Construction of collaborative system for predicting and optimizing the performance of centrifugal pumps based on machine learning, and experimental verification

In the context of high-efficiency energy utilization, the research of centrifugal pump performance based on machine learning mainly follows two technical routes: the first involves establishing a performance parameter prediction model using neural networks, while the second focuses on parameter optimization of core components such as impellers through intelligent optimization algorithms. Usually considered as separate processes, this research integrates the CFturbo optimization concept to propose a multi algorithm based framework for centrifugal pump performance prediction and optimization system. The performance is predicted and optimized using a regression model constructed from two datasets. Considering the complex geometric configurations of the blades, it optimizes ten parameters of the impeller, including the inlet and outlet blade angles at both the hub and shroud, at the same time. The concept of High-Efficiency Operating Zone (HEOZ) was concurrently introduced to mitigate instability risks during off-design condition operations. Across both validation sets, the head and efficiency achieved mean absolute percentage errors below 4% and 2% respectively, while the model demonstrated exceptional performance in identifying HEOZ. Experimental validation further confirmed that the synergistic framework achieves superior accuracy and reliability, energy consumption can be saved by approximately 9.09%.