In order to improve the torque performance of the permanent magnet synchronous motor (PMSM), a novel PMSM with non-uniformly partitioned Halbach array (NU-HA) is proposed in this paper. The orthogonal experiments are carried out based on FEA to analyze the structural parameters sensitivities to the average torque (Ave-T) and torque ripple ratio (TRR) of the proposed PMSM with NU-HA. Based on abundant FEA parametric simulations, the Kriging surrogate model is adopted to establish the mathematical model of the PMSM with NU-HA, which may be used for the further optimization. It is found that the maximum relative error is less than 3%, and the determination coefficient is basically close to 1, verifying the accuracy of the mathematical model. In order to improve the Ave- T and reduce the TRR, the structural parameters of the novel PMSM are optimized by using the non-dominated sorting genetic algorithm II (NSGA II). Through multi-objective optimization, the TRR is significantly reduced from 34.2% to 2.05%, and the Ave- T is improved by 2.02%.