Hydrogen prediction in poultry litter gasification process based on hybrid data-driven deep learning with multilevel factorial design and process simulation: A surrogate model

Abstract

Gasification is one of the recommended processes for poultry litter valorization, and its success is largely dependent on process input parameters for syngas production. The quality of syngas, characterized by a higher heating value (HHV) and lower heating value (LHV), is significantly influenced by hydrogen. In this research, integration of Aspen Plus simulation and Convolutional Neural Network (CNN) have been done to estimate the hydrogen content in syngas. The gasification process has been optimized using the response surface method, and the results have been supported by the particle swarm optimization (PSO) technique. The statistical analysis of the response surface model revealed that the optimal process parameters are around 408 °C, 2.0 (biomass to air ratio) BMR, and 2.0 bars. Interestingly, PSO led to nearly identical optimum values (400 °C, 2.0 BMR, and 2.0 bars), resulting in high-quality syngas. Furthermore, CNN exhibited a good predicting performance with a coefficient of determination (R2) exceeding 0.96, coupled with mean square error (MSE) and mean absolute error (MAE) of 0.01 and 0.05, respectively. This solidifies the integration of Aspen Plus simulations and CNN as an accurate surrogate model for predicting hydrogen levels during poultry litter gasification, enabling effective process optimization. Therefore, the proposed model serves as a reliable tool for predicting and optimizing the syngas produced during the gasification process of poultry litter, with potential applications in enhancing energy production and waste management practices.