Signalized arterial origin-destination flow estimation using flawed vehicle trajectories: A self-supervised learning approach without ground truth

Abstract

For alleviating arterial congestion, most control strategies provide progression for through and turning traffic. A prerequisite input is the arterial origin–destination (OD) flow pattern, which can be estimated based on connected vehicle (CV) trajectories. However, the existing estimation methods require the ground-truth historical OD flow, which is difficult to obtain. To address this issue, this paper develops a method to estimate real-time OD flow along a signalized arterial without ground truth. A model based on the Generative Adversarial Network (GAN) network is proposed, which incorporates long short-term memory (LSTM), attention mechanism, and convolutional neural network (CNN) to capture the temporal and spatial correlations between OD flow patterns. This model is trained with the proposed self-supervised without historical OD flow. The proposed model is extensively tested based on a realistic signalized arterial, and the results indicate sufficient accuracy for progression control.