Application of artificial neural networks in predicting biomass higher heating value: an early appraisal

Abstract

ABSTRACT Biomass higher heating value (HHV) is the maximum energy released by its complete oxidation. The aim of this mini-review was to synthesize the early efforts of researchers in the prediction of biomass HHV by artificial neural network (ANN) models. This was conducted to evaluate the progress of research, identify knowledge gaps and synthesize future perspectives in the research area. Multi-layer perceptron artificial neural network (MLP-ANN) was observed to be the most accurate ANN model for the prediction of biomass HHV. Model accuracy was more dependent on the ANN architecture than on the data size. Investigations based on ultimate analysis data (either singularly or combined with proximate analysis data) gave more accurate models. Evaluating more intricate and sophisticated ANN architectures could yield better models for biomass HHV prediction. ANN models based on chemical analysis and physical properties data are unreported and could be explored in future studies. There is likely to be a paradigm shift in biomass HHV prediction as soon as the more accurate ANN models become more popular with researchers in biomass energy. Abbreviations AI Artificial Intelligence; ANFIS-PSO: Adaptive Neuro-Fuzzy Inference System with Particle Swamp Optimization; ANN: Artificial Neural Networks; FNPLS: Network-based Fuzzy Partial Least Squares; HHV: Higher Heating Value; IFNPLS: Iterative Network-based Fuzzy Partial Least Squares; INNPLS: Iterative Neural Network Adapted Partial Least Squares; LHV: Lower Heating Value; MLP-ANN: Multi-Layer Perceptron Artificial Neural Network; NNPLS: Neural Network Adapted Partial Least Squares; PCA-ANN: Principal Component Analysis with ANN Paradigm; PCA-IFNPLS: Principal Component Analysis with Iterative Neural Network Adapted PLS; PLS: Partial Least Squares; R2: Coefficient of Determination; RMSE: Root Mean Square Error