Rod pump optimization using adaptive machine learning

A system for controlling operations of a sucker rod pump system of a production well that produces reservoir fluids from a reservoir, the system comprising:

• a gateway device disposed at a wellsite corresponding to the well;
• and
• a cloud-computing system disposed remotely from the wellsite and operably coupled to the gateway device;
• wherein the gateway device comprises at least one first interface that is operably coupled to the sucker rod pump system and at least one second interface that is operably coupled to the cloud-computing system, wherein the gateway device is configured to communicate time-series operational data characterizing operation of the sucker rod pump system to the cloud-computing system;
• and
• wherein the cloud-computing system is configured to 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 until next shutdown, wherein the processing of the time series operational data involves:
• filtering the time-series operational data;
• and
• extracting feature data from the filtered time-series operational data and using the feature data to train the machine learning model to predict the run time of the sucker rod pump system until the next shutdown, wherein the feature data characterizes both past behavior of the sucker rod pump system and current operating condition of the sucker rod pump system;
• and
• ii) use the trained 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.