Liang Zijun, Wang Ruihan, Zhan Xuejuan, Hu Tingting, Li Xiaoyan, Li Yuqi
School of Urban Construction and Transportation, Hefei University, Hefei, 230601, China.
Anhui Province Transportation Big Data Analysis and Application Engineering Laboratory, Hefei, 230601, China.
Sci Rep. 2025 Jul 1;15(1):22249. doi: 10.1038/s41598-025-06969-9.
In response to the common issue of traffic spillover at short-distance intersections, which tends to be overlooked due to its potential for self-dissipation, this paper conducts a study on predicting traffic spillover dissipation at short-distance intersections using the machine learning models. First, conditions for identifying traffic spillover and determining its dissipation at short-distance intersections are proposed based on traffic wave theory, and the traffic operation and data collection scenario for a short-distance intersection is built using traffic simulation software, VISSIM 11. Next, data sets required as inputs for the prediction model are collected and generated, based on queue length of stranded vehicles and the dissipation state of traffic spillover, to improve model interpretability. Finally, a two-stage prediction model for traffic spillover dissipation is constructed using a Bi-LSTM model. The first stage of the model predicts the queue length of stranded vehicles on the road segment, which is then used as feature data for the second stage to predict the spillover dissipation state between short-distance intersections. The results show that the model's prediction of the queue length of stranded vehicles in the first stage outperforms the DT, CNN, RF, LSTM, and GRU models, with a prediction accuracy of 93.4%, which verifies the feasibility of selecting the Bi-LSTM model in this paper. The model's prediction of traffic spillover dissipation state in the second stage outperforms single-stage prediction models, with the model achieving an accuracy of 92.88% in traffic spillover identification and 90.72% in traffic spillover dissipation state prediction. This validates the effectiveness of the two-stage prediction method proposed in this paper and is conducive to further improving the model's prediction accuracy. The proposed method can accurately predict traffic spillover and their dissipation at short-distance intersections, enabling the targeted selection of signal control strategies to address traffic spillover issues, thereby effectively improving the capacity of short-distance intersections.
针对短距离交叉口交通溢流这一因具有自我消散可能性而往往被忽视的常见问题，本文开展了利用机器学习模型预测短距离交叉口交通溢流消散的研究。首先，基于交通波理论提出短距离交叉口交通溢流的识别条件及其消散判定条件，并利用交通仿真软件VISSIM 11构建短距离交叉口的交通运行及数据采集场景。其次，基于滞留车辆排队长度和交通溢流消散状态，收集并生成预测模型所需的输入数据集，以提高模型的可解释性。最后，利用双向长短期记忆网络（Bi-LSTM）模型构建交通溢流消散的两阶段预测模型。模型的第一阶段预测路段上滞留车辆的排队长度，该排队长度随后用作第二阶段的特征数据，以预测短距离交叉口之间的溢流消散状态。结果表明，模型第一阶段对滞留车辆排队长度的预测优于决策树（DT）、卷积神经网络（CNN）、随机森林（RF）、长短期记忆网络（LSTM）和门控循环单元（GRU）模型，预测准确率达93.4%，验证了本文选用Bi-LSTM模型的可行性。模型第二阶段对交通溢流消散状态的预测优于单阶段预测模型，在交通溢流识别方面准确率达92.88%，在交通溢流消散状态预测方面准确率达90.72%。这验证了本文提出的两阶段预测方法的有效性，有助于进一步提高模型的预测精度。所提方法能够准确预测短距离交叉口的交通溢流及其消散情况，有助于针对性地选择信号控制策略来解决交通溢流问题，从而有效提高短距离交叉口的通行能力。
