The two-stage prediction method for traffic spillover dissipation at short-distance intersections based on Bi-LSTM

Area of Science:

• Transportation Engineering
• Machine Learning Applications
• Traffic Flow Theory

Background:

• Traffic spillover at short-distance intersections is a common yet often overlooked issue.
• Predicting the dissipation of such traffic congestion is crucial for optimizing intersection capacity.
• Existing methods may not adequately address the complexities of short-distance intersection spillover.

Purpose of the Study:

• To develop and validate a machine learning model for predicting traffic spillover dissipation at short-distance intersections.
• To improve the interpretability and accuracy of traffic spillover prediction.
• To enable targeted signal control strategies for mitigating traffic congestion.

Main Methods:

• Proposed conditions for identifying and determining traffic spillover dissipation based on traffic wave theory.
• Utilized VISSIM 11 for simulating traffic operations and collecting data.
• Constructed a two-stage prediction model using a Bidirectional Long Short-Term Memory (Bi-LSTM) network, incorporating queue length prediction in the first stage and spillover dissipation prediction in the second.

Main Results:

• The first stage of the Bi-LSTM model achieved 93.4% accuracy in predicting queue length, outperforming traditional models.
• The second stage achieved 92.88% accuracy in traffic spillover identification and 90.72% in dissipation state prediction.
• The two-stage prediction method demonstrated superior performance compared to single-stage models.

Conclusions:

• The proposed two-stage Bi-LSTM model effectively predicts traffic spillover and its dissipation at short-distance intersections.
• The method validates the feasibility and effectiveness of using machine learning for traffic spillover management.
• Accurate prediction facilitates targeted signal control, enhancing the operational capacity of short-distance intersections.