(365d) Machine-Learning Reduced Order Model for Cost and Emission Assessment of a Pyrolysis

Biomass pyrolysis is a promising approach for producing economical and environmentally-friendly fuels and bioproducts [1]. Biomass pyrolysis experiments show that feedstock properties significantly impact product yields and composition[2], [3]. Scientists are developing detailed chemical reaction mechanisms to capture the relationships between biomass composition and pyrolysis yields [4]. These mechanisms can be computationally intensive[5]. A few studies have investigated the use of machine learning (ML) algorithms in the estimation of pyrolysis products to circumvent the computationally expensive chemical reaction mechanisms and computational fluid dynamics (CFD) models[6], [7]. However, to the best of the author’s knowledge, none has integrated ML models into process modeling tools for real-time estimation of pyrolysis products to assess the cost and emissions of a pyrolysis refinery.

In this study, we investigate the use of a machine learning reduced order model (ROM) for assessing the pyrolysis products used in estimating costs and emissions of a pyrolysis biorefinery in real-time. We developed a Kriging-based ROM to predict pyrolysis yields of 314 feedstock samples based on the results of a detailed chemical kinetic pyrolysis mechanism. The ROM is integrated into a chemical process model for calculating mass and energy yields in a commercial-scale (2000 tonne/day) biorefinery. All Kriging models achieved excellent accuracy with an average coefficient of determination score of 99% when the predicted results were compared to the chemical kinetics results. The ROM estimated biofuel yields of 65 to 140 gallons per ton of dry biomass. This results in biofuel minimum fuel-selling prices of $2.62 to $5.43 per gallon and emissions of -13.62 to 145 kg of CO2 per MJ of biofuel. The ROM achieved an average mean square error of 1.8e-20 and a mean absolute error of 0.53%. These results suggest that ROMs can facilitate rapid feedstock screening for biorefinery systems.

Figure 1: Machine learning (Kriging), lifecycle cost (TEA), and emissions (LCA) prediction framework for chemical process simulations with machine learning reduced-order models implemented using the CAPE-OPEN standard. Rounded rectangles represent data, filled squares represent software and sub-processes, and the diamond represents a decision step. Process outcomes include minimum fuel-selling price (MFSP) and greenhouse gas (GHG) emissions.

FUNDING

The authors would like to acknowledge the support of the U.S. Department of Energy through the grant DE-EE0008326.

REFERENCES

[1] D. O. Hall and J. I. House, “Biomass: a modem and environmentally acceptable fuel,” Sol. Energy Mater. Sol. Cells, vol. 38, no. 1–4, pp. 521–542, Jan. 1995, doi: 10.1016/0927-0248(94)00242-8.

[2] W. Li, Q. Dang, R. C. Brown, D. Laird, and M. M. Wright, “The impacts of biomass properties on pyrolysis yields, economic and environmental performance of the pyrolysis-bioenergy-biochar platform to carbon negative energy,” Bioresour. Technol., vol. 241, pp. 959–968, 2017, doi: 10.1016/j.biortech.2017.06.049.

[3] P. A. Meyer, L. J. Snowden-Swan, S. B. Jones, K. G. Rappé, and D. S. Hartley, “The effect of feedstock composition on fast pyrolysis and upgrading to transportation fuels: Techno-economic analysis and greenhouse gas life cycle analysis,” Fuel, vol. 259, no. September 2019, p. 116218, 2020, doi: 10.1016/j.fuel.2019.116218.

[4] E. Ranzi, P. E. A. Debiagi, and A. Frassoldati, “Mathematical Modeling of Fast Biomass Pyrolysis and Bio-Oil Formation. Note I: Kinetic Mechanism of Biomass Pyrolysis,” ACS Sustain. Chem. Eng., vol. 5, no. 4, pp. 2867–2881, 2017, doi: 10.1021/acssuschemeng.6b03096.

[5] P. Mellin, E. Kantarelis, and W. Yang, “Computational fluid dynamics modeling of biomass fast pyrolysis in a fluidized bed reactor, using a comprehensive chemistry scheme,” Fuel, vol. 117, no. PART A, pp. 704–715, 2014, doi: 10.1016/j.fuel.2013.09.009.

[6] H. Merdun and I. V Sezgin, “Modelling of pyrolysis product yields by artificial neural networks,” Int. J. Renew. Energy Res., vol. 8, no. 2, pp. 1178–1188, 2018, [Online]. Available: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85049086039&partnerI....

[7] Q. Tang et al., “Prediction of Bio-oil Yield and Hydrogen Contents based on Machine Learning Method: Effect of Biomass Compositions and Pyrolysis Conditions,” Energy & Fuels, Aug. 2020, doi: 10.1021/acs.energyfuels.0c01893.