Interpretable Scientific\nMachine Learning Approach\nfor Correcting Phenomenological Models: Methodology Validation on\nan ESP Prototype

The electric submersible pump (ESP) is widely used in\noil extraction\nprocesses and is recognized for its effectiveness as an artificial\nlift technique in the petroleum sector. Developing and improving predictive\nmodels for these systems can contribute to optimizing operational\nefficiency and maximizing oil production. These improvements facilitate\nESP performance control and monitoring, potentially making the extraction\nprocess more economically and environmentally sustainable. This paper\naims to develop a framework for creating an interpretable corrective\nmodel for the ESP phenomenological model using experimental data.\nNeural networks were employed to analyze the complexities and nonlinearities\nof the processes, followed by symbolic regression to generate a simplified\nand interpretable equation to improve the model’s predictive\ncapacity. An uncertainty assessment of model parameters was performed\nusing Markov chain Monte Carlo (MCMC) and propagated to the model\nprediction. Additionally, a regression model was created directly\nfrom the process data for comparison purposes. The comparative analysis\nindicated that the approach incorporating neural networks to generate\nsynthetic data, followed by symbolic regression, improved the model’s\nability to predict key variables such as intake pressure, choke pressure,\nand production flow.