Road surface friction prediction using long short-term memory neural network based on historical data

Abstract

Road surface friction significantly impacts traffic safety and mobility. A precise road surface friction prediction model can help to alleviate the influence of inclement road conditions on traffic safety, Level of Service, traffic mobility, fuel efficiency, and sustained economic productivity. Laboratory-based methods were used in most previous studies related to road surface friction prediction model development which are difficult for practical implementations. Moreover, for the existing studies about data-driven method development, the time-series features of road surface friction have not been considered. Thus, to utilize the time-series features of road surface friction for predictive performance improvements, this study employed a Long-Short Term Memory (LSTM) neural network to develop a data-driven road surface friction prediction model. According to the experiment results, the proposed prediction model outperformed the other baseline models in terms of three metrics. The impacts of the number of time-lags, the predicting time interval, and adding other relative variables as training inputs on predictive accuracy were investigated in this research. The findings of this study can support road maintenance strategy development, especially in winter seasons, thus mitigating the impact of inclement road conditions on traffic mobility and safety.