Design and application of a hybrid predictive control framework for carbon capture in pressurized circulating fluidized bed coal-fired processes

Amid escalating climate change and pressing carbon neutrality goals, integrating pressurized circulating fluidized bed (PCFB) coal combustion with microchannel CO2 absorption offers a promising approach for enhanced carbon capture. To address challenges in operating-condition identification, mode switching, and control performance, this paper proposes a hybrid predictive control framework. An improved Long Short-Term Memory (LSTM) model, featuring CEEMDAN-based data preprocessing, a Multidimensional Channel Attention Mechanism (MDCAM), and an adaptive time–frequency domain loss function, achieves over 95 % recognition accuracy across 60 %–100 % load ranges. Coupling a Koopman operator with dictionary learning ensures smooth transitions among modes, reducing the RMSE to 0.0057 and limiting overshoot to 0.41 % under extreme conditions. Validation in a microchannel CO2 absorption setting demonstrates strong generalization, with RMSE values of 0.0137 and 0.0155 for constant and step-change kLa conditions, respectively, and a computation time of about 250 ms per step. These findings underscore the framework's potential to bolster dynamic control in PCFB-based microchannel carbon capture systems, contributing to carbon neutrality targets.