Jul 2026· 2026 5th International Conference on Energy and Electrical Power Systems (ICEEPS)· pp. 28-34· 0 citations· 15 references
Abstract
The rapid proliferation of distributed energy resources and the increasing complexity of modern distribution networks have imposed unprecedented challenges on power supply service command systems. Traditional approaches suffer from data fragmentation across heterogeneous business systems, shallow semantic understanding of customer complaints, and delayed outage perception. This paper proposes an integrated intelligent framework that addresses these challenges through three synergistic modules: (1) a lightweight multi-source heterogeneous data fusion engine that constructs a multi-level distribution network Knowledge Graph (KG) spanning User-Meter Box-Branch-Transformer Area-Feeder hierarchies; (2) a fine-tuned Bidirectional Encoder Representations from Transformers (BERT)-based natural language processing module for automatic tri-level (Red/Orange/Yellow) sensitivity classification of power service work orders; and (3) a Spatio-Temporal Graph Convolutional Network (ST-GCN) that leverages the KG topology as spatial adjacency for proactive outage detection at the meter-box level. Experimental results on simulated datasets demonstrate that the proposed framework achieves 94.7% F1-score in work order classification and reduces outage detection latency from tens-of-minutes level to minute-level, with 91.3% F1-score in proactive outage detection. The framework provides a viable pathway toward data-intelligent transformation of power supply service command operations.
Increasing penetration of renewable energy in distribution networks severely disrupts traditional source-load aggregation, particularly under complex operating conditions. This paper proposes a condition aware dynamic source-load aggregation(CA-DSSA,), validated on the IEEE 33-bus distribution system across seven distinct operating conditions. The core innovation is a differentiable soft graph partitioning layer (SGPL) whose zone boundaries are continuously generated by an operating-condition encoder, enabling joint end-to-end optimization of partitioning and feature learning without any predefined zoning rules. A relational graph attention network (R-GAT) captures heterogeneous bus and branch semantics, while a dilated temporal convolutional network (D-TCN) models multi-scale temporal dynamics. Comparative experiments against DCRNN and STGAT demonstrate that CA-DSSA reduces Mean Absolute Percentage Error (MAPE), achieves a prediction interval coverage probability (PICP) above 95%.
Qinghe Sun, Ai-Hua Zhou, Min Xu et al.· International Conference on...· 0 citations
Maintaining IoT network reliability in mixed-mobility megacities like Ho Chi Minh City (HCMC) presents a formidable challenge due to high stochasticity and complex, non-Euclidean traffic topologies. Traditional monitoring systems predominantly treat sensors as isolated spatial points, frequently failing to differentiate between legitimate traffic congestion and latent cascading sensor anomalies. To address this gap, this paper proposes the Graph-Theoretic Spatiotemporal Transformer (GTST). The core innovation of the GTST framework is the integration of a spectral positional encoding scheme derived from the Laplacian eigenvectors of the transport graph, capturing the network's underlying functional structure. Evaluated using a high-fidelity synthetic dataset comprising 100,000 transactions across 28 urban nodes under extreme stochastic noise, empirical results demonstrate the GTST model prioritizes the mitigation of critical false negatives. It achieves a peak fault recall of 0.96, matching computationally heavy deep-learning baselines with a vastly lower computational footprint. Crucially, the GTST framework accomplishes this with an ultra-low inference time of 0.0061ms per transaction, proving its superior viability and efficiency for real-time edge deployment in noisy urban zones.
Thai Ky Trung Pham, Deshinta Arrova Dewi· International Conference on...· 0 citations
Experiments demonstrate that HG-FL achieves superior performance compared to centralized, flat federated, and graph-only baselines, and highlight the framework’s scalability, robustness, and effectiveness in achieving cross-domain generalization and service-level assurance within privacy-preserving ITS environments.
Mohammed Saad Javeed, Jannatul Maua, Md Shafiq Ullah et al.· IEEE Open Journal of Intelli...· 0 citations
Accurate network traffic forecasting is fundamental to Quality of Service enforcement, proactive congestion control, and dynamic resource allocation in modern backbone and software-defined networks. However, existing approaches often lack adaptability to non-stationary traffic patterns and fail to provide a consistent comparative evaluation across diverse models under unified experimental conditions. This paper presents an adaptive, data-driven framework that integrates Bidirectional LSTM (Bi-LSTM), LSTM, Gated Recurrent Units (GRU), Random Forest, XGBoost, Support Vector Regression (SVR), and classical ARIMA regressors for short- and medium-term traffic forecasting. The proposed architecture couples multi-scale temporal feature extraction with a feedback-driven online retraining loop, enabling continuous adaptation to distributional shifts. Extensive experiments are conducted on two publicly available datasets CAIDA Equinix backbone traces and the MAWI traffic archive comprising over 72 hours of flow-level measurements at one-minute resolution. Bi-LSTM achieves the lowest RMSE of 0.0287 Gbps and the highest $R^{2}=0.9714$, outperforming ARIMA by $\mathbf{7 5 . 2 \%}$ and vanilla LSTM by $\mathbf{8 . 0 \%}$. All results are confirmed via paired Diebold-Mariano (DM) tests and Student’s t-tests $(p \lt 0.01)$. System inference latency of 2.3 ms per batch satisfies real-time SDN control-plane requirements. Code and preprocessing scripts will be made publicly available to ensure full reproducibility.
E. Chithra, G. C. Bharathi, Sankara Rao Allada et al.· International Conference on...· 0 citations
In response to the problems such as high coupling degree of multi-source heterogeneous data, complex spatiotemporal correlations, difficulty in depicting the evolution mechanism of abnormal emissions, and the discreteness of control response links in the ecological governance system under the low-carbon goal, an intelligent control platform architecture for ecological governance modernization is proposed. The platform adopts edge-cloud collaboration technology, builds a monitoring unit graph model, a unified encoding mechanism for multi-source data, and a low-carbon constraint mapping method, to achieve standardized access, spatiotemporal alignment, and fusion processing of sensing monitoring data, emission data, meteorological data, and equipment operation data. At the model layer, an ecological state propagation and estimation model is established based on the spatiotemporal graph network, multi-modal feature fusion and abnormal scoring functions are used to identify abnormal emissions, and multi-objective optimization algorithms are used to complete the scheduling of governance tasks under low-carbon constraints. The platform further combines the stream-batch integrated processing link and the model service orchestration mechanism to form a closed-loop technical process of state perception, risk warning, scheduling solution, and feedback update, which can provide support for the intelligent, low-carbon, and engineering implementation of ecological governance.
Yan Liang· International Conference on...· 0 citations
Network Digital Twins (NDTs) enable proactive network management and optimization by predicting system behavior before control actions are applied to live infrastructures, supporting critical operations in Internet Service Provider (ISP) networks and wide-area networks (WANs). However, to anchor the superior performance NDTs promise to provide, key enabler techniques are required. Given that mobile networks are modeled as graphs, graph-based architectures such as graph neural networks (GNNs) have shown promising performance in modeling network behavior. This work proposes a novel Hybrid Spatial-Temporal Graph Neural Network (HSTGNN) architecture. Unlike single-branch GNN approaches, we propose a multi-scale design that combines three complementary message-passing paradigms: local neighborhood aggregation, spectral filtering, and learnable attention-based weighting. When benchmarked against other approaches, the proposed HSTGNN achieved superior performance delivering a coefficient of determination score of approximately 0.8816, 17.5\% better than the best baseline ChebNet. Furthermore, HSTGNN achieved the lowest Mean Absolute Error (MAE) of 0.0300, and Root Mean Squared Error (RMSE) of 0.0458, significantly outperforming baseline frameworks and certifying the proposed framework's capability in enabling NDTs.
J. Sengendo, Fabrizio Granelli· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
