Machine learning-driven prediction of coal pyrolysis product distribution and multi-objective optimisation of Pareto frontier

Coal occupies a large proportion of global energy consumption; improving the efficiency of its thermochemical conversion is essential for energy security, low-carbon transition, and value-added utilisation. Pyrolysis is the leading step of coal utilisation. Predicting the distribution of pyrolysis products is the basis of pyrolysis process optimisation and targeted product regulation. Traditional experimental methods are time-consuming and costly. Existing kinetic prediction models suffer from structural complexity or limited accuracy, which cannot meet the needs of industrial production and scientific research. To achieve accurate and low-cost prediction of coal pyrolysis product, an enhanced hybrid prediction model named SSA-BOA-BPNN was constructed by integrating the sparrow search algorithm (SSA) and the butterfly optimisation algorithm (BOA) with the back-propagation neural network (BPNN). The average R2 values reached 0.9508, 0.9699, 0.9294, 0.9758 for the three-phase product distribution, tar component distribution, tar fraction composition, and pyrolysis gas composition, respectively. The prediction accuracy and generalisation performance were substantially outperforming conventional BPNN, single-algorithm optimisation models and other commonly used models. To provide guidance for the regulation of pyrolysis products and process optimisation, the optimal pyrolysis conditions for typical bituminous coal and lignite corresponding to the optimal range of optimisation objectives (including tar yield and lighter tar proportion, the contents of aromatic hydrocarbon and phenolic compound in tar, gas yield and tar yield) were obtained by Multi-objective Pareto frontier optimisation. The research results realised the accurate, efficient and low-cost prediction of coal pyrolysis product distribution, which provided an important reference for promoting the development of coal thermochemical conversion technology.