Comparison of neural network and response surface methodology techniques on optimization of biodiesel production from mixed waste cooking oil using heterogeneous biocatalyst

The present research aims to optimize the process parameters of biodiesel production from mixed waste cooking oil, consisting of sunflower and palm oil. The biocatalyst was synthesized from Chinese broccoli, Cauliflower and Napa cabbage vegetable wastes and potassium hydroxide-impregnated biocatalyst was synthesized through the incipient wet impregnation method. Artificial neural networks (Scaled Conjugate Gradient, Bayesian Regularization and Levenberg Marquardt Neural Network) and RSM-based central composite design are used to optimize the parameters in biodiesel production. The obtained predicted results from ANN and RSM models are compared with experimental results. The experimental results showed that biocatalyst impregnated with KOH has produced a maximum yield of 94.7 % at a reaction temperature of 60 degrees C, time of 120 min, catalyst loading of 7 wt%, 1:9 M ratio and 900 rpm stirrer speed. In addition, among the three algorithms of ANN, BRNN neural network had better prediction results in biodiesel production. At the same time, LMNN, and SCGNN are nearly similarly efficient on the other hand the BRNN model has lower error values. The ANN and CCD models have demonstrated lower RMSE and MAPE such as 0.001 to 0.018, and 3.22-4.37 % respectively, and high R2 values in the ranges of 0.97 to 0.99, indicating the excellent reliability of these two approaches for process optimization. Hence, the above study recommended a biocatalyst impregnated with KOH as a catalyst to achieve maximum biodiesel yield. Further, ANN and RSM techniques are suggested for biodiesel process optimization and yield prediction.