State estimation (SE) plays a crucial role in the monitoring of smart grids. However, as an important cyber-physical system (CPS), smart grid is currently facing various anomalies that pose a serious threat to the accuracy of the state estimation process. This paper proposes a new framework named Grid2Vec (Grid To Vector) for detecting abnormal types in the power system based on graph representation learning. The proposed method involves constructing a statement sequence using a random walk of nodes on the power system graph and training a classification neural network by optimizing the node vector. The trained network can then accurately determine the type of anomaly present in the power system. The proposed Grid2Vec model is scalable to large-scale graph networks and can improve learning efficiency through parallel walking. The effectiveness of the proposed model was verified through numerical experiments conducted on real power systems. The results indicate that the Grid2Vec model exhibits high precision and efficiency in detecting several fault types amongst diverse noise and assault scenarios. It achieves an average detection accuracy of 100.00%, 99.97%, 99.49%, and 99.21%. The level of precision demonstrated is notably superior to the relative error attained by current methodologies.

Because of the high permeability of renewable energy, achieving the real-time state of the distribution system, including the topology, essential parameters and operation condition, is now crucial to its delicate analysis and determination. In the active distribution system, the frequently changing topology of the network impedes the accurate identification of the line parameters, which is sensitive to the natural disaster factors. Inspired by the latest progress of deep learning, we propose a dynamic graph fusion network (DGFN) architecture to tackle the topology and line parameter joint identification problem. The DGFN model is based on the graph deep learning method, which constructs a new topology according to the changes of node features through the radius graph method, and forms the corresponding adjacency matrix, which helps to improve the line parameters identification accuracy. The architecture can be used for any distribution system, only the adjacency matrix and node feature data need to be inputted. Masses experiments show that the effectiveness and robustness of the proposed approach under the distribution system topology changes.

Accurate cloud and cloud-shadow segmentation is a crucial step in optical remote sensing image preprocessing, playing a significant role in subsequent applications such as land-cover classification and change detection. However, the complexity of cloud/shadow shapes and noise interference (e.g., snow and ice, buildings, complex backgrounds, and atmospheric optics) make this task challenging. Although existing deep learning methods have achieved remarkable results in cloud segmentation tasks, a better balance between computational efficiency and segmentation accuracy is still needed. Traditional deep learning models have good detail and generalization capabilities due to their local feature extraction ability and spatial invariance, but they are relatively weak in processing global context information, leading to false positives and false negatives in complex scenarios. Encoders based on state space models (such as VMamba) can effectively capture global context through long-range dependency modeling, but there is still room for optimization in computational efficiency. Additionally, complex attention mechanisms (such as CBAM) can improve feature representation capability, but the large number of parameters limits the deployment efficiency of models. This paper conducts a systematic architectural exploration of the MCloudX cloud segmentation network, seeking a balance between efficiency and accuracy from three directions: backbone network modernization, encoder efficiency optimization, and attention mechanism lightweighting. Through comprehensive ablation experiments on SPARCS and L8-Biome datasets, we systematically evaluate the independent and synergistic effects of each component and validate them on Biome_3 and SPARCS datasets. Experimental results show that the proposed optimization configuration (ResNet50+LocalMamba+ECA-Net) significantly improves computational efficiency while maintaining comparable accuracy to the baseline. We name this optimization configuration LECloud, providing valuable empirical references for future research on efficient remote sensing segmentation architectures.

High-precision wind power prediction improves grid stability and reduces curtailment losses. Existing methods face three limitations: static graphs cannot capture dynamic spatial correlations under weather changes, time series models miss multi-scale temporal features, and frequency-domain analyses lack physical constraints. We propose: (1) a dynamic distance correlation weighted graph that adaptively combines geographic and power correlations for weather–terrain coupling; (2) a spatio-temporal-frequency fusion framework integrating graph networks, bidirectional GRUs, and a patchwise sparse time–frequency module; (3) a turbine power curve-constrained frequency mixer for physical consistency. On the SDWPF dataset, our model achieves MAE reductions of 37.47–43.32% and RMSE reductions of 37.93–42.70% versus baselines, outperforming state-of-the-art methods. The approach demonstrates superior performance in complex spatio-temporal scenarios.

Electric vehicles (EVs) are essential for addressing dynamic path planning and charging scheduling problems in transportation-energy coupled networks. However, conventional traffic assignment models struggle to accurately capture EV charging behaviors due to limited temporal resolution, insufficient power-transportation integration, and inadequate representation of queue dynamics. This article proposes a mixed-integer quadratic programming (MIQP) model based on virtual service nodes (VSNs), which decomposes the charging process into distinct waiting and service phases, thereby enabling fine-grained spatiotemporal capture of EV charging behaviors and their queue dynamics. We model each charging station (CS) as a logical triplet (entry, service, and exit) connected by wait and charge arcs. Within the VSN framework, charging pile states and queue dynamics are characterized using discrete time steps and binary variables with linear constraints, which facilitates precise modeling of operational constraints. Furthermore, a grid-transportation collaborative equilibrium mechanism incorporates time-of-use pricing into the optimization objective, thereby incentivizing peak-valley load balancing. The experimental results demonstrate that the VSN strategy reduces average and maximum waiting times by 91.0% and 84.5%, respectively, while enhancing CS utilization balance and smoothing grid load profiles, ultimately improving the overall service capacity of the coupled network.

With the widespread adoption of Electric Vehicles (EVs) and the deep integration of transportation and energy systems, EVs and their charging infrastructure have become key enablers for the electrification of transportation and the decarbonization of the power grid. However, the capacity limitations of the distribution network and the mismatch between charging demand and the spatiotemporal availability of charging facilities have become increasingly significant. This work proposes a novel traffic-energy integration framework by constructing a Coupled Meteorology-Transportation-Distribution Grid Hierarchical Graph Model (CMTD-HGM) to optimize EV charging scheduling and energy management. Specifically, Graph Neural Networks (GNNs) are utilized to analyze the complex spatiotemporal dependencies between meteorological conditions, road conditions, and energy consumption, incorporating a linearized DistFlow model into the graph framework to accelerate power flow calculations and enable efficient coordination of transportation and power grid flows. Furthermore, this paper quantifies and assesses the dynamic hosting capacity of the distribution network for EVs under extreme weather conditions. Experimental results demonstrate that the CMTD-HGM reduces the energy consumption prediction error (MAE) to 0.17 kWh, and improves node voltage deviations by 6.82% after optimizing the charging station layout, providing theoretical support for EV charging infrastructure planning.