Comparative Study between Cascade-Forward Neural Network and Feedforward Neural Network Models for Enzymatic Polymerization Process

Abstract

Enzymatic polymerization has gained attention as a more environmentally friendly alternative to conventional chemical methods, as it avoids the use of fossil fuel-derived monomers and requires milder operating conditions. However, the reaction kinetics and mechanism of the polymerization are complex. The process is also sensitive to the operating conditions, as harsh operating conditions would negatively impact the desired molecular weight of the resulting polymer. To address this, the process model must be developed accurately to predict the molecular weight under suitable conditions. This study compares the effectiveness of two types of artificial neural networks (ANN), namely cascade-forward neural network (CFNN) and feedforward neural network (FFNN) to determine the better-performing ANN type for the enzymatic polymerization process. A total of 84 experimental datasets obtained from laboratory batch study were used, which include the inputs of reaction time, reaction temperature and reactor impeller speed, as well as polymer molecular weight as the output. The effects of these operating inputs on the outputs were also investigated through sensitivity analysis.  Twelve training algorithms for both CFNN and FFNN were compared and evaluated in terms of mean square error (MSE), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), regression (R), determinant coefficient (R²), and accuracy. CFNN model with the Levenberg-Marquardt backpropagation (LM) achieved the highest accuracy of 99.97%, yielding the least MSE, RMSE, MAE, MAPE and R of 2.36, 1.53, 0.80, 0.03%, 1, respectively. The sensitivity analysis showed that both temperature and speed have a significant negative impact on the molecular weight of polymer compared to the time input.