Electric Parameters Assist Rod-Pump Condition Diagnosis, Production Metering

Abstract

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 204785, “Research and Application of Rod-Pump Working Condition Diagnosis and Virtual Production Metering Based on Electric Parameters,” by Ruidong Zhao, SPE, Cai Wang, and Hanjun Zhao, SPE, PetroChina, et al. The paper has not been peer reviewed. The conventional configurations of the Internet of Things (IoT) related to pumping wells include an electric parameter indicator and dynamometer. If working-condition diagnosis and virtual production metering of well pumping can be realized through electrical parameters, the need for dynamometers can be eliminated or reduced, with a potentially major effect on reducing investment and improving coverage of the IoT in oil wells. In the complete paper, an integrated multimodel diagnosis method is proposed. The diagnostic and virtual-production-metering method and software based on electrical parameters have been applied in many of the operator’s oil fields. Introduction In China, more than 90% of the more than 400,000 oil wells are sucker-rod-pump (SRP) wells. At present, with the popularity of IoT in the oil field, SRP wells have accumulated massive amounts of production and operation data such as polished rod load, displacement, and electric-motor-power parameters. The working performance diagnosis and production calculation for these wells mostly are based on dynamometer cards, whose IoT costs are relatively high and whose widespread applications are limited. In contrast, the acquisition of electric parameters is much cheaper and easier than is the case with dynamometer cards, but a lack of corresponding theories and efficient methods to make use of electric parameters persists. If a power curve can be transformed to a dynamometer card accurately in the same well working condition, well diagnosis and production calculation can be realized by the generated card. At present, few reports on transforming power curves to dynamometer cards exist. The approach to generate a dynamometer card from an electric power curve is to convert 1D timeseries data to 2D load and displacement-series data. Building the mathematical transformation model to generate cards from power curves is difficult because of uncertainties related to pumping units during working. Thus, transformation methods, by virtue of deep learning, seem more promising—specifically, transforming electric power to dynamometer cards by deep learning.