Closed clintonsteiner closed 1 year ago
Taking a look at this now
https://github.com/NREL/hive/pull/182 Link to PR lives here, will probably want to play around with some of the assumptions we are making
Manhattan scenario time increased to 232.56 seconds from 184.81 seconds
Manhattan scenario time increased to 232.56 seconds from 184.81 seconds
and i'm seeing why, the manhattan scenario is using "shortest_time_to_charge" charging dispatch. which, i mean, "shortest_time_to_charge" is a lot more sensible of a heuristic, but, a good deal more compute intensive and implies more perfect information for drivers about charge times at stations. was there a discussion amongst pycon folks to make this change?
there's some tuning that could improve on these runtimes via the h3_search_resolution. the default charge search performs a 2-level expanding ring search, and the size of the search rings (h3_search_resolution parameter in HIVE) changes how big those hexes are in the search. if the rings are very small, we see more hexes in each expanding ring, but less stations per ring. that would do better in a denser scenario. this is set manually; in the future, we could target some average stations per hex and set it programatically, but, eh, scenario development is something we don't do everyday to warrant that kind of approach yet.
No sorry, just using that shortest_time_to_charge to test the changes in a compute intensive way. Removed this and opened up a fixed PR
Closed by #184
To deal with the grid throttle scenario, what might be a good thing in the future would be to pass some max charge time argument. that at least can be set to int((sim.end_time - sim.sim_time) / sim.timestep_duration_seconds)
So in the scenario where the grid is throttled during charging, then 10 minutes later begins working again, we want to ensure we dont exit too early.
See https://github.com/NREL/hive/issues/146 for more backstory