Abstract: In response to the relatively limited research on health evaluation and prediction of the entire system of mine hoists, a health evaluation index system and comment set for mine hoists have been established. The health evaluation and prediction system for mine hoists has been designed. A fuzzy comprehensive evaluation method for the health of mine hoists is proposed to address the issues of insufficient utilization of monitoring data from various components of mine hoists, and the inability of health evaluation results to meet actual production needs. The method introduces relative degradation degree to characterize the health of different types of indicators of the hoist. The method uses health degree to quantify the health of mine hoists. The fuzzy comprehensive evaluation method is used to calculate the health of mine hoists. The analytic hierarchy process (AHP) is improved by replacing the 1-9 scale with an exponential scale to reduce computational complexity. The method uses CRITIC objective weighting method and combines subjective and objective weights to calculate the comprehensive weights of each subsystem and indicator. Based on the fuzzy comprehensive evaluation calculation process and the maximum membership principle, the health evaluation results and fault causes of the mine hoist are obtained. On the basis of the health evaluation results of the mine hoist, the Harris hawks (HHO) algorithm is used to optimize the important parameters of the support vector regression (SVR) model. The HHO-SVR model is constructed to predict the health of the mine hoist, improving the accuracy of the health prediction results. The experimental results show that the fuzzy comprehensive evaluation method can accurately evaluate the health of the hoist. Compared with particle swarm optimization support vector regression (PSO-SVR), genetic algorithm optimization support vector regression (GA-SVR), and grey wolf algorithm optimization support vector regression (GWO-SVR) models, the prediction results of the HHO-SVR model are closer to the actual values and have better prediction performance.