Introducing machine learning and hybrid algorithm for prediction and optimization of multistage centrifugal pump in an ORC system

The isentropic efficiency of the working fluid pump has a significant impact on the overall performance of the organic Rankine cycle (ORC) system. Based on machine learning, this paper proposes an experimental data-driven isentropic efficiency prediction model. Meanwhile, S-fold cross validation algorithm and smoothing factor circulation screening technology are used to improve the predictive ability of the model. The prediction accuracy of the optimized model and the unoptimized model are compared with each other. The influence of several operating parameters on isentropic efficiency are analyzed. In addition, with intelligent algorithm, the boundary values of operating parameters are determined. The genetic algorithm (GA) and particle swarm optimization (PSO) are combined into a GA-PSO hybrid algorithm. Subsequently, the hybrid algorithm is integrated with the machine learning model to predict and optimize the isentropic efficiency under full operating conditions. The highest isentropic efficiency reaches up to 58.73%. The prediction and optimization of the isentropic efficiency of multistage centrifugal pump under full operating conditions provides not only a useful guidance on assuming pump efficiencies in theoretical analysis, but also a meaningful reference for obtaining the optimum overall operating performance of the ORC system. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This study proposed an experimental data-driven isentropic efficiency prediction model based on machine learning, and used algorithms and technologies to enhance predictive ability. By analyzing the influence of operating parameters on isentropic efficiency and determining boundary values using intelligent algorithms, the prediction and optimization of isentropic efficiency of multistage centrifugal pumps was successfully achieved.