Moving vehicles interact with IoT devices deployed in cities and establish social relationships to provide proactive and intelligent services for smart cities. For example, big-data-driven accurate and timely traffic speed prediction systems play an important role in empowering Intelligent Transportation Systems (ITS) in smart cities. The reason is that it is the foundation of modern traffic management and traffic control. Most of the existing advanced traffic speed prediction models are Spatial-Temporal hybrid models. They improve the predicting accuracy by leveraging Graph Convolutional Network (GCN) and Recurrent Neural Network (RNN) to extract spatial and temporal features from the traffic speed data, respectively. However, these models have complex structures and high computational costs. To improve the accuracy of prediction and reduce the cost of model training, we propose a hybrid model, Spatial-Temporal Gated Graph Attention network (ST-GGAN), based on Graph Attention mechanism (GAT) and Gated Recurrent Unit (GRU). Such a method has a simpler structure, lower computational costs, and higher predicting accuracy. The experimental results show that our model's performance is better than the existing advanced models on a real-world dataset.
