Optimizing pretreatment of Leucaena leucocephala using artificial neural networks (ANNs)

Abstract

This study deals with optimizing the pretreatment process for Leucaena leucocephala wood. Total 51 experiments were conducted. Three parameters were optimized i.e. catalyst concentration (1–3%), duration (120–300 min) and temperature (100–150 °C). Improvement in saccharification efficiency was evaluated using commercial cellulases. Single response, total reducing sugar yield, was estimated at the end of the experiments. The results of these experiments were analysed by three algorithms of neural networks i.e. Bayesian Regularization Neural Network (BRNN), Scaled Conjugate Gradient Neural Network (SCGNN), and Levenberg Marquardt Neural Network (LMNN). Among the three algorithms of ANN, BRNN gave most accurate predictions for total reducing sugar yield. At some points, BRNN and LMNN are almost equally efficient but BRNN had lower values for all error functions. SCGNN and LMNN are supervised in nature but Bayesian uses probabilistic optimization. Probably that makes this technique better than others.