Spatial-temporal feature alignment for traffic flow prediction using light-gated recurrent unit and graph convolution network

In the evolving realm of urban planning, Intelligent Transportation Systems (ITS) become pivotal, necessitating refined methodologies for accurate traffic flow prediction. Current models tackle the need for advanced methods to address the inherent complexity, nonlinearity, and randomness of traffic flows by applying advanced deep-learning models. Recent traffic flow prediction depends on Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) to extract and learn temporal and spatial data. However, due to the highly complex patterns and correlations in traffic data, standard RNN and CNN layers struggle to learn easily, affecting the quality of the learning process. Additionally, many deep neural network models fail to extract the complex graph structures of traffic networks, missing critical high-order correlations among nodes. This research introduces ST Features Alignment with Light Gated Recurrent Unit and Graph Convolution Network (STF-GGRU) for traffic flow prediction. This model contains five effective modules: input module, feature extraction, spatial, temporal, and prediction module. This model does not regard sensor data as uniform or interchangeable. Rather, it acknowledges and capitalizes on the distinct attributes of each sensor's data, resulting in a more refined and detailed analysis. The model's efficiency is rigorously evaluated using metrics like Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). In the PeMSD4 dataset, it achieves the lowest RMSE of 27.18, MAE of 16.2, and MAPE of 9.67%, outperforming established models. The model excels in both short-term and long-term prediction scenarios, consistently achieving the lowest error values across different time horizons. In the PeMSD4 dataset, RMSE values range from 28.18 to 32.23, and in the PeMSD8 dataset, from 12.01 to 16.3. The model consistently achieves the lowest MAE and MAPE values, effectively capturing intricate Spatial-Temporal (ST) traffic patterns. The model shows varying efficiency in handling peak and non-peak traffic conditions, indicating the need for tailored algorithms for each sensor location.

Text FK 2024 68 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (7MB) Text FK 2024 68.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (8MB)

Actions (login required)

View Item