The invention aims at a turbo expander impeller in large air separation equipment and particularly relates to an optimal design method of a large turbo expander impeller blade structure with defect consideration. On the basis of the original impeller stress analysis, a defect factor is added, impeller parameters are subject to genetic optimization operation by a generalized regression neural network and a genetic algorithm-based multi-goal optimization algorithm, and Pareto optimal solutions which are uniformly distributed are finally obtained to serve as impeller blade optimization parameters; the actual working features of an impeller are integrated in an optimization process, so that the impeller is high in overall strength and structural rigidity and high in practicability. Under the condition that the requirement on the strength of the impeller is met, by the method, stress concentration under defect action is reduced, the service life of the impeller is prolonged, meanwhile, the weight and the rotation inertia of the impeller are reduced, and materials are saved.