Application of TabPFN model on the energy performance improvement of high-power multistage centrifugal pump

With the continuous growth of global electricity demand, supercritical boiler feed pumps (multistage centrifugal pumps), which are critical components in thermal power systems, consume significant amounts of energy during operation. Therefore, reducing their energy dissipation is crucial. However, pump optimization often faces challenges such as high computational costs and low efficiency. This study proposes a novel baffle design installed within the second interstage passage of a supercritical boiler feed pump. Based on this design, six key control parameters of the third- and fourth-stage impeller blades were selected as variables, generating 200 sample cases. The TabPFN model was used to construct surrogate models for predicting head and entropy generation. This study innovatively combines TabPFN with SHAP to conduct an interpretability analysis for pump optimization. Subsequently, the trained TabPFN model was integrated with the NSGA-II algorithm for multi-objective optimization. Energy dissipation in the flow field was visualized using entropy generation theory. Findings reveal that after installing the baffle, the pump achieved a 1.2 % efficiency improvement while maintaining nearly constant head, with an 8.2 % reduction in turbulent entropy generation and a 2.9 % decrease in wall entropy generation. The optimized impeller blades exhibited an increased wrap angle, enhancing fluid flow control. Compared to the baseline design, the optimal solution total entropy generation dropped by 18.6 % and efficiency improved by 4.8 %, with negligible head variation. This study provides a valuable reference for TabPFN in fluid machinery and insights into energy-efficient design optimization for multistage centrifugal pumps.