QUALITY CONTROL OF ECHOGRAMS WITH MACHINE LEARNING BASED APPROACH FOR DETERMINING THE DYNAMIC LEVEL IN PRODUCTION WELLS

The paper presents a new approach to assessing the reliability of the results obtained from the interpretation of echograms, which is used to determine the dynamic level of fluid in production wells. The comprehensive method combines both the evaluation of the quality of the obtained echograms and the consideration of additional historical parameters. Analysis of the conducted cases showed that it is not always possible to accurately recognize the response from the liquid level. This may be due to a number of reasons, such as external interference when recording the signal or its low amplitude. Incorrect echograms cannot be identified without the involvement of experts or the application of analytical methods; therefore, a machine learning model based on a convolutional neural network is used to assess the quality of the raw data. Additionally, an analysis of the correctness of the obtained dynamic level value is conducted in terms of its consistency with historical data from the well and the pump setting depth. The overall assessment of the reliability of the interpretation is calculated automatically, allowing for efficient processing of large volumes of information and making informed conclusions. The developed approach was tested on field data from reservoirs with low-permeability formations. The results demonstrated a high degree of convergence with expert assessments, confirming the reliability and applicability of the proposed method for quality control of echogram interpretation.