On-line testing of the McMaster incident detection algorithm under recurrent congestion

The work reported here represents an elaboration of the logic for incident detection identified in previous work conducted at McMaster University. The improved incident detection logic has gone through three levels of testing; data from the Freeway Traffic Management System on the Queen Elizabeth Way in Ontario were used. An improved logic that could recognize and then ignore recurrent congestion and that could identify incidents that occurred within recurrent congestion was developed and tested off-line. The data used for this stage of the work consisted of 39 days from early summer 1990. The results were sufficiently promising that the algorithm was then installed on-line, and its results were reported to a file instead of to the system operator. Following a period of initial testing and revision to the algorithm and parameters, a major on-line test was conducted during 64 normal weekdays from March 12 to June 18, 1992. The algorithm detected 19 of 28 incidents, a 68 percent success rate. For the 19 incidents, the algorithm time to detection averaged 2.1 min after the time recorded in the operator's log; the median time to detection was 1 min later than for the operator. The false alarm experience was 20 in the 64 days of test, or one in every 6.4 operator shifts.