Spatiotemporal Network for Accurate Traffic Flow Prediction in Intelligent Transportation Systems Using Generative AI

Traffic flow prediction plays a critical role in Intelligent Transportation Systems (ITS), enabling better traffic management and trip planning. Accurate prediction not only aids city managers in optimizing traffic operations but also empowers individuals to make informed travel decisions. However, achieving precise traffic flow predictions is challenging due to the complex spatiotemporal dependencies inherent in traffic data. Recent advancements in Generative AI and deep learning have shown promise in addressing these challenges, with convolutional neural networks (CNNs) being commonly employed for extracting spatial and temporal features. Despite their effectiveness, CNNs are limited in fully capturing complex spatiotemporal dependencies and are computationally intensive when stacked to address global dependencies, leading to slower convergence rates. To overcome these limitations, we propose ST-WaveMLP, a globally aware spatiotemporal network model for traffic flow prediction. Leveraging Generative AI based ST-Wave MLP introduces the repeatable ST-WaveBlock structure, which is based on multilayer perceptron’s (MLPs). Each ST-WaveBlock includes two key modules: The Spatiotemporal Global Attention Component (SGAC), which captures global spatial and local temporal dependencies, and the Spatiotemporal Local Attention Component (SLAC), which captures local spatial and global temporal dependencies. This design ensures strong spatiotemporal representation learning, enabling effective modelling with only 2–4 stacked blocks. Extensive experiments on four real-world traffic flow datasets demonstrate that Generative AI based ST-Wave MLP achieves state-of-the-art performance, offering improved convergence and prediction accuracy. Compared to existing methods, Generative AI based ST-Wave MLP improves prediction accuracy by up to 9.57% and enhances convergence speed by up to 30.6%, making it a robust solution for traffic flow prediction in modern ITS empowered by Generative AI. When compared to the previous best models, the proposed model demonstrated training speed gains of 30.6%, 12%, and 28.6% on the TaxiBJ, TaxiJN, and TaxiGY datasets, respectively.