Theoretical Models Constructed by Artificial Intelligence Algorithms for Enhanced Lipid Production: Decision Support Tools

Abstract

Theoretical models that predict the lipid content of microalgae are an important tool for increasing lipid productivity. In this study, response surface methodology (RSM), RSM combined with artificial neural network (ANN), and RSM combined with ensemble learning algorithms (ELA) for regression were used to calculate the maximum lipid percentage (%) from Chlorella minutissima (C. minutissima). We defined one set of rules to achieve the highest lipid content and used trees.RandomTree (tRT) to simulate the process parameters under various conditions. Among the various models, results showed the optimum values of the root mean squared error (0.2156), mean absolute error (0.1167), and correlation coefficient (0.9961) in the tRT model. RSM combined with tRT estimated that the lipid percentage was 30.3% in wastewater (< 35%), lysozyme (≥ 3.5 U/mL), and chitinase (< 15 U/mL) concentrations, achieving the best model based on experimental data. The optimal values of wastewater concentration, chitinase, and lysozyme were 20% (v/v), 5 U/mL, and 10 U/mL, respectively. Also, the if-then rules obtained from tRT were also used to test the process parameters. The tRT model served as a powerful tool to obtain maximum lipid content. The final rankings of the performance of various algorithms were determined. Furthermore, the models developed can be used by the fuel industry to achieve cost-effective, large-scale production of lipid content and biodiesel.