Abstract: The mechanical parameters of surrounding rock are one of the critical indicators in stability evaluation. However, existing methods that analyze all strata within a model often result in overestimated parameter values. To conduct a more refined study on the zonation characteristics of surrounding rock mechanical parameters, a novel approach combining surrounding rock zonation methods with parameter inversion models has been proposed, introducing a machine learning model for the inversion of mechanical parameters considering zonation degradation. This model employs the Coronavirus Herd Immunity Optimization (CHIO) algorithm to optimize the penalty factor and kernel function width of the Least Squares Support Vector Machine (LSSVM), significantly enhancing the precision and stability of parameter inversion. The effectiveness of the proposed method has been validated through theoretical solutions and engineering applications. Utilizing the Zhangji Mine in the Huainan mining area as a case study, five different hybrid machine learning models were compared regarding their prediction accuracy and generalization capabilities for surrounding rock mechanical parameters. The results demonstrate that the CHIO-LSSVM method achieves higher accuracy in parameter prediction. Finally, by integrating field-measured deformation data, parameter inversion analysis considering surrounding rock zonation degradation was conducted, and the inversion accuracy was validated through forward calculation results, indicating that this model is suitable for the refined inversion of zonation parameters in deep tunnel surrounding rock.