Global trip times table

[[I may have this now in the do-it-yourself part.]]

Recall that the missing information is the expected trip time for a given starting time. One option is to generate a global trip times table, i.e. for each time slice and each origin-destination pair the information about the trip time for a departure time within that time slice. This table would be generated from actual performance of simulated travelers/vehicles, that is, all travelers/vehicles departing during the time slice from the same starting location to the same destination would be included, for example by averaging. The table would then be used by the activities generation module to provide estimated trip time information.

The main disadvantages of this approach are:

In a large network, there are easily several hundred thousand links, corresponding to several hundred thousand potential origins/destinations. That is, for a single time slice, our table would have more than $10^5 \times 10^5 = 10^{10}$ entries, corresponding to 40 GByte per time slice, which is clearly too much for most current computing environments.

Going along with the last is that in such a network, with a realistic number of $10^7$ travelers, most entries of the trip time table would be left empty, implying some other method to fill the missing cells.

Implementation

For our simulations, this could be implemented as follows:

From the events file, generate a table of 5min-by-5min origin-destination trip times. That is, for each origin-destination pair and for each 5min bin, you average the travel times of vehicles during that 5min bin.

For example, if there were, between 8:00 and 8:05 (planned departure times), two vehicles traveling from link 100 to link 1900, and the trip took them 30 and 32minutes, respectively, then the expected trip time for a departure between 8:00 and 8:05 is 31 minutes.

Generating this table would concern the system integration specialists.

That table now is read into the activities generation module, and the departure time choice is based on that information.

This would concern the route/acts gen specialists.

If there is information missing between time bins, then interpolate. If there is information missing for early or late times, think about some intelligent solution.