Modeling and prediction of biodiesel production by using different artificial intelligence methods: Multi-layer perceptron (MLP), Gradient boosting (GB), and Gaussian process regression (GPR)

Abstract

In this study, different distinct approaches of machine learning (ML) including Multi-layer perceptron (MLP), Gradient Boosting with DT (GBDT), and Gaussian process regression (GPR) were employed in order to predict the amount of Papaya oil methyl ester (POME) biodiesel production. To optimize the POME production, yield of these models were optimized with focus on maintaining generality and enhancing the prediction accuracy. The influencing transesterification factors on the biodiesel manufacture like the temperature of reaction (℃), amount of sodium hydroxide as catalyst (wt.%), treatment time (min), and methanol to papaya oil molar ratio were chosen as the inputs. NaOH was employed as a catalyst at the phase boundary for the reaction between papaya oil and short chain alcohols. Considering the MAPE criterion, the MLP, GBDT and GPR models have shown the error rates of 8.9670E-02, 2.0324E-01 and 7.2080E-02, respectively. Similarly, the GPR process gets the best R2 criterion score of 0.996, followed by GBDT with 0.989 and MLP with 0.971. The Mean Absolute Error (MAE) also shows the best model is the Gaussian process, which has an error rate of 4.7. In addition, the optimal POME yield production value was estimated through the proposed method to be about 99.96%, in the optimized values of 64 ℃, 0.875 wt%, 7.375 min, and 10.875 for the temperature reaction (℃), amount of catalyst, treatment time, and methanol to papaya oil molar ratio, respectively.