The current coverage of oil wells with telemetry does not allow timely determination of deviations in the operation of about 40% of electric submersible pumps. To solve this problem, a model of virtual sensors has been developed that allows the prediction of temperature and pressure growth at the pump intake in the absence of submersible sensors based on modern big data processing and machine learning technologies. The developed models of virtual sensors are embedded directly into the process control system, which allows notifying the technologists and operators about a possible reduction in the planned average pump operating time and their possible failures for various reasons.