AI Methods for Modeling the Vacuum Drying Characteristics of Chlorococcum infusionum for Algal Biofuel Production

Algal-based biofuels offer distinct advantages over other types of biofuels currently available within the fuel industry. However, one important disadvantage is that over their entire life cycle, they consume significant amounts of energy through cultivation, pretreatment, and production. Under pretreatment, drying is an energy-intensive yet highly critical process in standardizing the production of algal biofuel products. The current study proposes the use of artificial intelligence (AI) methods in optimizing the efficiency of a vacuum drying process. Previously, vacuum drying was modeled using least-squares regression methods, which captured the general linear or non-linear trend of the samples, but secured poor accuracy for individual sample points. In addition, these methods are unsuitable for online parameter optimization. Three AI-based models were developed to model the vacuum drying process, specifically an artificial neural network (ANN), a support vector machine (SVM), and an extreme gradient boosting machine (XGB). Based on error values, the ANN (RMSE = 0.0437) and XGB (RMSE = 0.0308) outperformed polynomial regression, and all models obtained meaningful lower values than multivariate linear regression (MLR). There is a notable difference in the ability of XGB to approximate individual sample points, particularly at high and low tail-ends of the dataset. Overall, the AI methods exhibited higher accuracy in estimating the drying characteristics for the chosen strain of algae. The current study may be extended to optimization by relating the control parameters to energy consumption, and automation based on the mathematical model.

Automated summary

AI methods, including ANN and XGB, demonstrate higher accuracy in optimizing the efficiency of a vacuum drying process for algal-based biofuels. These models outperform traditional regression methods and show notable improvement in approximating individual sample points, especially at high and low tail-ends of the dataset. This study suggests potential for further optimization and automation based on the AI models developed.