Physics-Coupled Spatio-Temporal Active Learning for Dynamical Systems

Abstract

Spatio-temporal forecasting is of great importance in a wide range of dynamical systems applications, such as earth science, transport planning, etc. These applications rely on accurate predictions of spatio-temporal structured data reflecting real-world phenomena. A stunning characteristic is that the dynamical system is not only driven by some physics laws but also impacted by the localized factor in spatial and temporal regions. One of the major challenges is to infer the underlying causes, which generate the perceived data stream and propagate the involved causal dynamics through the distributed observing units. Another challenge is that the success of machine learning based predictive models requires massive annotated data for model training. However, the acquisition of high-quality annotated data is objectively manual and tedious as it needs a considerable amount of human intervention, making it infeasible in fields that require high levels of expertise. To tackle these challenges, we advocate a spatio-temporal physicscoupled neural networks (ST-PCNN) model to learn the underlying physics of the dynamical system and further couple the learned physics to assist the learning of the recurring dynamics. To deal with data-acquisition constraints, an active learning mechanism with Kriging for actively acquiring the most informative data is proposed for ST-PCNN training in a partially observable environment. Our experiments on both synthetic and real-world datasets exhibit that the proposed ST-PCNN with active learning converges to near optimal accuracy with substantially fewer instances.

A water track laser Doppler velocimeter for use in underwater navigation

“…The ocean current velocity includes a large-scale, slowly-varying, spatially steady component and a small-scale, spatiotemporally local perturbation. For the large-scale component, the preload ocean maps based on the ocean circulation models [26,27] or Artificial Intelligence algorithm [28,29] can provide forecasts for up to 6 d with an average spatial resolution of 3 km and a temporal resolution of 3 h or less. The small-scale current is usually approximated as a Markov process [30,31].…”

The Gulf of Mexico in trouble: Big data solutions to climate change science

“…With the extent of observational data available along the GOM, machine learning is used in predicting the severity of weather events (Ramachandra, 2019). Data-driving machine learning approach is used to predict the Loop Current evolution and the Loop Current ring formation in the Gulf of Mexico (Wang et al, 2019), including a forecast of the sea surface height of the Loop Current System (Zeng et al, 2015;Wang et al, 2021), and a forecast of velocity structures of the Loop Current and its eddies (Huang et al, 2021;Muhamed Ali et al, 2021;Huang et al, 2022b). All ocean basins are experiencing sea level rise and warming due to climate change and global warming, so predicting and understanding the sea-level rise is also done using the machine learning approaches (Roshni et al, 2019;Morovati et al, 2021;Nieves et al, 2021;Tur et al, 2021).…”