Prediction of higher heating value of hydrochars using Bayesian optimization tuned Gaussian process regression based on biomass characteristics and process conditions

Abstract

Hydrochars are valuable resources obtained from hydrothermal carbonization (HTC) of biomass. To optimize the reaction conditions of HTC, extensive experimentation is required, which is both costly and time consuming. In order to reduce the time and cost, this study develops new predictive models for higher heating values (HHV) of hydrochar based on Gaussian Process Regression (GPR), Ensemble, and Decision Tree (DT) algorithms using Bayesian Optimization (BO). This approach reduces prediction errors by combining GPR, Ensemble, and DT with BO. This is the first study on the application of BO for the hyperparameter selection as the basic learner. BO-GPR converged during training with the lowest Mean Absolute Error (0.1783) compared to BO-Ensemble (0.5128) and BO-DT (0.7430). The efficiencies of the algorithms were measured by the Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and coefficient of determination (R2). During the testing, BO-GPR generated MAE, RMSE and R2 of 0.4435, 0.5961 and 0.9705, respectively, and performed better than BO-Ensemble and BO-DT. The Nemenyi test showed that BO-GPR, BO-Ensemble, and BO-DT were statistically different in terms of their prediction ability. BO-GPR outperformed the other two methods.