Online trained controller for Electrical Submersible Pumps in liquid–gas flow

Abstract

The Electrical Submersible Pump (ESP) is an artificial lift method widely used in the oil industry. More than 150,000 wells worldwide are equipped with ESPs to produce cost-effective oil rates. An ESP unit usually includes an electric motor, a protective element, a gas separator (optional), and a multistage centrifugal pump. When free gas produced in the well reaches the pump stages, problems such surging and gas-locking can occur, leading to unexpected shutdowns in the oil production. In this work, a control based on Artificial Neural Networks (ANN) with online training is proposed as a solution to reduce these unplanned shutdowns. The proposed controller was designed using the direct inverse control method and the data obtained from an ESP working with a liquid–gas flow. The results of this work show that the controller can drive the operation of the ESP away from the regions where surging and gas-locking would occur. Since the controller is an online trained ANN, it was able to control the operation of two different pumps, namely the model P100 pump, the model P47 pump, and the model P23 pump.