Multiple day-of-year (DOY) signal timing plans (STPs) are developed to deal with day-to-day traffic fluctuations during a particular peak hour. Once the operating intervals of these plans are scheduled, they stay valid for the next several months or even years. A STP deployed in this way is rigid regarding any day-to-day changes in the peak-hour traffic and might not be the best option (among the developed DOY plans) for the prevailing traffic. In this paper, we select an appropriate DOY STP for the next peak hour based on estimated traffic profile of that peak hour. We first applied
-means clustering to a historical data set to identify the dominant profiles and develop optimal STPs for the extracted demand profiles. We then utilized recent traffic information to predict the profile of the next peak-hour traffic and select the STP relevant for the estimated profile. For predictions of an upcoming peak-hour traffic profile, we tested multiple methods in the context of the following: (i) within-an-hour prediction where all traffic information available 15–30 min before the start of peak hour is utilized; and (ii) next-day prediction where traffic information from previous (but not of the investigated) days is considered. The results of the analysis, based on a 3-year of volume data along a 5-mi corridor, show that adequate signal plans can be selected for 65% and 75% of the days in a year in the context of a next-day and within-an-hour prediction, respectively. The results merit the further development of this idea.
