Real-time forecasting of intersection turning movements is a critical requirement for predictive traffic signal control at signalized intersections. Several traffic flow forecasting models have been developed in the past; however, most of them have focused only on the road segment-level traffic instead of the turning movement flow (TMF) at intersections. Therefore, in this paper, we propose a new TMF forecasting model, which consists of a combination of two neural network models, namely the stacked bidirectional long short-term memory and the traditional multi-layer perceptron model; this combination will enable effective learning for both short- and long-term time-varying patterns. Moreover, extensive computational experiments, using two years of turning movement counts at 22 intersections in the city of Milton, Ontario, Canada, explore the performance advantage of the proposed model in comparison with several state-of-the-art base models for forecasting accuracy, robustness, and transferability.
Accurate and timely intersection-level turning movement flow (TMF) prediction is fundamental to achieving real-time signal optimization and traffic control. Traditional traffic flow prediction usually focuses on road segmentation-level prediction based on historical traffic data. However, TMF prediction is more challenging due to the high non-linearity and intrinsic complexity. Moreover, most of the current studies do not consider the effects of external characteristics. This paper proposes a parallel adaptive feature fusion framework (PAFF) for the accurate prediction of intersection-level turning movement flow. This framework consists of a four-parallel network structure with an improved graph convolution model and an adaptive multi-source feature fusion mechanism to learn spatiotemporal correlations of traffic flow features. Fifteen external factors such as weather and geographical information are also included as inputs, and the correlations are studied. Finally, in order to validate the performance of the model, this paper designs experiments to compare the accuracy of different models using real-world data from 67 intersections in the city of Waterloo, Ontario, Canada. The experimental results show that the method is more accurate than other benchmark models. This study also explores the influence of external factors on the prediction results, and findings show that multi-source features have a positive effect on prediction accuracy.
14 January 2026 in Artificial Intelligence for Transportation