.png){kind=icon}
.png){kind=icon}
Return
Using deep learning methods to optimize freeway traffic flow sensor spacing based on empirical data
.png){kind=icon}
PRE
AI
DOI：10.1080/15472450.2025.2578670.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate traffic state estimation is crucial for intelligent transportation systems. Optimizing sensor spacing on freeways is essential to balance traffic state estimation accuracy against sensor installation and maintenance costs. This study aims to optimize freeway traffic flow sensor spacing using deep learning methods based on empirical data. MobileNetV3, long short-term memory (LSTM), gated recurrent unit (GRU), stacked auto-encoders (SAEs), and transformer are employed to train traffic speed estimation models, and a Staking ensemble strategy is introduced to further reduce estimation errors. Results indicate that the transformer model achieved the best overall performance. Using this model, the mean absolute percentage error (MAPE) and mean absolute error (MAE) of traffic speed estimation were evaluated for different sensor spacings across different freeway section types and traffic conditions, and the recommended sensor spacings for various scenarios were provided. Then a case study incorporating economic feasibility analysis demonstrated that a 1000-m sensor spacing is the most cost-effective configuration for a 20-km basic freeway section. Furthermore, transferability and robustness tests confirmed the model’s generalizability to other freeways and its resilience to random data loss, underscoring its potential for scalable deployment in intelligent transportation systems. Overall, these findings provide practical guidance for freeway operation and management agencies in determining optimal sensor spacing by balancing estimation accuracy and project cost.
Keywords:
deep learning
freeway
sensor spacing optimization
traffic speed estimation
Journal
J
IF:
2.8
Papers:
63
Citations:
0