AUTHOR=Song Ning , Tian Hao , Nie Jie , Geng Haoran , Shi Jinjin , Yuan Yuchen , Wei Zhiqiang TITLE=TSI-SD: A time-sequence-involved space discretization neural network for passive scalar advection in a two-dimensional unsteady flow JOURNAL=Frontiers in Marine Science VOLUME=Volume 10 - 2023 YEAR=2023 URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2023.1132640 DOI=10.3389/fmars.2023.1132640 ISSN=2296-7745 ABSTRACT=Numerical simulation of turbulent flow is a great challenge because it contains extremely complicated variations with a high Reynolds number. Usually, it requires very high resolution grids in order to capture the very fine changes during its physical process to achieve accurate simulation, which will result in a vast number of computations. This issue has been a bottleneck problem for a long period until the deep learning solution is proposed to utilize big scale grids with adaptively adjusted coefficients during the spatial discretization procedure instead of traditional methods that adopt small grids with fixed coefficients so that the computation cost is dramatically reduced and the accuracy is preserved. This breakthrough is considered as a significant improvement in the numerical simulation of turbulent flow. However, these previously proposed deep-learning-based methods always predict the coefficients with only consideration of spatial correlation among grids, which provides relatively limited context and thus cannot describe patterns along the temporal dimension sufficiently, implying that the spatiotemporal correlation of coefficients is not well learned. To address this problem, we propose Time-Sequence-Involved Space Discretization Neural Network(TSI-SD) to extract grid correlations from spatial and temporal views together. This novel deep neural network is transformed from a classic CONV-LSTM backbone with careful modification by adding temporal information into 2D spatial grids along the x-axis and y-axis separately at the first step and then fusing them through a post-fusion neural network. After that, we combine TSI-SD with the finite volume format as an advection solver for passive scalar advection in a 2D turbulent flow. Compared to previous methods that only consider spatial context, our method can achieve higher simulation accuracy, while the computation is also decreased since we find that after adding temporal data, one of the input features, the concentration field is redundant and should be no more adopted during the spatial discretization procedure, which results in a sharp decrease of parameter scale and achieves high efficiency. Comprehensive experiments, including comparison with SOTA methods and sufficient ablation studies, are carried out to verify the accurate and efficient performance and discuss the advantage of the proposed method.