The impeller serves as the core aerodynamic component of multi-blade centrifugal fans, playing a pivotal role in determining their overall energy efficiency. However, traditional two-dimensional planar deformation design methods exhibit significant limitations when addressing complex internal flows. To overcome the constraints of conventional blade designs with axial invariance or uniform deflection, this study proposes a three-dimensional parametric design method based on axial distortion laws for multi-objective optimization of axially distorted blades. Using non-uniform rational B-spline curves, non-uniform deformation control is achieved along the blade height to accommodate variations in the inlet and outlet angles of airflow at different blade heights. During the optimization process, the optimal Latin hypercube sampling method is employed to generate sample points, while the Northern Goshawk optimization radial basis function neural network model and the non-dominated sorting genetic algorithm II are utilized for multi-objective optimization. Three non-dominated solutions are selected from the Pareto optimal set, and their validity is verified. Compared to the original design, the optimized balanced solution achieves a 15.51% increase in flow rate and a 1.01% improvement in efficiency. Notably, on the Pareto frontier, the maximum increase in the flow rate reaches 19.89%, and the maximum improvement in efficiency is 1.43%. Furthermore, comparative analysis of the flow field characteristics before and after optimization reveals that axially distorted blades effectively redistribute blade load by adjusting the inlet and outlet angles at different blade heights. This approach suppresses flow separation on the suction surface, reduces airflow blockage in the volute tongue region, and weakens the secondary flow within the volute, thereby improving the overall aerodynamic performance.