A comparison of machine learning algorithms for estimation of higher heating values of biomass and fossil fuels from ultimate analysis

Higher heating value (HHV) is one of the most important parameters to consider while obtaining energy efficiently from fuels. It provides means to estimate the quality of the fuel. However, measuring the HHV of fuels requires advanced devices, is expensive and time consuming. Thus, sufficient estimation of the HHV is crucial for development of fuel technologies and numerous studies have been performed about this subject. In this study, several machine learning algorithms (DTR, SVR, GPR, RFR, Multi-Linear Regressions, and Polynomial Regression) were utilized to construct models for estimation of the HHV from the largest open dataset of ultimate analysis of fuels on a dry, ash-free basis. The mathematical analysis is conducted for numerous different solid, liquid, and gaseous fuels such as biomass, biochar, municipal solid waste, kerosene, gasoline, fuel oil, algae, natural gas etc. 10-fold cross validation method was used to assess the validity of the constructed models in an unbiased manner. The R-2, adjusted R-2, RMSE, N-RMSE, and AAE values of each model were computed for performance evaluations. Finally, the results were compared with the literature, and advantages and disadvantages of each method were discussed in terms of both computational complexity and prediction accuracy. The RFR and DTR models performed exceptionally well in estimation of HHV of all classes of fuels with R-2 values of 0.9814 and 0.9664, respectively. In addition, the statistical values of the ultimate analysis for each constructed class of fuel are investigated for each class of fuel and an extensive Krevelen diagram was produced from the largest dataset to date, which illustrated the relationships between the atomic O/C ratio and the atomic H/C ratio of the fuels.