Graph neural networks for multivariate time series regression with application to seismic data

Abstract

Machine learning, with its advances in deep learning has shown great potential in analyzing time series. In many scenarios, however, additional information that can potentially improve the predictions is available. This is crucial for data that arise from e. g., sensor networks that contain information about sensor locations. Then, such spatial information can be exploited by modeling it via graph structures, along with the sequential (time series) information. Recent advances in adapting deep learning to graphs have shown potential in various tasks. However, these methods have not been adapted for time series tasks to a great extent. Most attempts have essentially consolidated around time series forecasting with small sequence lengths. Generally, these architectures are not well suited for regression or classification tasks where the value to be predicted is not strictly depending on the most recent values, but rather on the whole length of the time series. We propose TISER-GCN, a novel graph neural network architecture for processing, in particular, these long time series in a multivariate regression task. Our proposed model is tested on two seismic datasets containing earthquake waveforms, where the goal is to predict maximum intensity measurements of ground shaking at each seismic station. Our findings demonstrate promising results of our approach—with an average MSE reduction of 16.3%—compared to the best performing baselines. In addition, our approach matches the baseline scores by needing only half the input size. The results are discussed in depth with an additional ablation study.

Prediction of the SYM‐H Index Using a Bayesian Deep Learning Method With Uncertainty Quantification

“…It can be seen from Table 6 that SYMHnet outperforms its three variants. SYMHnet‐B is the second best among the four models, implying that a GNN is effective in solving time series regression problems (Bloemheuvel et al., 2022). SYHMnet‐G, which contains a BiLSTM network but no GNN, does not perform well.…”

Deep Learning for Time Series Classification and Extrinsic Regression: A Current Survey

“…GCNs are most commonly used with electroencephalogram (EEG) data where the location of EEG electrodes on our head can be represented naturally using a graph, instead of treating the EEG signal as a multivariate time series. Some of these applications are epilepsy detection [45], seizure detection [46] and sleep classification [47] Besides EEG, GCNs have also been applied to engineering applications such as machine fault diagnosis [48], slope deformation prediction [49] and seismic activity prediction [50]. Applications that use GCN to model their time series, such as the above, often require the time series to have multiple dimensions that makes sense spatially, such as EEG and sensors placed in the ground.…”

GRAPES: Earthquake Early Warning by Passing Seismic Vectors Through the Grapevine

“…Here, we introduce a DL model to continuously predict ground motions across a seismic network. We build upon previous work using Graph Neural Networks (GNN) (van den Ende & Ampuero, 2020; Bloemheuvel et al., 2022) and transformer models (Münchmeyer et al., 2021a, 2021b) for EEW. Our algorithm, GRAph Prediction of Earthquake Shaking (GRAPES; Clements (2023b)), predicts future earthquake shaking using the previous 4‐s of acceleration waveforms across a seismic network (Figure 1).…”