Collapse is one of the major disasters in the construction of mountain tunnels. Because there are many factors affecting the collapse, and the weight of each factor varies greatly, even some factors are unnecessary or redundant. At present, the commonly used evaluation methods not only fail to screen these unnecessary or redundant factors, but also rely too much on expert experience and subjective evaluation to determine the weight, resulting in low accuracy and poor reliability of risk assessment results. Based on this, considering the advantage of rough set theory in data mining, index screening and importance computation, the risk assessment of mountain tunnel collapse is constructed as the decision information table of rough set. However, in the engineering experiment, it is found that the attribute reduction and weight calculation based on the traditional dependency degree can not meet the requirements, and there is a problem that the weight of calculation is zero or the reduction result is too much and can not be taken out. In view of the above problems, a method based on conditional information entropy is proposed, which introduces conditional information entropy into the definition of attribute importance and weight. At the same time, the most important conditional attributes are taken as the starting point, and attributes are gradually added to realize attribute reduction. The method established in this paper can not only extract the main influencing factors from a large number of influencing factors, but also eliminate the relatively redundant or unimportant factors. At the same time, it can calculate the objective weight of each factor. Then combined with fuzzy comprehensive evaluation method, the risk evaluation model of mountain tunnel collapse based on rough set and conditional information entropy is established and successfully applied to the Xiucun tunnel. It shows that the model is reliable and practical, and provides a new research idea for the risk assessment of mountain tunnel collapse.
