Research On Relevance Vector Machine Model For Prediction And Recognition Of Non-linear Behavior Of Rock And Soil

Geotechnical engineering is an important part of civil engineering.In recent years,with the development of economy and the aggravation of urbanization,infrastructure construction has begun to develop underground.Large-scale geotechnical engineering problems are reaching unprecedented heights.However,with the increase of the scale and complexity of the project,the uncertainties of underground engineering is more and more various.Predicting the non-linear behavior of rock and soil accurately has great significance for protecting people's lives and property and reducing engineering cost.Traditional theoretical solutions to geotechnical engineering problem prediction are often limited by various aspects,such as the difficulty of engineering investigation,long time and high cost.With the progress of science and technology,more and more intelligent methods have been applied in geotechnical engineering.Support Vector Machine and BP Neural Network have been applied widely,but due to the shortcomings of the method itself,forecasting results are often unsatisfactory,so more efficient and accurate forecasting methods need to be put forward.A new idea of applying relevance vector machine to geotechnical engineering is proposed.As a new machine learning method,relevance vector machine is seldom used in geotechnical engineering at present.Compared with other machine learning methods,relevance vector machine has the advantages of strong generalization ability,simple parameter optimization,high prediction accuracy and good robustness.At the same time,the model is more sparse and the choice of kernel function is more free.It can establish accurate mapping relationship between complex factors and predicted values,which provides an effective way to solve geotechnical engineering problems.In sum,the main works and results are listed as follows:1.Aiming at the limitations of traditional methods,the relevance vector machine model for regression of non-linear behavior of rock and soil mass is established.The hot issues in geotechnical engineering,such as debris flow velocity,slope stability are selected to verify the applicability of the model.A prediction model of average velocity of debris flow based on relevance vector machine is established,and the advantages of the model are compared with the results of other intelligent methods.At the same time,a non-linear mapping relationship between slope stability index and its influencing factors is established based on the principle of relevance vector machine.The stability of the model is further verified by mathematical analysis of average relative error,mean square error,confidence interval and correlation coefficient.2.In view of the fact that the relevance vector machine model can only be divided into two classifications at first,but there are many kinds of classifications needed in practical engineering,a one-to-one relevance vector machine model is proposed to realize multi-classification,and the probability output results are given.The mapping relationship between rock burst grade and its influencing factors is established through training of a small number of learning samples by relevance vector machine and the probability output of new samples with only known influencing factors is given to a certain rock burst grade.The relevance vector machine recognition model of the failure mode of layered dangerous rock is established and compared with other intelligent methods.The results prove that the relevance vector machine can effectively identify the non-linear behavior of rock and soil,and the results are satisfactory.3.Relevance vector machine model predicts fast,but trains slowly,in order to solve the problem,the advantages of principal component analysis are fully utilized to reduce the dimension of input factors,remove redundant information between data,obtain independent input samples,reduce the impact of related data on the accuracy of the model,and improve the efficiency of the model.In prediction,the regression model of TBM tunneling speed is established,and the advantages of the improved model are verified by comparing with the original model.In recognition,the classification model of surrounding rock stability is established.The results show that the optimized correlation vector machine model is more accurate and efficient.