Prediction of the higher heating value of biomass based on multiple classification methods

In general, measuring the higher heating value (HHV) of biomass by experimental methods is time-consuming and complex. These problems can be effectively avoided by studying HHV prediction of biomass. This study innovatively predicts the HHV of biomass from a new perspective, that is, biomass data samples are classified according to three different classification methods: biomass categories classification, K-means clustering classification, and ash and volatile matter (VM) content classification, and then predicting the HHV of biomass. It demonstrated that the predictive result of the classification methods was better than that of the non-classified. In the three methods, the method of classifying biomass by ash and VM content showed the best effect, with an average absolute error (MAE) of 0.755 MJ/kg and an average absolute percentage error (MAPE) of 4%. It revealed that modeling after sample classification can effectively improve the prediction accuracy of biomass HHV. It is a simple and feasible strategy and method. The improvement of prediction accuracy may be through different classification methods, which makes it easier for the model to capture the relationship between specific types of HHV, ash, VM, and fixed carbon (FC) in the prediction process, thus improving the efficiency of the model. Meanwhile, this study will provide a reference significance for the prediction of the products and properties of biomass in thermochemical conversion technology.