| Summary: | In this study, three novel ensemble algorithms, namely, simple averaging, weighted averaging, and meta-learning ensemble algorithms were employed to predict the higher heating value of biomass. These strategies were implemented in two main stages. In the first stage, four heterogeneous standalone models: an artificial neural network, a multivariate regression, a support vector regression, and an adaptive neuro-fuzzy inference system (ANFIS) were developed to predict the higher heating value. In the second stage, the outputs of the standalone models were aggregated for ensemble learning implementation. Seven input combinations of the biomass proximate analysis components formed the proposed models’ inputs. In the pre-ensemble phase, the ANFIS model having ash and volatile matter as an input combination presented the most accurate performance based on the Willmott's index of agreement of 0.9741 and the mean square error of 0.0032. The ensemble algorithms demonstrated improvement in the overall prediction performances with the meta-learning ensemble ranked superior for an average error decrease of up to 15% when ash, volatile matter, and fixed carbon served as the model's input combination. The findings of this work provide a robust foundation for the use of ensemble algorithms in the prediction of the biomass higher heating value.
|
|---|