As the flow structure in the centrifugal pump is complicated, it always causes serious noises, which has become an important problem in environmental protections. Some works have been completed for optimizing and reducing radiated noises of centrifugal pumps, but optimized algorithms are traditional and easily fall into local extreme values, so that final results are not always the optimal. For overcoming disadvantages of traditional algorithms, this paper proposed a novel MLGA-PSO (Multi-layer Genetic Algorithm-Particle Swarm Optimization) algorithm to make an optimization for noises and hydraulic performance of centrifugal pumps. This algorithm starts from the organizational structure of individuals and separates the global search from the local search, which can not only accelerate the convergence speed, but also avoid reducing the global search ability. The algorithm could effectively overcome the contradiction between global search ability and convergence speed. Inlet diameter, impeller blade outlet width, blade outlet angle and amount of blades are as designed variables. In order to verify advantages of the proposed MLGA-PSO algorithm in global search ability, optimizing speed and stability, GA (Genetic Algorithm), PSO (Particle Swarm Optimization) and GA-PSO (Genetic Algorithm-Particle Swarm Optimization) algorithms are chosen to carry out the compared experiment. Results show that the proposed MLGA-PSO algorithm has higher efficiency and accuracy. Finally, total noises of the optimized noise in the near-field and far-field using MLGA-PSO algorithm are 181 dB and 74 dB, respectively. Total noises in the near-field are reduced by 4.7 %, while those in the far-field are reduced by 16.9 %. It is clearly that the optimized centrifugal pump presents an obvious noise reduction effect.