Prediction Of Subgrade Deformation Based On Neural Network Fusion Model And Pavement Additional Effect Analysis In Seasonal Frozen Area

The subgrade is the substructure and load-bearing foundation of the road,if the deformation is too large,it will lead to the instability of the structure,affect the driving safety and shorten the service life of the road.Therefore,to predict the deformation of roadbed scientifically and reasonably and control it in a reasonable range is the key to ensure the stability of road structure and operation safety.With the formation of the long-life road construction concept of "durable surface layer,long base and permanent subgrade",it is particularly important to carry out the research on the frost heave and thaw settlement of subgrade in the seasonal frozen soil area.However,the freeze-heaving and thawing of subgrade is affected by many factors such as geological conditions and climatic conditions.Therefore,it is of great practical significance to explore a method to predict the freeze-heave and thaw subsidence deformation of subgrade based on influencing factors.This paper studies the freeze-thaw characteristics of silty clay based on the engineering environmental data of the test section.BiLSTM-BP neural network models and BiLSTMAttention neural network models are constructed using the freeze-heave and melt subsidence observation data of the subgrade of Yusong high-speed Fuyu connection line,respectively,and the models are merged with the stacking algorithm.The freeze-heave and thaw subsidence deformation of subgrade is predicted,and the accuracy of the model prediction is analyzed.Based on the finite element theory,the mechanical response of pavement structure layer under the action of melt settlement is analyzed.Specific work contents are as follows:(1)The road damage forms caused by subgrade frost heave and thaw subsidence in seasonal frozen soil area are analyzed.The freeze-thaw properties of silty clay with low liquid limit are analyzed from four aspects: composition,microstructure,physical properties,thermal parameters and engineering mechanical properties.The results show that the silty clay with low liquid limit is prone to freeze-heave and thaw settlement,and the CBR value,shear strength and resilience modulus of the soil sample decrease after freeze-thaw.(2)Using the average relative error(MAPE)as an evaluation index,the commonly used prediction models are: GM(1,1)model,curve fitting method,BP neural network model and BiLSTM neural network model were compared and selected,BP model and BiLSTM model with high prediction accuracy were selected and improved,and a single model of BILSTMBP and BILSTM-attention neural network was constructed.The applicability of the two models in the prediction of subgrade settlement is analyzed,which provides a theoretical basis for the establishment of the fusion model.(3)The fusion algorithm of stacking models is proposed,which integrates BiLSTM-BP and BiLSTM-Attention neural network models.The mean square error(MSE),root mean square error(RMSE)and mean absolute error(MAE)of single model and fusion model are compared and analyzed.The results show that the fusion model algorithm has the advantages of two single models,and has higher accuracy in the prediction of frost heave and melt subsidence deformation of roadbed.(4)A subgrade and pavement structure model was constructed based on ABAQUS finite element software,and the mechanical response of asphalt pavement structural layers was analyzed under different melt settlement depth and radius conditions.The results showed that the stress distribution law of the upper layer,the lower layer and the bottom layer of the pavement were similar,but the stress types were different: At the action of vehicle load,the upper and lower lamination stress is the largest,and the bottom base tensile stress is the largest.The tensile stress of the upper and lower layers is the largest,and the compressive stress of the base layer is the largest.The vertical strain on the top of the subgrade reaches the peak value at the center of the melt plate,and the vertical strain reaches the maximum at the melt settling edge.At the same depth of melting,the stress and strain decrease with the increase of melting radius.Under the same melting radius,the stress and strain increase gradually with the increase of melting depth.The inner bottom base of the melting plate will produce fatigue damage first,and the upper and lower layers will produce fatigue damage first when the melting plate is outside.In a word,this paper deeply discusses the frost heave and thaw subsidence deformation of road subgrade in seasonal frozen areas,proposes a subgrade deformation prediction method based on multiple factors,and uses BiLSTM-BP neural network model and BiLSTMAttention neural network model to conduct model fusion prediction.The results show that the model fusion algorithm can effectively improve the prediction accuracy and has practical application value.Based on the finite element theory analysis,the mechanical response law of pavement structure under the action of subgrade melt settlement provides a theoretical basis for promoting the construction of long-life pavement in the seasonal freezing area.