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

Area of Science:

• Waste-to-energy technologies
• Sustainable resource management
• Applied machine learning

Background:

• High waste generation necessitates efficient management and bioenergy production for net-zero CO2 goals.
• Feasibility studies are crucial for optimizing waste conversion processes, especially with diverse waste streams.
• Funding challenges, particularly in developing countries, hinder experimental investigations.

Purpose of the Study:

• To develop a machine learning-based statistical model for predicting product yields from waste thermochemical conversion.
• To guide researchers in selecting optimal conversion processes (slow pyrolysis, fast pyrolysis, gasification) based on waste composition and operating conditions.
• To minimize experimental runs, saving time and financial resources.

Main Methods:

• Utilized a dataset of 144 published experimental samples.
• Applied machine learning to build statistical models correlating product yields with waste composition and operating parameters.
• Developed a decision matrix based on the statistical models for process selection.

Main Results:

• Statistical models were established with a 95% confidence level.
• Models predict product yields for slow pyrolysis, fast pyrolysis, and gasification.
• A decision matrix guides process selection based on waste carbon and hydrogen content.

Conclusions:

• Machine learning models effectively predict bioenergy yields, aiding process selection.
• Gasification is favored for moderate carbon (40-46%) and high hydrogen content waste.
• Slow pyrolysis is preferred for moderate carbon waste with lower hydrogen, while fast pyrolysis suits high carbon waste (>47%).