Performance optimization and prediction model of pump as turbine based on Latin hypercube sampling

To explore the mechanism of how key geometric parameters affect the performance of pump as turbine, three correlation analysis methods—Pearson correlation coefficient analysis, Spearman correlation coefficient analysis, and standardized linear regression model analysis—are employed to analyze the correlation between key geometric parameters and performance parameters of 18 pump models, and 4 geometric parameters are identified as the optimization directions. Then, based on the Latin hypercube sampling method, 60 numerical calculation models are established. Flow calculations are performed on each turbine to obtain its external characteristics, and the influence law of the 4 geometric parameters on the external characteristics is studied in depth. Meanwhile, two regression models are further constructed to predict the efficiency of pump as turbine. The results show that the blade wrap angle has a strong negative correlation with efficiency, the impeller outlet width has a strong positive correlation with efficiency, and the impeller diameter has a weak positive correlation with efficiency. The fit goodness of the constructed full quadratic polynomial regression models ranges from 0.92 to 0.941. The adjusted model with variable elimination has better generalization ability, with an average relative deviation of only 1.52%.