Road Surface Condition Prediction using Long Short-Term Memory Neural Network based on Historical Data

Road surface condition significantly impacts traffic safety and mobility. A precise road surface condition prediction model can help to improve traffic safety, Level of Service, traffic mobility, fuel efficiency and sustained economic productivity. Most related previous studies are laboratory-based methods which are difficult for practical implementation. Only a few of them developed data-driven road surface temperature prediction methods. However, the impact of time-series have not been considered, and road surface conditions are not determined only by road surface temperature. This study deployed a long-short term memory neural network to develop a data-driven road surface condition prediction model based on historical data which was collected by a RCM-411 sensor in Finland. The proposed prediction model outperformed the other conventional models, Support Vector Machine (SVR), Random Forest (RF) and Feed-forward NN in terms of mean absolute error (MAE), mean squared error (MSE) and mean absolute percentage error (MAPE). The influence of (1) the number of time lags, (2) the time interval between time steps and (3) additional features, such as water, pavement temperature, etc. in predictive performance were also analyzed. According to analysis results, the proposed prediction model got the highest level of prediction accuracy with 7-time lags. The predictive performance dropped gradually as the time interval between time steps grew larger. Adding additional features, like water thickness on the road and road surface temperature did not improve the prediction accuracy of the proposed model, which contributes to the elevated complexity of the model. Findings of this study can help improve traffic mobility and traffic safety by predicting road surface condition more accurately, especially in winter. Future work includes an improved LSTM model by accommodating additional features and flexible time interval between time steps.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.
• Corporate Authors:

  Transportation Research Board

  ,    
• Authors:
  • Pu, Ziyuan
  • Liu, Chenglong
  • Wang, Yinhai
  • Shi, Xianming
  • Zhang, Chao
• Conference:
  • Transportation Research Board 98th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2019-1-13 to 2019-1-17
• Date: 2019

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 6p

Subject/Index Terms

• TRT Terms: Data mining; Mathematical prediction; Neural networks; Surface course (Pavements); Texture
• Subject Areas: Data and Information Technology; Highways; Maintenance and Preservation; Pavements;

Filing Info

• Accession Number: 01697454
• Record Type: Publication
• Report/Paper Numbers: 19-03118
• Files: TRIS, TRB, ATRI
• Created Date: Mar 1 2019 3:50PM