Deep learning-assisted pulsed discharge plasma catalysis modeling

Abstract

In this paper, a multi-layer feed-forward deep neural network was introduced into the plasma/plasma catalysis kinetics modeling. The deep learning-assisted modeling enables the initial input parameters for kinetics simulation, such as reduced electric field (E/N), to be extracted from specific experimental data after the neural network has been well-trained. The specific amplitudes of E/N and time t are set as the input data of the deep neural network, and the target product densities in time t calculated by the kinetics modeling are set as the output data. Replacing the kinetics simulation, the neural network can efficiently predict the target product densities under the fresh amplitudes of E/N. This method is validated in the plasma model for CH4/Ar pulsed discharge and the plasma catalysis model for N2/H2 pulsed discharge. The results indicate that the extended results calculated by the neural network are in good agreement with the numerical results calculated by the kinetics model and the relative error is 1.15 × 10–3 and 4.19 × 10–4, respectively, which might provide the possibility to assimilate experimental data and simulated data for optimizing research processes and integrating research results.