Additional data for post-game analyses

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Additional data for post-game analyses

markus
Colleagues --

I've been tinkering with post-hoc analyses of Power TAC tournament data, specifically with analyzing the impact that broker actions have on customer choices and welfare. It seems to me that one useful piece of information in that area is currently missing from the Power TAC logs, namely the hourly consumption that customers models would have chosen in absence of competitive tariff markets (i.e., their consumption under a single, monopolistic, flat tariff before load shifting).

Two ways in which that that data could be generated come to mind:

(1) Power TAC itself could ask the customer models for their "unshifted" consumption and log the data, or

(2) One could run an additional "monopolistic game" with only one fixed tariff for each competitive game and take the "unshifted" consumption data from there. (Of course, the reproducibility issues that Daniel brought up on this list would have to be sorted out for this to work).

For analysis purposes, (1) is clearly the simpler approach but, as John correctly points out, it breaks the current convention that Power TAC only logs its own state and no ancillary data for analysis purposes. In other words, if a lot of such requests for additional data surface over time, it might bloat the server logs.

I'm curious: Are there any other data analysis efforts out there? Is the "unshifted" consumption data the only thing that's absent or do you see any other needs? How do the customer model and server developers feel about the two options?

Thanks,

Markus
Reply | Threaded
Open this post in threaded view
|

Re: Additional data for post-game analyses

grampajohn
Administrator
markus wrote
... It seems to me that one useful piece of information in that area is currently missing from the Power TAC logs, namely the hourly consumption that customers models would have chosen in absence of competitive tariff markets (i.e., their consumption under a single, monopolistic, flat tariff before load shifting).
I think it's hopeless to try to gather this type of data in-game, because it's not a stateless system. As soon as a shift happens, the rest of the data stream is not comparable. It might be possible to log shifts. We already log curtailments. But I'm not sure what you could learn from that, because the shifts are not independent events. They occur in the context of previous shifts and curtailments.

On the other hand, it would be easy to run another game offline, with the same random number streams and no brokers other than the default broker. You would get the same weather, and the customer behavior would be what you are looking for - the sequence of customer actions in the absence of variable and TOU rates, and in the absence of curtailments.

Does this make sense?

John
Reply | Threaded
Open this post in threaded view
|

Re: Additional data for post-game analyses

markus

> On the other hand, it would be easy to run another game offline, with
> the same random number streams and no brokers other than the default
> broker. You would get the same weather, and the customer behavior
> would be what you are looking for - the sequence of customer actions
> in the absence of variable and TOU rates, and in the absence of
> curtailments.
>
> Does this make sense?

Sorry John, I only noticed the question at the end of your post when I
looked at it again today.

Yes, that does make sense. At the same time, it raises the importance of
making sure that games are fully reproducible before the coming
tournaments. I would suspect that the best comparative analyses come out
of comparing tournament results against a monopolistic setup, since it
will be at tournaments that brokers will compete at their fiercest.
Also, we'll probably observe the richest lineup of strategies during the
official tournaments.

Cheers,

Markus