STUMP-Net: a spatiotemporal uncertainty-aware mixture probabilistic network for long-term SST forecasting on the Korean Peninsula

Solar irradiance forecasting is fundamental and essential for commercializing solar energy generation by overcoming output variability. Accurate forecasting depends on historical solar irradiance data, correlations between various meteorological variables (e.g., wind speed, humidity, and cloudiness), and influences between the weather contexts of spatially adjacent regions. However, existing studies have been limited to spatiotemporal analysis of a few variables, which have clear correlations with solar irradiance (e.g., sunshine duration), and do not attempt to establish atmospheric contextual information from a variety of meteorological variables. Therefore, this study proposes a novel solar irradiance forecasting model that represents atmospheric parameters observed from multiple stations as an attributed dynamic network and analyzes temporal changes in the network by extending existing spatio-temporal graph convolutional network (ST-GCN) models. By comparing the proposed model with existing models, we also investigated the contributions of (i) the spatial adjacency of the stations, (ii) temporal changes in the meteorological variables, and (iii) the variety of variables to the forecasting performance. We evaluated the performance of the proposed and existing models by predicting the hourly solar irradiance at observation stations in the Korean Peninsula. The experimental results showed that the three features are synergistic and have correlations that are difficult to establish using single-aspect analysis.

STG-OceanWaveNet: Spatio-temporal geographic information guided ocean wave prediction network

Physics-informed spatio-temporal graph neural networks for evapotranspiration prediction: Case of the Korean Peninsula

Energy consumption forecasting is a crucial and challenging task in intelligent energy systems , particularly with the growing adoption of electric vehicles (EVs). Accurate energy demand forecasting relies on analyzing historical data and integrating appropriate factors. Recently, the application of graph convolutional networks (GCNs) has gained prominence in time series forecasting. In light of these developments, we introduce the Probabilistic Dual-Adaptive Spatio-Temporal Graph Convolutional Network (DAS-GCN). Unlike previous approaches, it synergistically interacts with two novel blocks considering user station switching behavior and correlations: Bi-Spatial Graph Integration Unit (Bi-SGI) leverages the spatial intelligence in GCNs, and Adaptive Probabilistic Graph Integration Module (A-PGI) utilizes user transition behavior to dynamically selects the appropriate Poisson probability distribution for the adjacency matrix . This dual approach enables comprehensive and enhanced data analysis by facilitating the learning of charging network and node connectivity, allowing for accurate forecasting of complex EV charging demands. Extensive experiments were performed on real-world datasets, integrating load duration and user transition patterns into energy load forecasting for a comprehensive energy load study. Experimental results show that DAS-GCN model significantly outperforms traditional methods and recent spatiotemporal forecasting approaches in predictive accuracy for both short- and long-term predictions.

Probabilistic spatio-temporal graph convolutional network for traffic forecasting

Deep learning not only enables end-to-end learning of fault features and adaptive feature extraction, but also deeply uncovers the inherent degradation patterns embedded in the data, demonstrating great potential in the field of remaining useful life prediction for rotating machinery. However, the degradation behavior of aviation bearings exhibits high uncertainty. Existing models often face challenges such as difficulty in extracting degradation features, high computational complexity, and poor interpretability. To address the aforementioned challenges, this paper proposes a lightweight spatiotemporal fusion framework for probabilistic prediction of aviation bearings, named the LSTP-Net, which is referred to as the Lightweight Spatio-Temporal Probabilistic Network. This study designs 3 core modules that work in synergy to construct a high-precision and robust aviation bearing life prediction model. First, a lightweight spatial feature extraction module is designed to efficiently capture local dependencies and spatial patterns from vibration signals. Subsequently, the extracted local degradation features are fed into a lightweight temporal feature extraction module. This module employs global sequence modeling to learn and embed the aforementioned spatial features into temporal information, thereby comprehensively extracting the spatiotemporal features that characterize performance degradation. Finally, the probability assessment module characterizes the uncertain information within the degradation data with interval estimates, quantifying the uncertainty in predictions. The experimental results indicate that, compared with 5 mainstream methods such as Squeeze-Bi-GRU-GPR-Net, the proposed model demonstrates significant advantages in terms of interval quality, and computational efficiency.