A CNN-LSTM-PSO-based data-driven closed-loop intelligent control model for deep coalbed methane extraction

Abstract

Deep coalbed methane (CBM) extraction systems are characterized by strong parameter coupling, delayed response and reservoir-damage risk under complex deep-reservoir conditions. To address these issues, a data-driven CNN-LSTM-PSO closed-loop intelligent control model is proposed for the integrated wellbore-surface extraction system. CNN extracts spatial features from multi-source sensing data, LSTM predicts the temporal evolution of extraction parameters, and an improved PSO algorithm optimizes extraction negative pressure, valve opening and flow distribution under engineering constraints. Field tests show that the model achieves a coefficient of determination ((𝑅 2 ) of 0.955. Compared with conventional manual control, average productivity increases by 19.2%, unit energy consumption decreases by 22.1%, reservoir damage decreases by 46.3%, and response time is shortened by 90%.