Machine learning methods for understanding coal pyrolysis

Pyrolysis is a critical step in the thermal conversion of coal. A deep understanding of its mechanisms is essential for accurately predicting product yield and promoting clean, efficient coal conversion. However, the development of technologies for precise product yield prediction has been limited due to the heterogeneity of coal structures and the complexity of pyrolysis processes. Machine learning (ML), with its unique ability to handle complex, multidimensional and nonlinear systems, has attracted growing interest in coal pyrolysis research. This article provides a systematic review of the latest advances in applying ML models to coal and biomass pyrolysis, emphasising their significant potential for data-driven product yield prediction. While the 'black box' nature of ML models enables high-precision modelling, limitations exist in explaining the mechanisms of pyrolysis pathways. Therefore, exploring ML-driven methods to uncover pyrolysis mechanisms is crucial to bridge the gap between predictive modelling and understanding the scientific basis of coal conversion. To address core issues in the current coal pyrolysis field, such as fragmented, multi-source, heterogeneous data, and insufficient cross-scale correlation analysis, we propose constructing a high-throughput pyrolysis characterization platform and innovating data association mining techniques to establish a robust, multidimensional database. This would enable the development of practical ML models suitable for laboratory and industrial settings, reducing the experimental workload and the costs of industrial trial and error.