The FRONTIER project has developed an Autonomous Network Management Engine (ANTME) to improve multistakeholder collaboration for a more efficient traffic management. This is done by providing real-time data, automatic response plans generation for specific types of incidents/events, and suitable arbitration models to find agreements among the different operators. In addition, to enhance the capabilities of the traffic operators for taking better decisions, intelligent services have been incorporated in the backend of ANTME: traffic state predictions and automatic incident detection. This blog post will focus on traffic state predictions developed by Eurecat Technology Center.
The traffic state predictions used in FRONTIER start from gathering the road information (i.e. geometry, capacity, speed limits) and its corresponding traffic data. The traffic data is not always homogeneously distributed on the road, so a first traffic state estimation model should be applied to fill the gaps.
This approach begins by mapping the traffic data, which is usually represented at point level, into segments of fixed length. Then the empty segments are filled using spatial interpolation with the closest available data in each time instant. This process is used for both historical data and real time data.
From the historical data, a Machine Learning model has been trained and fine tunned to predict the traffic in each segment of the selected road. The ANTME provides two types of traffic prediction: short-term and long-term. The short-term predictions take into account the most recent data of the segment to predict the traffic in a short time window of 1 hour. On the contrary, the long-term prediction considers the calendar information and offers 1 day and 1 week traffic prediction based on the historical behavior for each type of day. While the short-term provides useful information for real-time actions, the long-term is thought for planification proposes.
The suitable workflow developed for implementing this prediction module and the use of standards like DATEX II, has allowed its fast extrapolation from the first testing pilot site in Athens to two other different scenarios in Oxfordshire and Antwerp.
Finally, the long-term prediction module is also used to evaluate the traffic state under different circumstances. An auxiliar module integrated in the ANTME allows to check how an action taken by an operator affected on the road.
Basically, after an action is applied, the corresponding authority can obtain information of the % of variation of the traffic flow, speed or occupancy in the affected area based on the real-time data. This provides to the authority useful information about the effectiveness of the response plan. Also, this tool provides information about the normal traffic state at a certain moment, mainly comparing the real-time data with the normal traffic state predicted at the same point for the same type of day at the same time instant.