Study On Landfill Leakage Based On BP Neural Network

Contaminated sites not only pose a serious threat to human health,but may also cause environmental problems such as groundwater pollution.Therefore,it is particularly urgent to solve these problems.Because of its low permeability,geomembrane is resistant to testing and is not easy to crack,and has high cost performance,it is used in horizontal anti-seepage of landfills,forming an anti-seepage body and a waterproof curtain for leachate.The high density resistivity method has the advantages of wide application and high spatial and temporal resolution,and can be used for the detection and risk assessment of contaminated areas of landfills,providing the basis for remediation of sites.However,the high-density resistivity method has several common issues.For instance,it relies on the accuracy of the data for initial model selection,and it transforms complex nonlinear problems into linear ones,resulting in suboptimal fitting and getting trapped in local minima during inversion.With technological advancements,artificial neural network algorithms have also been continuously improved.This study focuses on the application of the Backpropagation(BP)neural network in three-dimensional inversion of contaminated sites.The BP neural network possesses unique learning and nonlinear approximation capabilities,making it highly promising in this regard.It can effectively address the challenges faced by traditional inversion methods and provide more detailed structures,thereby significantly enhancing the inversion results.The GRU(gated recurrent neural network)neural network has been widely used in time series forecasting in recent years,effectively reducing the risk of overfitting and dealing with gradient explosion problems.In this paper,GRU is used to predict the apparent resistivity time series data of the leakage dispersion model,and the simulation model and field examples are verified.The results show that GRU can well realize the prediction of apparent resistivity time series.By applying the predicted apparent resistivity data to the BP neural network,the diffusion process of pollutants can be effectively monitored,so as to better predict the occurrence,development and change trend of pollutants.The main work done in this thesis is as follows.(1)Study and learn the basic theory of resistivity method and the basic derivation of forward modeling problems,etc.,and then develop a resistivity device system for leakage and diffusion based on common electrode array devices.(2)Construct a model by the forward method,generate single-anomaly and doubleanomaly resistivity data samples,and implement the three-dimensional resistivity method inversion using BP neural networks.(3)Study the theoretical knowledge of GRU neural network,use GRU to predict the apparent resistivity time series of the pollution dispersion model,and put the apparent resistivity prediction results into the BP neural network for inversion to realise the monitoring of pollution dispersion.(4)By acquiring monitoring data from the monitoring system at the simulated landfill site of the Academy of Environmental Sciences,the BP and GRU neural networks were used to analyse the pollution dispersion.The inversion results were also verified by AGI software,confirming that the BP and GRU algorithm is effective for monitoring the pollution dispersion of hazardous waste landfills.The research of this thesis shows that using BP and GRU to monitor pollution diffusion has high accuracy,which provides some reference experience for future research.