Document Type : Review Paper
School of Energy Science and Engineering, Central South University, Changsha, Hunan 410083, People’s Republic of China.
Faculty of Engineering and Natural Sciences, Tampere University, Tampere, Finland.
Thermochemical treatment is a promising technique for biomass disposal and valorization. Recently, machine learning (ML) has been extensively used to predict yields, compositions, and properties of biochar, bio-oil, syngas, and aqueous phases produced by the thermochemical treatment of biomass. ML demonstrates great potential to aid the development of thermochemical processes. The present review aims to 1) introduce the ML schemes and strategies as well as descriptors of the input and output features in thermochemical processes; 2) summarize and compare the up-to-date research in both ML-aided wet (hydrothermal carbonization/liquefaction/gasification) and dry (torrefaction/pyrolysis/gasification) thermochemical treatment of biomass (i.e., predicting the yields, compositions, and properties of oil/char/gas/aqueous phases as well as thermal conversion behavior or kinetics); and 3) identify the gaps and provide guidance for future studies concerning how to improve predictive performance, increase generalizability, aid mechanistic and application studies, and effectively share data and models in the community. The development of biomass thermochemical treatment processes is envisaged to be greatly accelerated by ML in the near future.
- Machine learning (ML) schemes and descriptors of input and output features were examined.
- ML can predict yields, compositions, and properties of oil/char/gas/aqueous phases.
- ML predictive performance for wet and dry thermochemical processes is different.
- Improving ML predictive performance, generalizability, and application are future needs.