Methods to generate an expected turning traffic flows matrix for road junction analysis

In dependence upon a given geometric configuration, an actual or forecasted number of vehicles arriving at an intersection can turn left or right, otherwise proceed straight through it. This article can be placed in between a research working paper, and a kind of informational brief report. Indeed, it deals with forecasting techniques for estimation of intersection turning movements. Such issue is crucial, both in network planning and in traffic engineering, while its applications span from traffic safety and environmental impacts, to signal timing, roundabout design and setting of traffic control strategies. The number of vehicles making each movement on an existing intersection can be manually collected, especially when operational analyses are undertaken. Nevertheless, when the intersection is at the planning or design stage, an estimation process is required. In its first part, this paper provides a brief literature review of some of theoretical and practical methods focused to forecast the intersection turning movements. Such a review is limited on methods used to distribute the incoming and exiting traffic volumes within the different intersection branches, then generating an estimation of the so-called intersection O/D matrix of turning flows. The second part of the paper is experimental. Two different, but similar, heuristic procedures have described. Then, they have successively applied to some selected intersection real traffic data sets, and the respective computational performances were compared. Namely, the first one is known as proportion methods, while the second one is called as the difference, or deviation, method. Each method of these two starts from an initial matrix, and through iterative steps it reaches the best estimate of the matrix of turning flows, with respect to a given distribution model. Test intersections with their related set of real traffic data have been used as input, and each of the two procedures, as described in advance, was applied to the same numeric nstances. The obtained values were compared in respect to few selected performance indicators. Finally, the computational results were displayed and discussed. On this basis, some insights are drawn and useful remarks for application and future research have been addressed.