Researchers
Zhengyuan Zhu
Mark Kaiser
About the research
Winter highway maintenance is an annual multi-billion dollar operation aimed at improving the safety and mobility of the highway system. To help the winter highway maintenance agencies optimize the usage of resources, it is important to develop a performance measurement system that can evaluate how well maintenance activities have been performed. In Iowa Highway Research Board Project TR-491, researchers developed a performance measure based on average vehicle speed, which takes into account severity of the storm. The model uses six categorical variables to define a storm and compute the acceptable traffic speed drop.
A previous Iowa Department of Transportation (DOT) agreement developed a sequential Bayesian dynamic model based on the model in TR-491, which is capable of predicting the acceptable drops during the storm, and allow uncertainty in input variables (sensor measurements) to propagate into uncertainty in speed reduction. One limitation of the sequential Bayesian model is that its uncertainty measure does not account for model uncertainty and the uncertainty in human-weather interaction. The model in TR-491 is based on survey of expert opinion, and its uncertainty is not considered in the original development and our follow up work. The uncertainty in human behavior under different weather conditions is also not considered due to lack of time.
The Iowa DOT is interested in refining this sequential Bayesian model to produce more accurate real-time prediction of traffic speed drops with better uncertainty measures so that it can be used to evaluate the performance of snow/ice removal efforts and the effectiveness of different snow removal methods. Ultimately, the DOT is interested in using this model to help managers reallocate resources to optimize objective functions such as minimizing the total costs or the speed drops. The DOT is interested in developing a dynamic model capable of predicting in real-time acceptable drops in traffic speed on highways during major weather events with realistic uncertainty measures. The primary usage of such model is to evaluate the performance of highway winter maintenance operations and optimize resource allocation.
The researchers developed a model to relate weather variables to traffic flow changes at a local level. Weather station data and maintenance crew reports were used to develop an empirical adaptive stochastic model using a Bayesian formulation. Data from early time segments provide a prior quantification of the expected effects of weather variables on traffic speed over subsequent time segments. Data in the next time segment are then used to adjust these quantifications to reflect observed traffic speeds during that period. Thus, rather than explicitly determining numerous temporally dependent interactions, the main effects associated with key factors are allowed to undergo small shifts over time to fit the data. The model incorporates an autoregressive error structure to reflect temporal dependencies in observations that occur at reasonably high frequencies.