Rod pump optimization using adaptive machine learning

Abstract

Approaches for controlling operations of a sucker rod pump system of a production well that produces reservoir fluids from a reservoir, which employs a gateway device disposed at a wellsite corresponding to the well and a cloud-computing system operably coupled to the gateway device. The gateway device may communicate time-series operational data characterizing operation of the sucker rod pump system to the cloud-computing system. The cloud-computing system can receive and store the time-series operational data communicated from the gateway device and repeatedly perform operations that i) process the time-series operational data to train a machine learning model that predicts run time of the sucker rod pump system given feature data extracted the time-series operational data and ii) use the resultant machine learning model to determine new operating parameter data for use in controlling the sucker rod pump system, wherein the new operating parameter data maximizes production from the well.