API centrifugal pump in refinery plant is used to transfer for crude oil, semi-finished products and end products. Because pump is required for each process section or unit, many pumps are supplied and installed in refinery plant. However, pump design condition are different depending on the individual process conditions, fluids, and purposes, and such design conditions have generally changed depending on the economics and the concept of the process design. Therefore, accurate and rapid real-time design is required to cope with changes in design conditions. Also, for this purpose, it is necessary to comprehensively collect and review the big data to be handled in the design process, and to predict the proper process design specifications desired by the user. In this study, we selected variables and levels that affect the API centrifugal pump design (pump efficiency, pump shaft power) by using artificial neural network(ANN) suitable for the prediction model development. After that we have developed the most reliable model by changing the combination of variables using the MATLAB Toolbox with code modification.
Through the analysis of the data plane obtained from the developed pump specification prediction model, the process conditions (flow rate, differential pressure) in which the optimum specification performance is realized are shown as a result. In this study, it was confirmed that the optimal pump process prediction model can be predicted by the pump model predictive model without direct comparison.