Research On Engineering Measures And Cost Of Landslide Treatment Based On Machine Learning

With the accumulation of landslide treatment experience and landslide treatment data,the modernization and technicalization of landslide treatment have gradually become the direction of development.If similar typical landslide treatment engineering cases are obtained as a reference in the process of landslide treatment,there can be examples to follow in the decision-making of treatment measures and the investment of treatment cost,and the efficiency and correctness of decision-making can be greatly improved.Therefore,based on the data of landslide treatment projects in Sichuan Province since 2010,the thesis carries out research on landslide treatment.The landslide factors are deeply analyzed,and the qualitative recommendation model of landslide treatment measures is built based on machine learning algorithm.The regression algorithm and recommendation algorithm are used to study the parameters of treatment measures quantitatively.The thesis carry out the research on landslide treatment cost,and build the recommendation model of landslide treatment cost by studying the influencing factors of landslide treatment cost.The main work and related achievements of the thesis are as follows.1.The landslide data are collected and processed,and the map data with landslide spatial information are introduced to describe the landslide in detail.The data of 609 landslide treatment projects from 2010 to 2021 are collected and sorted out,rich potential influencing factors of landslide treatment are extracted,and the features of landslide maps are extracted by machine learning methods such as classification and clustering.The results show that the description of landslide becomes more precise after integrating the features of landslide maps.In the content-based recommendation model,the recommendation result error MAE is reduced by 5 % after integrating map features,which shows that the introduction of landslide map data has certain value.2.Based on the recommendation algorithm,the qualitative recommendation of landslide treatment measures is studied.After factor analysis,the landslide information model is constructed.Through the construction,training and experiment of two qualitative recommendation models,the application effects of content-based recommendation algorithm and hybrid recommendation algorithm in qualitative recommendation of landslide treatment are compared.The results show that the MAE of the qualitative recommendation model based on content recommendation algorithm and hybrid recommendation algorithm are 17.2 % and 16.2 % respectively.The recommendation effect of the latter is better than the former.3.The parameters of anti-sliding pile in landslide treatment measures are studied quantitatively by using regression algorithm and recommendation algorithm.Through factor analysis,the factors with great correlation with the parameters of anti-sliding pile are selected.The regression model is trained and tested by three regression algorithms:Random forest,Ada Boost and XGBoost.After ten-fold cross verification,the XGBoost regression model with the best regression effect is selected,and the mean absolute error(MAE)is 0.25 m(spacing of class A anti-sliding pile),2.15 m(length of class A antisliding pile)and 4.20(number of class A anti-sliding pile).The same data and the hybrid recommendation algorithm are used for comparative experiments,the mean absolute error(MAE)is obtained: 0.33 m(spacing of class A anti-sliding pile),2.18 m(length of class A anti-sliding pile)and 4.48(number of class A anti-sliding pile).In terms of accuracy,the regression algorithm is better than the recommendation algorithm,but the recommendation algorithm has great explanatory advantages.4.The recommendation model of landslide treatment cost is built.Based on the recommendation algorithm,the recommendation model of landslide treatment cost is built.By recommending similar treated typical landslide cases for the landslides to be treated,the reference comparison of total treatment cost is provided.In the ten-fold cross validation of 400 landslides to be treated,the mean absolute percentage error(MAPE)of the landslide treatment cost recommendation model is 26.5 %.At the same time,the model is also verified by some examples.The error percentage of the recommended results in the validation result is less than 50 %,indicating that the cost recommendation has a certain reference value.