There are three observations I’d like to share with the folks in the professional community who are involved in traffic signal coordination, and I’m hoping to hear if others have noticed anything similar.  

First, when implementing coordination plans that have been developed with an optimization model, I often find ways to improve operations during the fine tuning step.  The last step of the process is to input the final timing settings back into the model and run it one last time to get the MOEs for the final settings.  Despite my own on-site observations and travel time run data, the model sometimes outputs MOEs that look worse than the optimization settings that previously came from the model.  

Second, some agencies in my vicinity have been implementing adaptive systems over the last few years.   One agency in particular has a very talented gentleman who has been in charge of signal coordination for many years and is very capable.   My observations are that the operations with the adaptive systems aren’t much better than he used to implement, sometimes even worse.   The part that disturbs me, though, is that when he gets calls to improve traffic flow at a specific location, he response might now be to verify that the adaptive system was working properly on the day in question, then let the caller know he would not be able to produce anything better than the adaptive system.   We have other agencies in the area using a similar response.

Third, in my conversations with people in the general population, it seems pretty clear to me that people waiting to make a minor movement are more patient when there is a lot of traffic moving on the current green phase and can be very impatient when there is no traffic moving at all.   In extreme cases, the model MOE report can indicate a significant improvement even though the verbal public input can be quite negative.  For example, “I had to wait 10 minutes to get through that light today” (even though the cycle length is 150 seconds).

It seems to me we are relying on a set of assumptions in our models (whether off-line or on-line within an adaptive system) that are due for update.   I’d appreciate hearing the experiences of others.