Evaluation Of Technical Condition Of In-service Highway Slope Based On Bayesian Logistic Regression

In recent 30 years,with the rapid improvement of the expressway network in Guizhou Province,there are many high fill and deep excavation slopes.Up to now,with the increase of service time,a considerable part of the slope has experienced problems such as aging of the supporting structure and deterioration of service performance,which has brought a large number of economic losses and casualties.Based on the actual situation that the service performance of highway slopes in mountainous areas of Guizhou Province deteriorates year by year,it is urgent to study the hazard assessment and hazard prevention technology of highway slopes in service.With the development of computer,artificial intelligence algorithm and geographic information system have become important tools for the evaluation of slope technical condition.Therefore,combined with Bayesian inference,logistic regression model,geographic information system and other technical methods,this paper carried out the research on the technical status evaluation of highway slope engineering in service in Guizhou Province.The main research results are as follows:(1)The terrain and natural landform,stratigraphic structure and its lithology,geological structure and meteorological characteristics in Guizhou Province were investigated,and the internal and external factors affecting the stability of highway slope were studied.Taking prestressed anchor cable frame,which takes up the largest proportion of the supporting form of slope in service in Guizhou Province,as an example,the common diseases of the slope are studied and the evaluation index system is established.(2)A total of 171 sets of slope data were collected to analyze six factors affecting slope stability(weight,cohesion,internal friction Angle,slope height,slope Angle and pore pressure ratio),and the slope stability state was analyzed by using machine learning method of Bayesian logistic regression.The effects of data preprocessing methods(data standardization,normalization to [0,1],normalization to [-1,1])and three prior distributions(normal distribution,Cauchy distribution,Student T distribution)on the model optimization were studied.The results showed as follows: 1)the prediction results obtained after the normalization of the data were close to those obtained after the standardization in accuracy;2)With different prior distributions,there is little difference in model optimization results,but it is found that the mean value and standard deviation of each prior distribution will affect the posterior results of regression coefficient(intercept).3)When the data preprocessing method is normalized to [-1,1] and the prior distribution is normal,the prediction accuracy of the model is the highest,and the average AUC value is 0.860.(3)Taking the prestressed anchor cable frame supporting structure as an example,the stability state,the working condition of the prestressed anchor cable,the damage of the frame beam,the vegetation coverage rate and the daily maximum rainfall were selected as the evaluation indexes,and the advantages and disadvantages of the fuzzy comprehensive evaluation model and the Bayesian logistic regression model in the evaluation of the technical condition of the slope were compared and analyzed.The latter not only weakens the subjectivity of the former evaluation,but also achieves more than 91.2% classification accuracy in the verification set.(4)By using Python language,Java Script language and HTML5 technology,an intelligent cloud platform based on Web GIS is developed to evaluate the technical condition of the inservice slope in Guizhou Province,which provides a convenient visual operation platform for the management and evaluation of the in-service highway slope in Guizhou Province.Taking Renhuai-Zunyi Route in Guizhou Province as an example,The application of cloud platform in online domain slope is illustrated.