,the input data set of the deep learning model is formed,which solves the problem of extremum prediction accuracy of deep learning data.(3)Finally,this thesis constructs a time-series deep learning model for non-point source pollution,forms the composite SOD-VGG-LSTM model combining VGG,SOD and LSTM models and verifies the validity of the model.Experimental results indicated that the proposed model had the highest accuracy in the extreme value prediction compared with the mechanism model,the maximum relative error between the predicted value and observed value was 7.82%.The comparison with the auto regressive integrate moving average model(ARIMA),support vector regression(SVR),and recurrent neural network(RNN)show superiority of our proposed model,the RMSE,MAE,and SMAPE indexes of the established model were 0.261,0.225,and 1.41%,respectively.In conclusion,through the analysis on water quality history time-series data and spatial information,this study constructs a composite model for water quality prediction integrated the deep learning method.After experiments,it is validated that the method proposed in the thesis generate more correct extremum prediction results and reflect the influence of spatial characteristics on the water quality changes caused by non-point source pollution when compared with other prediction techniques,thus providing new research ideas and methods for the future water quality change prediction of non-point source pollution as well as reliable technical guidance for water environment quality protection.