Real-Time Traffic Congestion Prediction Considering Safety Factors: A Novel Online Learning Method with Attention Mechanism and Multi-LSTM-Based Integrated Learning

Real-time traffic congestion prediction is a crucial part of proactive traffic management, as it can gain more time for the responses of traffic systems such as route guidance, control, and enforcement. However, traffic congestion prediction often features congestion-related indicators, neglecting the potential connection between traffic safety and congestion. In addition, traffic congestion prediction relies on a large amount of historical data, making it impossible for the model to be updated in real-time. Therefore, this study proposes a novel online learning method with attention mechanism and multi-LSTM-based integrated learning that takes traffic safety factors as features and uses average delay per kilometer as the output. Further, the model is applied to two datasets (each 12-hour) from one expressway in China. The results show that: 1) the proposed method can predict traffic congestion in real-time; 2) the proposed method has better prediction performance than the baseline methods; 3) the proposed method has good adaptability in real-time. The findings of this study can provide new ideas and application cases for the future implementation of proactive traffic management.