In view of the flaw of significant performance fluctuations and low efficiency found in switched reluctance motor drives, a multi-objective optimization strategy, which combines a prediction model optimized using support vector machines (GA-SVM) with the third-generation non-dominated genetic algorithm (NSGA-III), has thus been proposed. The simulation results show that the proposed method can significantly improve the average torque and efficiency of switched reluctance motors, with its torque ripple reduced. By establishing a simulation model of a switched reluctance motor, and using sensitivity analysis to select parameters with high influence factors as decision variables, the switched reluctance motor is sampled using hyper-Latin square sampling. The response values are calculated using finite element analysis, with the GA-SVM and NSGA-III algorithms combined to perform multi-objective optimization on the motor. The optimized data is weighted with weight coefficients, thus obtaining the optimal solution. The effectiveness of the proposed method can be verified by the simulation results.