Spatiotemporal Residual Graph Attention Network for Traffic Flow Forecasting

Zhang, Qingyong; Li, Changwu; Su, Fuwen; Li, Yuanzheng

doi:10.1109/jiot.2023.3243122

Cited by 15 publications

(3 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In electric load forecasting, there are some essential factors such as weather activities, humidity, calendar meters, temperature changes, wind speed and etc. to effect on the electricity load behavior [14]. In this section, we introduce the proposed electric load forecasting architecture based on machine learning approach in smart grids.…”

Section: Methodsmentioning

confidence: 99%

Enhanced K-NN with Bayesian Optimization Algorithm for Predicting Energy Efficiency of Smart Grids in IoT

ZHAO,

Yuanchao,

Xiaopeng

et al. 2023

Preprint

View full text Add to dashboard Cite

With the increasing number of end users using electricity in modern cities, smart grids have some critical problems for energy efficiency and managing renewable energy resources. Therefore, electricity load forecasting is an important strategy to avoid power disconnection and power communication damages in smart grids. On the other hand, the Internet of Things (IoT) devices can collect appropriate data from end users with smart metering. In addition, smart devices can check and gather critical data from power stations using smart sensors and actuators to provide the Quality of Service (QoS) factors and energy efficiency of electricity transmission in the smart girds. Due to the huge data transmission, machine learning is useful for evaluating electricity load forecasting and power stability in smart grids. For detecting electric load forecasting and energy consumption factors, this paper develops Bayesian Optimization for K-nearest neighbor (BOKNN) with the hyper-parameters function to detect the electricity load forecasting and estimated power consumption factors in smart grids. The proposed technique optimizes the performance of electricity load forecasting and enhances the structure of the machine learning efficiency with high model accuracy. Two real-data sets as our case studies for Short-Term Electricity Load Forecasting are applied to evaluate the proposed BOKNN as existing experimental testbeds. The simulation results illustrate that the proposed BOKNN provides optimized high accuracy with 88.33% and 98.13% correlation coefficient and minimum Mean Absolute Error (MAE) with 0.04% for existing datasets. The experimental results show that the suggested BOKNN model greatly outperforms the state-of-the-art prediction algorithms.

show abstract

Section: Methodsmentioning

confidence: 99%

Enhanced K-NN with Bayesian Optimization Algorithm for Predicting Energy Efficiency of Smart Grids in IoT

ZHAO,

Yuanchao,

Xiaopeng

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…To enhance the stability of the self-attention mechanism, we incorporate the multi-head attention mechanism [16] for spatial feature extraction. This integration serves to mitigate any potential loss in the spatial information captured by the graph attention network (GAT), ensuring a more robust and reliable performance.…”

Section: Spatiotemporal Synchronous Graph Attention Neural Networkmentioning

confidence: 99%

PDASTSGAT: An STSGAT-Based Multipath Data Scheduling Algorithm

Xue,

Wu,

Han

et al. 2024

Algorithms

View full text Add to dashboard Cite

How to select the transmitting path in MPTCP scheduling is an important but open problem. This paper proposes an intelligent data scheduling algorithm using spatiotemporal synchronous graph attention neural networks to improve MPTCP scheduling. By exploiting the spatiotemporal correlations in the data transmission process and incorporating graph self-attention mechanisms, the algorithm can quickly select the optimal transmission path and ensure fairness among similar links. Through simulations in NS3, the algorithm achieves a throughput gain of 7.9% compared to the PDAA3C algorithm and demonstrates improved packet transmission performance.

show abstract

“…The effect of hyperparameters is weakened [10], and the influence of human factors is reduced [11]. For example, recurrent neural networks (RNNs) [12] and their variant long short-term memory (LSTM) [13], as well as Gated Recurrent Units (GRUs) [14], have achieved great results in performing temporal features extraction [15]. Graph convolutional networks (GCNs) [16], which evolved from convolutional neural networks (CNNs) [17], are commonly used for learning the temporal features of traffic data.…”

Section: Introductionmentioning

confidence: 99%

Gated Recurrent Unit Embedded with Dual Spatial Convolution for Long-Term Traffic Flow Prediction

Zhang,

Zhou,

et al. 2023

IJGI

Self Cite

View full text Add to dashboard Cite

Considering the spatial and temporal correlation of traffic flow data is essential to improve the accuracy of traffic flow prediction. This paper proposes a traffic flow prediction model named Dual Spatial Convolution Gated Recurrent Unit (DSC-GRU). In particular, the GRU is embedded with the DSC unit to enable the model to synchronously capture the spatiotemporal dependence. When considering spatial correlation, current prediction models consider only nearest-neighbor spatial features and ignore or simply overlay global spatial features. The DSC unit models the adjacent spatial dependence by the traditional static graph and the global spatial dependence through a novel dependency graph, which is generated by calculating the correlation between nodes based on the correlation coefficient. More than that, the DSC unit quantifies the different contributions of the adjacent and global spatial correlation with a modified gated mechanism. Experimental results based on two real-world datasets show that the DSC-GRU model can effectively capture the spatiotemporal dependence of traffic data. The prediction precision is better than the baseline and state-of-the-art models.

show abstract

Spatiotemporal Residual Graph Attention Network for Traffic Flow Forecasting

Cited by 15 publications

References 38 publications

Enhanced K-NN with Bayesian Optimization Algorithm for Predicting Energy Efficiency of Smart Grids in IoT

Enhanced K-NN with Bayesian Optimization Algorithm for Predicting Energy Efficiency of Smart Grids in IoT

PDASTSGAT: An STSGAT-Based Multipath Data Scheduling Algorithm

Gated Recurrent Unit Embedded with Dual Spatial Convolution for Long-Term Traffic Flow Prediction

Contact Info

Product

Resources

About