Probabilistic Modeling and Prediction of Dynamic Discharge Process in Multiphase Pumps

Abstract

To ensure the reliability of reciprocating multiphase pumps, it is necessary to predict the flow rate curve of the discharge process under different multiphase transportation conditions. Unfortunately, an accurate model describing the complicated characteristics is still not available. A modeling method of automatically selecting a probabilistic model is proposed for prediction of the discharge flow rate. A posterior probability index is proposed to evaluate the trained local Gaussian process regression (GPR) models. Additionally, to enhance the prediction reliability, the prediction variance-based index is explored to automatically choose a more suitable model from the selected local GPR and just-in-time GPR models for each new sample. Consequently, with limited samples, an efficient probabilistic modeling method is developed for online prediction of the discharge flow rate curve. The experimental results for a reciprocating multiphase pump validate its superiority.