<p>Urban traffic flow prediction is vital for mitigating traffic congestion and enhancing the efficiency of road traffic. In actual traffic scenarios, traffic flow is subject to the influence of a multitude of factors, including morning and evening rush hour, emergencies, etc., and its dynamic changes are complex. However, traditional methods and some existing technologies have problems such as poor real-time performance and low prediction accuracy. Therefore, the research innovatively proposes a spatiotemporal graph convolutional network prediction model grounded in variational mode decomposition optimized by an improved whale algorithm. This model extracts multi-scale spatiotemporal features of data through variational mode decomposition, and then optimizes the hyper parameters of the spatiotemporal graph convolutional network using an improved whale algorithm to better predict future urban traffic flow. The experimental results show that the prediction accuracy and stability of the proposed model are significantly higher than other comparative models on different datasets. In the comparison experiment of predicting traffic flow in the PeMSD8 dataset for the next 1&#xa0;h (12 5-min time steps), the growth rate of various evaluation indicators of the model was the slowest. After 1&#xa0;h, the average absolute error was only 16.5 ± 0.8 veh/5-min, the root mean square error was 33.4 ± 1.2 veh/5-min, and the average absolute percentage error was 17.9% ± 0.5%. The introduced model can effectively complete the task of urban traffic flow prediction and provide new technical support for the field of traffic flow prediction, which helps to accurately regulate traffic management and optimize traffic strategies.</p>