Abstract: In order to judge the stability of tunnel surrounding rock more accurately and effectively, a kind of random weight particle swarm optimization(RandWPSO)-least squares support vector machine (LSSVM) prediction model for limit displacement is introduced. The membership function of fuzzy probability model for surrounding rock stability is optimized and then a modified model is established. The fuzzy probability of surrounding rock stability can be calculated by inputting parameters, i. e. the predictive displacement value u, the standard deviation of predictive displacement, the predictive limit displacement value U and the standard deviation of predictive limit displacement, into the modified model. The model is verified by 8 engineering practices. The results indicate that: (1) The problem of irrational limit displacement selection existing in the traditional membership function is solved and the error caused by the simplifying linearization treatment is evaded by the modified model effectively. (2) The calculated stability probability is in good agreement with the actual situation and the reliability of surrounding rock stability evaluation results is higher. (3) The improved model is applied to the evaluation of tunnel surrounding rock stability and the calculated results can reflect the actual engineering situation well.

Key words: tunnel engineering, fuzzy probability, membership function, surrounding rock stability, RandWPSOLSSVM prediction model