Research On Deformation Prediction Of Deep Foundation Pit Based On Improved Least Square Support Vector Ensemble Model

With the development of the city and the increase of population,the demand of high-rise buildings and underground space is also increasing,so the construction of deep foundation pit is becoming more and more common.These deep foundation pits are usually located in dense urban centers,often close to buildings,traffic arteries,subway tunnels and various pipelines.Therefore,it is an important means to avoid accidents to take real-time and effective monitoring,analyze the safety performance of deep foundation pit and surrounding buildings and make timely forecast.The deformation prediction of deep foundation pit is to predict the deformation of deep foundation pit by various mathematical modeling and simulation methods on the basis of monitoring,so that appropriate measures can be taken to protect the surrounding environment and buildings.The deformation prediction of deep foundation pit can be realized by finite element analysis,computer simulation,neural network and other methods,which can predict the deformation in the excavation process of foundation pit and provide important reference for the risk assessment and control in the construction process.On the basis of least square support vector machine,by introducing particle swarm optimization algorithm,genetic algorithm,adaptive noise complete set empirical mode decomposition and influencing factors,combined with a deep foundation pit project in Jinan,different models are compared and analyzed.The main research contents of this paper are as follows:(1)Briefly describe the deformation mechanism in the construction process of deep foundation pit,summarize the types of deep foundation pit deformation,classify and quantify the factors affecting the stability of deep foundation pit,and establish the index system of influencing factors of deep foundation pit deformation.Combined with the example of a deep foundation pit excavation in Jinan,the deformation monitoring data in the construction process of deep foundation pit were collected and sorted out,the gross error of the data was eliminated by the criterion,and the non-equidistant sequence was transformed into equidistant sequence by cubic spline interpolation.Finally,the accurate cumulative variation curve of each monitoring point was drawn,and the deformation law of deep foundation pit was summarized and analyzed.(2)The development history of machine learning is reviewed,the basic principle of least square support vector machine is studied,and the mean relative error,mean square error and square mean root error are determined as the evaluation indexes of model prediction accuracy in the experiment;The advantages and disadvantages of particle swarm optimization algorithm and genetic algorithm are discussed,and an improved particle swarm genetic hybrid algorithm based on the two advantages is proposed.Experiments show that the improved particle swarm genetic hybrid algorithm proposed in this paper has greatly improved the global optimization ability and operational stability compared with particle swarm optimization algorithm and genetic algorithm.(3)Two improved least squares support vector ensemble models are proposed for deep foundation pit deformation prediction.In the first model,the complete set empirical mode decomposition technology of adaptive noise is applied to the data processing stage of the prediction model.The decomposed modal components are trained and predicted by the least squares support vector machine optimized by the improved particle swarm genetic hybrid algorithm.Then the final prediction result is obtained by superimposing the prediction results;In the second model,the quantitative data of the influencing factors of deep foundation pit deformation screened by the information entropy method were used as input data and imported into the least square support vector machine optimized by the improved particle swarm genetic hybrid algorithm for training and prediction.Experimental results show that the prediction accuracy of these two models is superior to other prediction models in terms of mean relative error,mean square error and mean square root error,which proves that the two models have applicability and superiority to deep foundation pit deformation prediction.(4)The shortcomings,improvement directions and unfinished research of the two improved least square support vector combination models proposed in this paper are analyzed,and the future development of deep foundation pit deformation prediction is prospected.