Context based spatial–temporal graph convolutional networks for traffic prediction

Knowledge-Based Systems, Год журнала: 2025, Номер unknown, С. 112933 - 112933

Опубликована: Янв. 1, 2025

Язык: Английский

---

Shkd: A framework for traffic prediction based on sub-hypergraph and knowledge distillation

Knowledge-Based Systems, Год журнала: 2025, Номер unknown, С. 113163 - 113163

Опубликована: Фев. 1, 2025

Язык: Английский

---

Missing Traffic Data Imputation based on Tensor Completion and Graph Network Fusion

Transportation Research Record Journal of the Transportation Research Board, Год журнала: 2025, Номер unknown

Опубликована: Апрель 30, 2025

During traffic data acquisition, missing often arise owing to equipment failures and network disruptions. Despite extensive research on imputation, two primary limitations persist: First, existing methods struggle fully integrate the spatiotemporal correlations low-rank structures inherent in data. Second, current has mostly focused completely at random (MCAR), with limited attention other patterns. We propose an innovative method, tensor completion graph fusion (TCGNF), address these challenges for imputation. This method initially utilizes preliminary imputation of Subsequently, it constructs road by leveraging Pearson correlation coefficient from historical physical distances between detectors. The then uses sampling aggregation (GraphSAGE) extract features networks fuse them. Finally, are trained generative adversarial (GANs) accurate Extensive experiments were conducted publicly accessible datasets validate efficacy TCGNF model. outcomes indicate that model demonstrates superior generalization capabilities, significantly outperforming state-of-the-art models concerning overall performance.

Язык: Английский

---

Optimal Graph Information Fused Graph Attention Network for Traffic Flow Forecasting

Journal of Advanced Transportation, Год журнала: 2025, Номер 2025(1)

Опубликована: Янв. 1, 2025

To manage and make decisions about intelligent transportation systems more efficiently, accurate traffic flow forecasting is necessary. Traffic has complex spatial correlation time dependence. Most current research models are based on a predefined graph structure with priori knowledge for prediction, which cannot well extract the hidden relationships in data. In this paper, we propose Optimal Graph Information Fused Attention Network (OGIF‐GAT). Specifically, learn actual connections between nodes through multigraph feature fusion structure. Next, design new attention network (GAT), improves problem of ignoring edge features traditional GAT model considers their when estimating each neighboring node pair: effect that distance factor correlation. addition, use temporal hybrid transformer (THT) to dependencies. Extensive experiments four public datasets (PeMS04, PeMS08, PeMS‐BAY, METR‐LA) demonstrate our achieves optimal level prediction accuracy all them shown have strong generalization ability. Compared STSGCN, mean absolute error (MAE) decreases by 7.9%, 10.3%, 33.2%, 19.6%, respectively.

Язык: Английский

---

Multi-step control method of traffic flow data quality based on spatiotemporal similarity at video frame rate

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 156, С. 111153 - 111153

Опубликована: Май 28, 2025

Язык: Английский

---

A hybrid model for missing traffic flow data imputation based on clustering and attention mechanism optimizing LSTM and AdaBoost

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Ноя. 2, 2024

Reliable traffic flow data is not only crucial for management and planning, but also the foundation many intelligent applications. However, phenomenon of missing often occurs, so we propose an imputation model to overcome randomness instability bands flow. First, k-means clustering used classify road segments with belonging same pattern into a group utilize spatial characteristics roads fully. Then, LSTM networks optimized attention mechanism are as base learner extract temporal dependence Finally, AdaBoost algorithm integrate all LSTM-attention reinforced impute data. To validate effectiveness proposed model, use PeMS dataset validation, rate from 10 60% under three modes, multiple baseline models comparison, which confirms that our improves stability accuracy imputing different scenarios.

Язык: Английский

---

CECS-CLIP: Fusing Domain Knowledge for Rare Wildlife Detection Model

Animals, Год журнала: 2024, Номер 14(19), С. 2909 - 2909

Опубликована: Окт. 9, 2024

Accurate and efficient wildlife monitoring is essential for conservation efforts. Traditional image-based methods often struggle to detect small, occluded, or camouflaged animals due the challenges posed by complex natural environments. To overcome these limitations, an innovative multimodal target detection framework proposed in this study, which integrates textual information from animal knowledge base as supplementary features enhance performance. First, a concept enhancement module was developed, employing cross-attention mechanism fuse based on correlation between image features, thereby obtaining enhanced features. Secondly, feature normalization amplifying cosine similarity introducing learnable parameters continuously weight transform further enhancing their expressive power space. Rigorous experimental validation specialized dataset provided research team at Northwest A&F University demonstrates that our model achieved 0.3% improvement precision over single-modal methods. Compared existing algorithms, least 25% AP excelled detecting small targets of certain species, significantly surpassing benchmarks. This study offers integrating rare endangered wildlife, providing strong evidence new perspectives field.

Язык: Английский

---

A tensor decomposition method based on embedded geographic meta-knowledge for urban traffic flow imputation

International Journal of Geographical Information Science, Год журнала: 2024, Номер 39(4), С. 788 - 816

Опубликована: Дек. 2, 2024

Accurate and reliable traffic flow data are essential for intelligent transportation systems; however, limitations arising from hardware communication costs often lead to missing data. Tensor decomposition is widely used address these issues. However, existing imputation methods employ a fixed geographic feature similarity matrix constrain the tensor process, which fails accurately capture spatial heterogeneity of flows, thus limiting accuracy robustness. This study proposes method embedded with meta-knowledge (Meta-TD) determine flows. The key innovation establishing dynamic relationship between then using flows process. Experimental results based on real urban demonstrated superiority Meta-TD over fifteen baseline models under random, block, long time-series patterns, achieving reductions in MAE, RMSE, MAPE 6.97–97.05%, 3.33–94.68%, 0.72–90.89%, respectively. Notably, maintained high sudden changes states, evidencing its robustness varying rates distribution patterns. adaptability makes it highly suitable complex environments.

Язык: Английский

---

Decomposing Spatio-Temporal Heterogeneity: Matrix-Informed Ensemble Learning for Interpretable Prediction

Knowledge-Based Systems, Год журнала: 2024, Номер unknown, С. 112906 - 112906

Опубликована: Дек. 1, 2024

Язык: Английский

---