Abstract: For the process of grinding T-type Q235 carbon steel with alumina flap disc, with downforce and grinding tool speed as input variables and material removal rate as output variable, the backpropagation (BP) neural network was optimized on the basis of the success history intelligent optimizer (SHIO) algorithm. The initial weights and thresholds were determined and the best parameters within a certain range were obtained. The SHIO-BP neural network prediction model of material removal rate was established and verified by orthogonal test. The results show that the average absolute error between the material removal rate predicted by SHIO-BP neural network and the test value was 0.929, and the R2 score was 0.990 3. The prediction was accurate, and the traditional BP neural network problems of parameter selection and easily falling into local optimal solution were solved.