Hybrid training approach for artificial neural networks using genetic algorithms for rate of reaction estimation: Application to industrial methanol oxidation to formaldehyde on silver catalyst

A novel reactor simulator for the methanol oxidation to formaldehyde on silver catalyst was presented in this paper, including an original kinetic model based on artificial neural networks. The neural network training was performed using genetic algorithms associated with standard back-propagation, in order to improve the training efficacy, eliminating the effect of random initial weights estimates. Experimental data for training (rates of reaction) were obtained from process data (conversion and selectivity), using a back-calculation procedure through a simplified deterministic model implemented in the reactor simulator. Process data are widely available at industrial plants or literature and the proposed approach improves the response time to train the neural network in cases where rigorous kinetic experimental work cannot be conducted due to resource limitations. The simulator containing the trained artificial neural network was successfully validated with literature and industrial data, especially at industrial operating conditions, where available deterministic kinetic models for this system have failed. The simulator presented here, as well as the procedure to train the neural net consist in a powerful tool for plant process engineers to optimize the formaldehyde silver reactor in a timely and economical fashion.