Machine Learning-Assisted Tailoring of Pore Structures in Coal-Derived Porous Carbons for Enhanced Performance

Porous carbon materials (PCMs) have emerged as key players in energy storage and environmental remediation thanks to their highly tunable pore structure parameters. However, traditional empirical approaches for optimizing these parameters are time-consuming and resource-intensive. Herein, a workflow is introduced that integrates machine learning-assisted pore parameters prediction with experimental validation for PCMs, thereby facilitating the identification of candidate PCMs for different application requirements. Coal-based activated carbon (CAC) is first employed to validate the effectiveness of this workflow, given its high tunability and extensive application potential. During the validation process, machine learning models are developed to establish a predictive map linking precursor and preparation parameters to the resulting specific surface area and total pore volume. The proposed strategy for CAC is applied to the design of a high-performance supercapacitor electrode (specific capacitance of 491.2 F g^-1 at 0.1 A g^-1) and a methylene blue adsorbent (adsorption capacity of 2196.8 mg g^-1 at room temperature), demonstrating its effectiveness. Furthermore, the workflow is extended to coal tar pitch-based carbon materials, a more complex system, and demonstrates encouraging outcomes. This work extends the strategies for controlling pore structure parameters of CAC and provides a transferable workflow for exploring other PCMs.