System for automated selection of equipment of installation of submersible electric centrifugal pumps to the well based on machine learning

Abstract

Currently, the largest oil companies in Russia are facing the problem of depletion of operated oil wells, which leads to an increase in the cost of produced raw materials. This stimulates the need to introduce better tools to increase the efficiency of downhole equipment of electric centrifugal pumps. This paper considers the construction of an automated software complex for selecting the characteristics of an electric centrifugal pump to a well based on artificial neural networks in the Python programming language using Tensorflow machine learning technologies. For analysis and formation of training samples data of production wells of Vankorskoye field are taken. The formed sample includes such variables as: production well flow rate, pump supply, watering, oil density, water density, depth of upper perforation holes, bottom hole depth, tubing lowering depth, dynamic level, formation pressure, wellhead pressure, fluid viscosity, feed coefficient, tubing outer diameter, tubing roughness, pipe wall thickness. These data directly affect the selection of the type and characteristics of electric centrifugal pumps, and have a mutual influence on each other. A calculation algorithm has been created. Data described, processed and prepared. The influence of optimization by selection of optimal characteristics of pump on power saving is considered. With the help of Tensorflow machine learning libraries, a model of the neural network has been created in the software PyCharm, predicting the optimal characteristics of a suitable electric centrifugal pump taking into account the influencing parameters.