Artificial Intelligence Methods for Predicting Pyrolysis Product Yields of Lignocellulosic Biomass

The study is devoted to the development of neural network models for predicting the yield of final products during pyrolysis of plant biomass. To train the models, we use data obtained from full-scale experiments on heating various types of raw materials in a cylindrical reactor. The main features are the chemical and physical characteristics of the biomass composition, the parameters of the temperature regime of the pyrolysis process. The predicted final yields of pyrolysis products are liquid pyrolysis products, including resin; solid pyrolysis products and pyrolysis gas in percentage. Based on these data, various machine learning models were developed. As a data set for training the models, we used 750 records from full-scale experiments. As a result, a model was selected from them that allows predicting the ratios of various fractions of the output product during pyrolysis with a sufficiently high accuracy RMSE = 4.5872, R² = 0.74 which opens up new opportunities for optimizing pyrolysis plants and increasing their energy efficiency.