Deformation Prediction Research Of Foundation Excavation Based On IPSO-RNN-LSTM

With the continuous improvement of China’s urban modernization level,the scale and depth of foundation pit buildings of subway and high-rise buildings show an overall upward trend.The deeper the excavation depth of the foundation pit is,the more complex the construction technology is and the more safety risks are.Moreover,there are roads,urban pipelines and buildings distribute around the foundation pit.The existence of these structures makes the soil around the foundation pit under the combined action of a variety of loads,which brings safety risks to the foundation pit construction.Therefore,the timely and accurate prediction of foundation pit deformation is very important to the safety of foundation pit construction.Based on the actual monitoring data obtained from a foundation pit project in Shaoxing,this paper analyzes the actual monitoring data of foundation pit and obtains the cause and law of foundation pit deformation,and then analyzes the characteristics of Recurrent Neural Network(RNN),Long-Short-Term Memory Neural Network(LSTM)and foundation pit data.A new foundation pit deformation prediction model RNN-LSTM is proposed.Finally,in order to solve the problem that it is difficult to determine the parameters of RNN-LSTM,an improved particle swarm optimization algorithm is proposed for the super-parameter selection of RNN-LSTM.The main research contents and conclusions are as follows:(1)Introduce to the foundation pit monitoring scheme,analyze the actual monitoring data of supporting pile deformation,column vertical displacement and surface settlement of the foundation pit,and explain the reasons for the changes of these monitoring data.(2)The principle of foundation pit deformation prediction model which is widely used at present is expounded,and the RNN-LSTM foundation pit deformation prediction model is put forward according to the characteristics of foundation pit monitoring data.In order to verify the effectiveness of RNN-LSTM network in nonlinear data of foundation pit,the comparative experiments of Support Vector Machine Regression Model(SVR),Moving Autoregression Prediction Model(ARIMA),Grey Theory Model(GM(1,1)),Back Propagation Neural Network(BP),RNN and LSTM are carried out.The results show that LSTM network is more suitable for foundation pit deformation prediction than other models.Then use RNN-LSTM and LSTM to predict and compare,the results show that RNN-LSTM is more suitable for foundation pit deformation prediction than other models.(3)In order to solve the problem of parameters determination of RNN-LSTM,an improved particle swarm optimization algorithm is proposed to optimize its parameters.Firstly,the shortcomings of particle swarm optimization(PSO algorithm)and genetic algorithm(GA algorithm)are described,and a particle swarm optimization algorithm with self-mutation mechanism(IPSO algorithm)is proposed.Secondly,three test functions are used to test the optimization ability of particle swarm optimization algorithm,genetic algorithm and improved particle swarm optimization algorithm.the final test results show that the improved particle swarm optimization algorithm is less precocious and has stronger optimization ability than traditional particle swarm optimization and genetic algorithm.After that,the foundation pit prediction experiments of RNN-LSTM,PSO-RNN-LSTM and IPSO-RNN-LSTM are designed,and the results show that the foundation pit deformation prediction ability of IPSO-RNN-LSTM is better than that of RNN-LSTM and PSO-RNN-LSTM.