Ranking the Feedstocks Using Neural Network-Based System for Biofuel Production

The progressive movement from fossil to renewable fuels like biofuels is one of this century's challenges. Identifying the right feedstock is the biggest challenge in generating high-quality, sustainable second-generation biofuels. The solution to these problems can be found in second-generation agro-industrial biofuels, as their use does not compete with food resources; it also allows low-commercial raw material processing and provides an alternative to the disposal of such raw material. This study explores an innovative approach to classifying feedstocks, using secondary literature data sources. Besides the high reliability of methods used, the study analysed researching over 20 parameters of 106 feedstocks. The research developed a rating system for Multi-Criteria Decision Analysis (MCDA), including weighting for each parameter based on expert opinion or statistical methods such as PCA. The ranking system output then be fed into a Multivariate Regression (MVR) and a Multilayer perceptron (MLP), to rank feedstock to produce the highest quality sustainable biofuels for a particular location. The above rating will first be based on particular location agroclimatic characteristics, including Rockhampton (e.g., rainfall, temperature, light intensity, etc.). The algorithm can then be used at other locations (e.g., Brisbane, Cairns and Townsville) to override the agroclimatic parameter. After comparing performance of MVR to MLP, the study suggests that MLP output produces statistically significant results referring to rank the feedstocks. ANN model implemented in Python proved adequate to classify the feedstock successfully. It achieved a high validation accuracy calculated by the R
²
metric at 0.68. This result suggests that adding more data gives added significance to the results. Outputs would support prospective buyers and also be a thorough feedstock analysis.