Abstract: Herein, a safety evaluation method of mountainous tunnel structures is established using an extreme learning machine (ELM) improved by multi-criteria compromise solution sorting method (VIKOR) and sparrow search algorithm (SSA). Previous researches on the influencing factors of tunnel structure safety in mountainous areas are analyzed to build an evaluation set, indicator system, and benchmarks for the safety of tunnel structures in mountainous areas.  Then, the analytic hierarchy process, entropy weight method + criteria importance through intercriteria correlation method, and game theory are employed to calculate the weights of evaluation indicators. The VIKOR decision model is adopted to quantify the safety levels of tunnel structures, and the constructed samples generated by MATLAB are transformed into training samples for machine learning. Based on the parameter optimization results, an SSA-ELM model is constructed, and 48 engineering instance samples, which had undergone field surveys and safety level determinations, are collected for simulation prediction. Comparative analysis is conducted with the original ELM and the particle warm optimization algorithm-optimized ELM model. The results indicate that the SSA-ELM model has higher prediction accuracy.

Key words: mountainous tunnel, tunnel structure safety evaluation, game theory, multi-criteria compromise solution sorting method, sparrow search algorithm, extreme learning machine