Research On Landslide Dynamic Prediction And Control System Based On Elman Neural Network

Landslide is one of the extremely destructive geological disaster,which seriously endangers the security of human lives and properties,destroys the global ecological environment,and wastes resources.Therefore,it is of great significance to further study the landslide prediction and control system to reduce the influence of landslide disasters.The evolution of landslides is affected by many factors,which has strong nonlinearity and uncertainty.The artificial neural network can establish the black box model of the system without the complicated mechanism analysis.Taking Baishuihe and Shiliushubao landslide for the study object in the Three Gorges reservoir region,the neural network is introduced into the study of landslide prediction and control system in this thesis.Firstly,with an improved BP algorithm considered the characteristics of time series data in landslide,the landslide single-step prediction model is established by using Elman dynamic neural network,which reduces the time complexity of model training and achieves the single-step prediction of Baishuihe and Shiliushubao landslides with higher accuracy.Secondly,based on the single-step prediction of landslide,by introducing the PID neural network and the fuzzy PID control algorithm,the landslide control model based on improved PID neural network(F-PIDNNC)is presented and established.Then a single-step prevention and control system for landslide is designed,which can effectively reduce the slope deformation speed.Finally,considering the problem of delayed control existing in the single-step landslide prevention system,a two-step prediction model and an improved F-PIDNNC model for landslide are established.Then the predictive control algorithm is introduced to design a two-step prediction control system for landslide.The control system can predict the landslide displacement two steps ahead with higher accuracy and integrates the two-step predictions to achieve more effective landslide control.The warning range is doubled and the control is more rapid and effective,which buys valuable time for landslide control.This thesis mainly studied the landslide prediction and control,which are the two main issues in landslide research.The neural network is used to predict the landslide deformation.According to the prediction results,a scientific control scheme can be obtained through the control model.The two aspects form a complete landslide prevention system,which has a certain reference value for the further landslide prevention study.