Strategies and applications for predicting flow using neural networks: a review

This paper introduces the current state of neural network technology for predicting fluid flow. In particular, we introduce (i) neural networks for spatiotemporal flow field predictions, (ii) neural networks that can learn from a small number of fluid data points, and (iii) aero- and hydrodynamic applications of artificial neural networks. The first topic discusses research on predicting unsteady flow fields using convolutional neural networks and generative adversarial networks. The second topic covers methods to increase the learning performance of neural networks when only a limited amount of data is available due to the high cost of fluid simulations and experiments. In the third topic, examples of applying neural networks in the fields of wind power and meteorology are introduced. This article will present the challenges faced in fluid flow prediction based on neural networks, as well as expectations for the positive changes that future technological advances will bring.