Thermal effect evaluation of tar-rich coal pyrolysis driven by product-oriented and machine learning

Abstract

The thermal effect of coal pyrolysis is a key parameter in the process of coal pyrolysis mechanism research and reactor design. At present, there are a lot of differences in the accuracy of evaluation methods for coal pyrolysis thermal effect. The most widely used and feasible method is based on simultaneous thermal analyzer (TG-DSC synchronous combination). However, this method still relies heavily on the measurement accuracy of the instrument, and there are also some differences in the accuracy of thermal effect calculation methods after the heat flow curve is obtained, resulting in a poor repeatability of the data and difficulty in adapting to the needs of large-scale production and process design. Tar-rich coal is high-quality raw material for coal to tar, it contains more hydrogen-rich structures such as aliphatic side chains and bridge bonds, it could be pyrolyzed to produce more tar during pyrolysis. The regulation of pyrolysis process is the key factor affecting the quality of tar produced by the pyrolysis of tar-rich coal, and the thermal effect of pyrolysis process is an important parameter for selecting pyrolysis process. Therefore, it is urgent to develop an accurate and efficient evaluation method of coal pyrolysis thermal effect for both tar-rich coal pyrolysis and generalized coal conversion. Based on the experimental results for the slow pyrolysis of tar-rich coal, combined with pyrolysis reaction mechanism and empirical formula, the product-oriented pyrolysis reaction system of tar-rich coal was constructed. The chemical reaction heat for the pyrolysis process of tar-rich coal was obtained by the thermodynamic calculation of the reaction system using the classical thermodynamic analysis method. Combined with the physical heat absorption in the pyrolysis process, the thermal effect in the low-temperature slow pyrolysis process of tar-rich coal could be obtained. Then, based on the reported coal pyrolysis experimental results and measured values of pyrolysis thermal effect, the random forest model of machine learning method was used to nonlinearly model and predict the thermal effect of coal pyrolysis. The results showed that based on experimental results, the thermal effect value for slow pyrolysis of tar-rich coal calculated by the product-oriented calculation was generally lower than that measured by TG-DSC synchronous combination, and the error was within 10%. The prediction accuracy for the thermal effect value of tar-rich coal pyrolysis based on machine learning algorithm was 0.935 2. Overall, the two prediction models of thermal effect of tar-rich coal pyrolysis constructed were practical and applicable.

Keywords

• tar-rich coal
• pyrolysis
• thermal effect
• product-oriented
• machine learning