Real-Time Performance Prediction of a Physics-Based Pump System Simulation Using Bayesian Optimization

This study aims to develop a model designed to effectively respond to fluctuations in energy demand. It involves constructing a physics-based simulation capable of real-time performance prediction for a two-stage centrifugal pump system. The model is composed of Modelica and a Reduced-order Model(ROM). It predicts key performance indicators such as pump head and flow rate in real-time. Additionally, a multi-parameter calibration framework using Bayesian optimization was applied to reflect the actual performance of the pump. This framework calibrates the parameters of the Modelica model. As a result, the error rates for flow rate and head were reduced to 2.87% and 3.58%, respectively.