Fine-tuning time-of-day partitions for signal timing plan development: revisiting clustering approaches

Time of day (TOD) control, i.e. applying different signal timings during specified time of day to accommodate temporal traffic patterns, is widely used in the operation of most signalized intersections. In the current practice of TOD partitions, clustering approach has been a popular choice for determining TOD breakpoints given its strength in identifying dissimilarity. However, there still exists an unsolved challenge where the outliers (e.g. rare traffic flow disturbance) may lead to frequent and sometimes drastic changes of TOD plans. Such TOD control can result in unstable traffic states and inefficient operations of signal control systems. This study investigated this issue by revisiting several classical clustering approaches, i.e. K-means, hierarchical and Fisher ordinal clustering. We examined the following factors that may have large impact on the partition results of TOD, namely data collection duration, multi-day and multi-phase choices, and time-dimension in the dataset. The performances of three clustering approaches were systematically compared via an experimental study using field data collected at one intersection in Shanghai, China. This study hopes to offer practical insights through a comprehensive analysis on TOD partitions and assist traffic engineers to fine-tune signal control for signalized intersections, which would remain as a dominant and mainstream traffic control tool for the current urban road networks.