Online Deep Transfer Learning and Multi-Sensor Analysis for Enhanced Underground Monitoring

Underground environments present unique challenges for monitoring hazardous conditions due to their complex and dynamic nature. Current prediction methods often suffer from data scarcity, especially during the early operation phases, and fail to adapt to changing conditions. This paper aims to address these issues by integrating multi-sensor analysis with online transfer learning techniques to enhance methane concentration monitoring in underground settings. The methodology involves training a baseline Long Short-Term Memory (LSTM) model on open-source data from a Polish coal mine and fine-tuning it using an online learning approach with data obtained from a Chinese coal mine. The results demonstrate significant improvements in prediction accuracy and reliability, especially during the initial stages of monitoring when data are sparse, achieving an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.93 when using the online transfer learning model. This study helps improve the safety and efficiency of underground monitoring systems by addressing initial data scarcity and enhancing model adaptability to dynamic environmental conditions.