Utilizing machine learning in predicting yields of products in biomass thermochemical conversion processes

Agricultural and municipal waste management is one of the most important topics nowadays due to the high rate of waste generation. Bioenergy is one of the key technologies to achieve the global net zero CO2 emissions goal. In the era of biorefineries, and especially for new types of waste and new waste blends, a feasibility study is required to identify the optimum waste conversion process which is usually preceded by experimental investigation. Especially for developing countries, securing the corresponding funds is usually challenging. In this research work, machine learning has been applied to develop a statistical model based on 144 samples of the published experimental data for different types of wastes undergoing three thermochemical conversion processes; namely, slow pyrolysis, fast pyrolysis and gasification. Statistical analysis has been performed and models were built with (95% confidence level) which correlate the various products' yields with the waste composition and operating conditions. These models provide a guide to researchers regarding the expected yield of products for each of the three studied processes so as to discard the non-promising processes for the used type of waste. This will accordingly minimize the number of required experimental runs; hence saving time and money. A decision matrix was also developed based on the statistical models of each of the studied thermochemical processes. It was concluded that when carbon content is moderate (40-46%, by mass), gasification process is favored if the hydrogen content is high; otherwise, slow pyrolysis is favored. On the other hand, if the waste has high carbon content (above 47%) then fast pyrolysis is favored.