Intelligent sensor networks for monitoring well conditions and reservoir pressure

Abstract

The article substantiates the relevance of implementing intelligent sensor networks (ISN) for continuous monitoring of well conditions and reservoir pressure under conditions of high depletion of the well stock and the inertia of traditional control. It formulates the goal of quantitatively assessing the operational, predictive, and economic advantages of ISN compared with periodic measurements and wellhead telemetry. The materials and methods include five-year observations of 50 production wells in Western Siberia, where 10 wells are equipped with pilot industrial ISN (DTS/DAS and an array of point pressure sensors with LoRaWAN), and 40 are monitored using traditional approaches, with ISN data sampling at 1-second intervals and a total data volume exceeding 10 TB. Data processing was carried out using wavelet filtering and the Kalman algorithm, correlation-regression analysis, k-means/DBSCAN clustering for anomaly detection, and LSTM recurrent neural networks for reservoir pressure forecasting, with accuracy comparison to hydrodynamic and ARIMA models. ISN drastically increase operational responsiveness: the average data acquisition time is reduced from 24.73 hours to 0.02 hours, processing delay from 125.40 minutes to 1.15 minutes, and the update frequency rises from 0.85 to 43,200 times per day, effectively shifting control from discrete to quasi-real-time. For 30-day forecasting, LSTM achieves MAPE = 1.12% and RMSE = 0.16 MPa with , surpassing hydrodynamic models (MAPE = 8.43%) and ARIMA on ISN data (MAPE = 3.12%), which increases confidence in operational decision-making. In critical complication diagnostics, the average detection time for water breakthrough, sand production, and cavitation is reduced from 72.5/48.2/24.0 hours to 0.5/0.2/0.1 hours, with classification accuracy of 98.6–99.8%, enabling prevention of incident escalation. Economic assessment shows a reduction in OPEX by 18.57%, an increase in operating efficiency from 92.4% to 98.1%, and a decrease in losses due to emergency downtime by 85.21% per well per year, yielding a positive integral effect. The discussion interprets the synergetic effects: dense ISN data streams improve model accuracy, accelerating and reliably verifying anomaly detection, which translates into increased technical readiness and cost reduction. The study concludes that a transition toward a digital oilfield with dynamic digital twins of wells and reservoirs is underway and outlines prospects for integrating ISN with geophysical data and expanding neural network architectures to identify multifactor dependencies.