基于超参数双调和方程代理模型的航空燃油离心泵多目标优化

Addressing in the problem of poor accuracy in constructing surrogate models due to the highly nonlinear performance parameters in aviation fuel centrifugal pumps, a biharmonic equation surrogate model with hyper-parameter optimization was proposed in this paper. Through variance-based sensitivity analysis method, the effectiveness of the hyper-parameters was validated, and the mechanisms of actions was explored. Based on comparison of particle swarm optimization, Bayesian optimization was used to quickly optimize the hyper-parameters so that the surrogate model automatically improved prediction accuracy. In the case verification, multi-objective optimization of centrifugal pumps was achieved based on the hyper-parameters biharmonic equation surrogate models for head and efficiency. The research results show that the hyper-parameters can achieve the variable structure of biharmonic equation surrogate model, and the high-order interaction effect of hyper-parameters is the main reason for the accuracy of the surrogate model. Bayesian optimization can achieve automatic optimization of the hype-parameters biharmonic equation surrogate models, with an acceleration ratio of 17.62 compared to particle swarm optimization. After optimization, the model prediction accuracy can be greatly improved, with the adjusted R square (R_adj²) increasing 17.1%, from 0.82 to 0.96. The loss function values of hyper-parameter biharmonic equation surrogate model for the head and efficiency are extremely small, which can describe the mapping relationship between structure parameters and head, efficiency. The centrifugal pump multi-objective optimization ultimately led to an increase in head of 0.041 m and in efficiency of 0.76%. In addition, the high velocity flow clusters in the channel are suppressed and the blade load gradient is significantly improved after optimization.