Uncertainty propagation to techno-economic outcomes through Gaussian process surrogate modeling of fast pyrolysis

Abstract

This study evaluates how predictive uncertainty in process outputs propagates into techno-economic indicators for fast pyrolysis. A multi-output Gaussian Process (GP) model was trained using 103 literature-derived data points to predict product yields and utility demands. Correlated samples from the GP predictive distributions were propagated through a simplified techno-economic analysis (TEA) model to obtain distributions of revenue, utility costs, EBITDA, and minimum selling price (MSP). Three feedstocks were evaluated at two temperatures. At 550 °C, the EBITDA distributions differed clearly between feedstocks. Wheat straw gave the most favorable operating outcome, with the highest mean EBITDA, 0.435 M$/y, and the lowest probability of negative EBITDA, 23.0%. Empty fruit bunches remained positive on average (mean EBITDA of 0.168 M$/y), but its higher drying cost increased the probability of negative EBITDA to 28.8%. Mallee bark had the broadest EBITDA distribution, indicated by a standard deviation of 1.365 M$/y, and the highest probability of negative EBITDA, about 46%. At 730 °C, EBITDA distribution widths became more similar, with standard deviations between 1.342 and 1.387 M$/y. However, empty fruit bunches showed the highest probability of negative EBITDA, 58.4%, mainly because of its higher drying cost. The MSP distributions provided a complementary output. The probability of requiring an MSP above 1000 $/t was highest for mallee bark at 550 °C and empty fruit bunches at 730 °C. Overall, propagating GP predictive uncertainty through TEA provides a practical way to quantify expected economic performance and downside risk under data-scarce conditions.