Optimization of Centrifugal Pump Impeller Based on Neural Network and NSGA-II Algorithm

Abstract

Abstract This article focuses on the multistage centrifugal pump secondary impeller as the research object, and the two performance parameters of head and efficiency are taken as the optimization objectives. Firstly, the range analysis of significant influence degree of impeller geometric parameters was carried out by partial factor test method. The blade inlet angle at the front and rear cover boards of the impeller, blade outlet angle and blade wrap angle as the optimization variables, two optimization schemes with different blade numbers were proposed. On this basis, the optimal Latin hypercube sampling method was used to extract samples from the two schemes, CFD technology was used to calculate the corresponding target value of each sample, BP neural networks with different structures were used to train the samples, and the agent model with the highest prediction accuracy was optimized as the population evaluation function of NSGA-II algorithm to search for optimization. The results show that the pump performance parameters have been improved after optimization of Scheme II, and the flow of the internal flow field conforms to the law. Finally, through the experimental verification, the optimized centrifugal pump head and efficiency increased by 3.21m and 3.05%.