Hybrid model for discharge flow rate prediction of reciprocating multiphase pumps. (October 2018)

• Record Type: Journal Article
• Authors: Deng, Hongying; Liu, Yi; Li, Ping; Zhang, Shengchang
• Is Part Of: Advances in engineering software. Volume 124(2018)
• Page Start: 53
• Page End: 65
• Publication Date: 2018-10
• Subjects: Reciprocating multiphase pump -- Discharge flow rate -- Hybrid model -- Probabilistic information -- Gaussian process regression; CFD computational fluid dynamics -- GPR Gaussian process regression -- MACV maximal absolute change of variance -- ME maximal error -- MFRE maximal flow rate error -- MV maximal MACV -- RMSE root-mean-square error; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2018.08.006
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: A hybrid model is structured for the discharge flow rate prediction of reciprocating multiphase pumps. Two probabilistic indices are proposed for stage recognition of the discharge flow rate curve. Individual modeling for each stage can better handle complex process using process and probabilistic knowledge. The proposed method is validated for better prediction performance and more efficient implementation. Abstract: Accurate description of the quick and nonlinear change of the discharge flow rate in reciprocating multiphase pumps is important but difficult. To ensure the reliability of reciprocating multiphase pumps, it is necessary to model the relationship between the discharge flow rates of a stroke and multiphase transportation conditions. A hybrid modeling method is proposed for practical use in this work. First, a Gaussian process regression (GPR) model is adopted to online predict the discharge flow rates. Then, the probabilistic information of GPR is used to divide the flow rate curve of a stroke into four stages for individual modeling. Additionally, the process knowledge of multiphase pumps is integrated into the modeling process. Furthermore, to capture the nonlinear characteristics of the mutation stage with limited samples, the local relationship between the input variables change related to opening points and the flow rates is constructed. Consequently, the process knowledge and probabilistic information are integrated to formulate a practical hybrid model. Experimental results show the superiority of the hybrid modeling method.