A Data-Driven ML Model for Sand Channel Prediction from Well Logs for UTES Site Optimization and Thermal Breakthrough Prevention: Hungary Case Study

This study presents a data-driven approach to predict the three-dimensional distribution of sand-rich channels in hydrocarbon reservoirs using well log data, aiming to optimize site selection for Underground Thermal Energy Storage (UTES) and manage hot and cold well pairs effectively. Leveraging detailed petrophysical datasets from 128 hydrocarbon exploration wells within the Szolnok Formation in southern Hungary, the developed machine-learning workflow—combining XGBoost regression and spatial residual correction—accurately delineated permeable channel systems suitable for thermal energy injection and extraction. The model achieved robust predictive performance (R 2 = 0.92; RMSE = 0.24), and correlation analyses confirmed significant relationships between predicted channels and sand content and shale content. Clearly identified high-permeability channel zones facilitated strategic well placement, significantly reducing the risk of premature thermal breakthrough and enhancing the reliability and efficiency of UTES operations.