Development of an optimal bilayered back propagation neural network (BPNN) to identify thermal behaviors of reactions in isoperibolic semi-batch reactors

Abstract

It is very important to identify the thermal behaviors of semi-batch reactors (SBRs) in isoperibolic operating mode. First, a default bilayered back propagation neural network (BPNN) is selected from multiple recognition algorithms to achieve this task, which is suitable for three kinds of reactions with arbitrary reaction orders: homogenous, kinetically−controlled, and diffusion−controlled liquid−liquid heterogeneous reactions. Then, it is further optimized by Bayesian regularization and genetic algorithm (GA). The result shows that the optimal bilayered BPNN undoubtedly has better recognition accuracy and generalization performance. It is found that the accuracy of the two test sets are 98.8% and 100%, respectively. Subsequently, a simple and user-friendly standalone desktop application program (app) is designed to easily use the developed optimal bilayered BPNN. Finally, a case of acetic anhydride hydrolysis reaction with acetic acid solvent is studied to prove the good validity of the bilayered BPNN and the great reliability of the standalone desktop app.