Speeding up turbulent reactive flow simulation via a deep artificial neural network: A methodology study