Study On Progressive Classification Of Tunnel Surrounding Rock Based On Multi-Source Data Fusion

The multi-source data information of the tunnel surrounding rock is produced by diverse methods of data collection of surrounding rock in tunnel,which leads to the uncertainties in the classification of tunnel surrounding rock.The theories of multi-source data fusion are introduced to fully consider the characteristics of the multi-source data of the tunnel surrounding rock,to reduce such uncertainties and make the result of the classification more reliable.According to the gradualness of the tunnel construction,the multi-source data of the tunnel surrounding rock is produced progressively,it is necessary to carry out the classification in a progressive way.This paper has made systematic research on the progressive classification of tunnel surrounding rock based on multi-source data information and multiple methods of the classification of tunnel surrounding rock.According to the theory of information entropy,the quality analysis of the multi-source data based on indices of the rock classification is proposed.The basic indices of the rock classification(including RQD,GSI,BQ,Q,RMR₈₉,and RMR₁₄)are divided into three categories,as indices of discontinuities,indices of physical mechanics,and indices of environmental factors.The empirical formulas for data conversion between similar indices in different methods is proposed based on existing research and regression analysis.The theory of information entropy is applied to calculate the information gain of the multi-source data of the tunnel surrounding rock.The importance of characteristic of the indices and the index of data quality are further calculated based on the information gain,to provide support for the progressive classification.Based on the improved D-S evidence theory,the RMR₁₄ and BQ methods of the classification of tunnel surrounding rock are proposed.The indices in RMR₁₄ are divided into intervals based on rock classification,and the basic belief assignment（BPA）is obtained by calculating the generalized distance between the data and the intervals through the Euclidean distance formula;in the method of BQ,the Euclidean distance formula is applied straight to obtain the BPA by calculating the generalized distance between the multi-source data and each possible values of the indices.Considering the features of the indices of RMR₁₄ and BQ,BPA is further processed by the construction of RBM as the matrix of belief reinforcement based on the divergence between each BPA.Murphy’s average belief theory is used to determine the average belief of BPA with evidence weight.The iterative data fusion is carried out with the average belief and the probability results of the classification are output for rating the level of tunnel surrounding rock,which are consistent with the results of the classification based on RMR₁₄ and BQ.According to the multi-source data generated gradually with the construction of tunnel,the model of progressive classification of tunnel surrounding rock with full consideration of each index of the methods of rock classification is proposed.The model of progressive classification is constructed based on BP neural network with 27 features,and 110 groups of data samples to train the neural network.To simulate the multi-source data of tunnel surrounding rock produced gradually,60 groups of multi-source data of tunnel surrounding rock are generated to test the trained model.The test results are consistent with the results of each method of rock classification,proving the accuracy of the model.Based on all the research above,a program for progressive classification of tunnel surrounding rock is developed to provide standard process.The applied research on the progressive classification of tunnel surrounding rock based on data fusion is done in the ZK78+352～ZK78+383 section of the Grand Canyon tunnel of E’han Expressway in Sichuan Province.The progressive classification is carried out four times based on the data collection.The results show that the tunnel surrounding rock in the section is in level III with the probability of 0.77,0.662 and 0.626,except the surrounding rock in ZK78+374～ZK78+378 section being in level IV with the probability of 0.612.It is necessary to take reinforcement measures to improve the safety of the tunnel construction in the ZK78+374～ZK78+378 section.The results are consistent with the results obtained from each method of rock classification,which verifies the accuracy of the proposed model.