Axial flow pump has been widely used in hydraulic engineering, agriculture engineering, water supply and sewerage works, and shipbuilding industry. In order to improve the hydraulic performance of pump under off-design working conditions, the influence of the inlet segment axial chord and inlet angle adjustment of the guide vane on the pump segment efficiency and flow filed was simulated by using the renormalization group (RNG) turbulent model based on the Reynolds-averaged Navier–Stokes equations. The results indicate that the inlet segment axial chord and inlet angle adjustment of guide vane have a strong influence on the pump segment efficiency. Considering the support function and hydraulic loss of the guide vane, the inlet segment axial chord is set to 0.25 times the axial chord of guide vane. On the basis of the inlet angle of the guide vane under design conditions, when the inlet segment angle is turned counterclockwise, the pump segment efficiency is improved in the lower flow rate region; moreover, the pump segment efficiency is improved in the larger flow rate region when the inlet segment angle is turned clockwise. As the conditions deviate from the design working conditions, the influence of the guide vane inlet angle on the pump segment efficiency increases. If the inlet segment angle is properly adjusted under off-design working conditions, the flow pattern in the guide vane is improved and the hydraulic loss is decreased, because the inlet segment angle matches with the flow direction of impeller outlet; consequently, the pump segment efficiency is increased.

The axial-flow pump system has been widely applied to coastal drainage pump stations, but the hydraulic performance optimization based on the contraction angles of the inlet passage has not been studied. This paper combined the computational fluid dynamics (CFD) method, machine learning (ML) algorithms and genetic algorithm (GA) to find the optimal contraction angles of the inlet passage. The 125 sets of comprehensive objective function were obtained by the CFD method. Three contraction angles and comprehensive objective function values were regressed by three ML algorithms. After hyperparameter optimization, the Gaussian process regression (GPR) model had the highest R² = 0.958 in the test set and had the strongest generalization ability among the three models. The impact degree of the three contraction angles on the objective function of the GPR model was investigated by the Sobol sensitivity analysis method; the results indicated that the order of impact degree from high to low was ${\theta }_3>{\theta }_2>{\theta }_1$. The optimal objective function values of the GPR model and corresponding contraction angles were searched through GA; the maximum objective function value was 0.963 and corresponding contraction angles were ${\theta }_1=13.34°$, ${\theta }_2=28.36°$ and ${\theta }_3=3.64°$, respectively. The results of this study can provide reference for the optimization of inlet passages in coastal drainage pump systems.