Application of deep belief network algorithm in safety evaluation of highway slope engineering
Aiming at the problem that the traditional slope engineering safety evaluation method cannot carry out the engineering safety evaluation of slope stability in line with the actual situation,In this paper,a safety evaluation method for highway slope engineering based on Deep Belief Networks(DBN)algorithm is proposed.In this paper,first,using MATLAB established a DBN neural network model.Then,using the collected middle and outer slope stability dataset,the bulk density,cohesion,internal friction angle,slope angle,slope height,and pore pressure ratio are input to the DBN algorithm as the characteristic values,and testing the DBN model,and the test results are compared with the slope stability safety evaluation method based on BP neural network algorithm.Finally,to test the versatility of the proposed DBN algorithm and its application in practical engineering,this paper uses the slope engineering safety influencing factors collected in Puyan project and different from the benchmark data set to carry out the safety evaluation of highway slope engineering.The results show that:(1)the accuracy of the DBN algorithm in predicting the slope state of the collected dataset is 100％;(2)When analyzing the example of arc failure slope,The average absolute error of the DBN network prediction safety factor arithmetic is 0.09,and the average relative error of the arithmetic is 0.07;(3)when analyzing the wedge failure slope example,the average absolute error of the DBN network prediction safety factor arithmetic is 0.01 and the average relative error of the arithmetic is 0.01;(4)the accuracy of the predicted slope dataset collected by the DBN algorithm is higher than that of the BP neural network algorithm;(5)when applied to the Puyan project,the accuracy of the DBN algorithm in predicting the slope risk level is 99.19％;(6)When applied to the Puyan project,the safety evaluation of the actual project by the DBN algorithm is closer to the reality than the index method.
万方数据
维普
