Research And Application Of The Prediction Model Of Tailings Dam Deformation Based On RF-SSA-LSTM

The prediction of tailings dam deformation is one of the important elements of safety management and maintenance in mining enterprises.Accurate prediction and timely maintenance measures are essential to prevent catastrophic dam failure accidents.Tailings dams are a complex system affected by many factors and dynamic changes.The interaction of various factors leads to severe deformation of the tailings dam,which in turn leads to dam failure accidents.However,the existing prediction methods only consider the influence of single factors on the tailings dam deformation,but do not take into account the synergistic effect of multiple factors,resulting in low accuracy of tailings dam deformation prediction,and the application of the actual project results in general.It is important to establish a comprehensive deformation prediction index system from a system perspective and to determine a prediction model with high applicability and accuracy when predicting tailings dam deformation.Therefore,it is of great theoretical and practical importance to make accurate prediction of tailings dam deformation.To establish an effective tailings dam deformation prediction model,this paper first analyzes the tailings dam deformation characteristics,analyzes the mechanism of different influencing factors,and explores the main system internal factors and environmental factors affecting tailings dam deformation.From the system perspective,a tailings dam deformation prediction index system integrating environmental influencing factors is constructed.Secondly,by analyzing the advantages and shortcomings of traditional models,we propose a tailings dam deformation prediction model based on RF-SSA-LSTM by integrating random forest(RF)feature screening algorithm,sparrow search optimization algorithm(SSA)and long short-term memory(LSTM)neural network prediction method.Finally,the quadratic algorithm and the missing forest algorithm are used to process the cleaning input data,and the RMSE,MAE and R~2are used as evaluation indexes to compare the prediction accuracy of the RF-SSA-LSTM prediction model with numerous prediction models by using the deformation monitoring data and meteorological variation data of the Lazigou tailings dam project to train the model.The results show that RF-SSA-LSTM has RMSE of 0.0129,MAE of 0.008,and R~2of0.980,which is better than SSA-LSTM and LSTM prediction models,and the predicted values fit better with the true value curve.The LSTM model using the SSA optimization algorithm gives better prediction results than the single deep learning model LSTM model,and is also more suitable for prediction of tailings dam deformation data than the four models LR,Xgboost,MLP,and Bayesian.The RF-SSA-LSTM prediction model uses random forest to optimize the input feature dimension of the model and reduce the deep learning model construction The SSA algorithm optimizes the different parameters of the LSTM,which in turn improves the prediction accuracy and robustness of the model.In this paper,the RF-SSA-LSTM based prediction model is established to fully explore the relationship between different time series data under the joint action of complex influencing factors of tailings dams,learn the long-term dependence of time series data and achieve good prediction effect,which is important for mastering the health condition of tailings dams and supporting the safe and stable operation of tailings dams.