Reconstructing flow fields from sparse measurements using a convolutional autoencoder integrated with an Informer model

This paper explores the use of sparse time-series data from flow systems, acquired through sensors or other means, to predict flow fields using deep learning techniques. This area of research holds substantial scientific significance and practical application value. The time-series data measured from different points typically contain spatial correlation and temporal features, which, when utilized effectively, can contribute to reconstructing flow fields. In this study, a convolutional autoencoder is applied to reduce the dimensionality of the flow field. Subsequently, an Informer neural network and a convolutional neural network are employed to extract low-dimensional representations of the flow field from the measurement data. A specially designed loss function bridges these latent features to establish a mapping between measurement point sequences and flow fields. The hybrid model is validated using data from both numerical simulations and experimental measurements. Results demonstrate that this method effectively predicts velocity and pressure fields from sparse data, showcasing its potential for practical flow field reconstruction tasks.