In order to improve the rapid wear of abrasive belt and its influence on the surface of titanium alloy when grinding titanium alloy and other difficult-to-machine materials with abrasive belt in the whole life cycle. A variable parameter optimization method of grinding parameters based on slow feed abrasive belt grinding mode is proposed. Firstly, the pre-experiment of abrasive belt wear in the whole life cycle of titanium alloy processing is carried out, and the processing parameters, abrasive belt wear quality and surface roughness of the processing process are collected, so as to prepare for the training of the model. Secondly, SVM algorithm is used to build a roughness prediction model, and NSGA-Ⅱ algorithm is used to optimize the processing parameters in the whole life cycle of slow-feed abrasive belt grinding. Finally, by comparing and analyzing the abrasive belt wear characteristics and the characteristics of titanium alloy surface roughness, morphological characteristics, microscopic characteristics and surface oxidation under variable parameter and fixed parameter grinding methods, the variable parameter grinding method in the whole life cycle of abrasive belt is verified. The results show that the accuracy of roughness prediction based on SVM algorithm can reach above 0.95, and the mean absolute error (MAE) is as low as 0.064. By comparison, it can know that the prediction accuracy of the algorithm is higher at the end of abrasive belt wear, because the sampling frequency at the end of abrasive belt wear is relatively high. The processing parameters optimized by NSGA-Ⅱ algorithm can effectively improve the surface quality. The roughness of the whole life cycle before optimization gradually decreases from 2.049 μm to 0.184 μm, and the roughness after optimization decreases from 1.549 μm to 0.494 μm; Moreover, the surface morphology and oxidation degree were detected by SEM and EDS. During the whole abrasive belt wear process, the plastic flow of fixed parameter grinding method is more than that of variable parameter grinding method, and the oxidation reaction degree is also greater. In addition, using ultra-depth-of-field equipment to detect abrasive belts in different wear periods, the topography of abrasive belts is obtained, and it is found that the proposed variable parameter optimization method can effectively improve the abrasive belt wear and reduce the rapid abrasive belt wear caused by slow feed grinding. The SVM algorithm proposed in this paper can predict the roughness and the NSGA-Ⅱ algorithm can optimize the parameters of grinding line in abrasive belt. The optimal solution of machining can be found through this algorithm, and the optimal machining parameters can be obtained through calculation. Through the comparative experiment of fixed parameter grinding mode and variable parameter grinding mode, the grinding contrast experiment shows that the variable parameter abrasive belt grinding method proposed in this study can effectively improve the grinding surface quality (The roughness is relatively low, the plastic flow on the surface of titanium alloy is small, and the oxidation reaction on the surface can also be improved.), slow down the abrasive belt wear and prolong the service life of the abrasive belt compared with the fixed parameter grinding method.