Application of machine learning methods for predicting well disturbances

Abstract

In the process of field exploration, along with regular flooding, a significant part of the wells is flooded prematurely due to leakage of the string and outer annulus. In an effort to intensify the flow of oil to the bottom of wells in field conditions, specialists often try to solve this problem by using various technologies that change the reservoir characteristics of the formation. Any increase in pressure that exceeds the strength of the rocks in compression or tension leads to rock deformation (destruction of the cement stone, creation of new cracks). Moreover, repeated operations under pressure, as a rule, lead to an increase in water cut and the appearance of behind-the-casing circulations. For that reason, an important condition for maintaining their efficient operation is the timely forecasting of such negative phenomena as behind-casing cross flow and casing leakage. The purpose of the work is to increase the efficiency of well interventions and workover operations by using machine learning algorithms for predicting well disturbances. Prediction based on machine learning methods, regression analysis, identifying outliers in the data, visualization and interactive processing. The algorithms based on oil wells operation data allow training the forecasting model and, on its basis, determine the presence or absence of disturbances in the wells. As a result, the machine forecast showed high accuracy in identifying wells with disturbances. Based on this, candidate wells can be selected for further work. For each specific well, an optimal set of studies can be planned, as well as candidate wells can be selected for further repair and isolation work. In addition, in the course of this work, a set of scientific and technical solutions was developed using machine learning algorithms. This approach will allow predicting disturbances in the well without stopping it.