Initial conditions for Power TAC simulation

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

Initial conditions for Power TAC simulation

grampajohn
Administrator
Dear Colleagues -

I have been looking at the CustomerBootstrapData type (see issue #364 and #365) as we convert to Java, and it seems to me that we need to have a more in-depth conversation about setting and communicating the initial conditions of the simulation, free of the intense time pressures of last June/July. Currently we communicate static simulation parameters in the Competition instance, and we run the customer models for two weeks under the default tariffs to generate load records that brokers can use to evaluate customers. But to intelligently build tariffs, it seems they would need to know not only the weather conditions that underly the customer data, but also the prices in the wholesale market that reflect the net load being served.

So here's a suggestion that would give us all of this data in some sort of reasonable way:

- we start the simulation n timeslots prior to the start of the competition, where n = 336 would be the two-week interval we are using for generating customer load. Without the overhead of brokers and communications, we could hope that this could be done online in less than 2 minutes; otherwise it could be done offline and packaged up as a config file.

- Data from the initialization period could be packaged up as the customer bootstrap data is packaged, as simple arrays of numbers indexed by timeslot. This could include consumption and production by customer model, weather conditions, and the price and quantity data from the market. Somehow the market data should be aggregated in a way that does not overly emphasize the effect of the default broker's trading strategies, for example by quoting simply the aggregate net prices and quantities of delivered power, rather than the ClearedTrade data from each traded timeslot.

I think this is an important issue, given the limited time brokers have to learn the environment of a given simulation. I would very much appreciate some feedback on what brokers should be able to see at the beginning of a simulation, and how we might generate and provide it in a reasonably convenient fashion.

Thanks -

John
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

chris.flath
Hi John - I like this proposal and think it should be well manageable and gives brokers a much better learning environment to start with - effectively all static learning is encapsulated in the dump and competition learning is then about learning other broker strategies and customer choice behavior.

One caveat I see is that market data would be somewhat uninteresting as the market price will be just the price of the marginal genco given aggregate demand (as conveyed by the default broker).
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
chris.flath wrote
One caveat I see is that market data would be somewhat uninteresting as the market price will be just the price of the marginal genco given aggregate demand (as conveyed by the default broker).
That's true, but there will be some hour-to-hour and day-to-day variability in both aggregate demand and genco capacity/price quotes, and so brokers will be able to see the price curve over this range of variability. Also, I would like to make the genco behavior a bit more interesting, perhaps by acquiring some real-market data, then condensing and scaling it to fit the parameters of the simulation.
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

chris.flath
grampajohn wrote
That's true, but there will be some hour-to-hour and day-to-day variability in both aggregate demand and genco capacity/price quotes, and so brokers will be able to see the price curve over this range of variability. Also, I would like to make the genco behavior a bit more interesting, perhaps by acquiring some real-market data, then condensing and scaling it to fit the parameters of the simulation.
Ok if we will implement some changes to GenCo behavior (and then probably also to defaultBroker trading) we should get more meaningful market data - actually it will be interesting to see how helpful it will be. We should perhaps consider weather-dependent large scale renewables as gencos - we would then have a corresponding feedback loop into the market.
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
chris.flath wrote
We should perhaps consider weather-dependent large scale renewables as gencos - we would then have a corresponding feedback loop into the market.
A member of the Minnesota team, Ryan Finneman, is working on a wind-park genco model. It should be ready in a few weeks.
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

Prashant Reddy
This all sounds good to me.  This should be obvious, but I'd just like
to highlight that brokers shouldn't make the assumption that the
bootstrap data necessarily looks like the capacity that those
customers will exhibit once the competition starts.  Depending on the
tariffs offered by the competition brokers, the actual capacity may be
quite different; i.e., the bootstrap data is only representative of
capacity under the initial conditions and it is up to the brokers to
figure out how the subsequent competition conditions are different
from the initial conditions and how those differences might affect the
customers' capacities.

Thanks,
Prashant
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
I have written up a more detailed initialization proposal on the participant wiki. Please let me know if this makes sense.
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
Here's an important question that arises from this proposal: When does the simulation clock start? In other words, what is the "base" time of the simulation, and at what time does the timeslot with serial number zero start? It seems that neither answer results in overall simplification for everybody, but I am inclined to have the clock start at the beginning of the initialization period, so brokers would see that two weeks have already passed when they log in. The other choice will make life difficult for at least some server components, and could be a nightmare for ex post analysis (you would have two timeslots with serial number 4, for example).

Does this make sense?

John
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

Prashant Reddy
Yes, makes sense to me (to have the brokers join after two weeks have elapsed).

Thanks,
Prashant
Reply | Threaded
Open this post in threaded view
|

RE: Initial conditions for Power TAC simulation

Wolf
Administrator

This makes total sense, and is done very frequently in evolutionary computing and other bootstrapped simulations. This will also allow us to determine how much bootstrapping data and time is enough, i.e. the simulation should reach some dynamically stable equilibrium with the default broker, and then the competition brokers should join.

 

Thanks,

 

Wolf

 

From: Prashant Reddy [via Power TAC Developers] [mailto:[hidden email]]
Sent: Monday, October 10, 2011 19:37
To: Wolf Ketter
Subject: Re: Initial conditions for Power TAC simulation

 

Yes, makes sense to me (to have the brokers join after two weeks have elapsed).

Thanks,
Prashant


If you reply to this email, your message will be added to the discussion below:

http://power-tac-developers.975333.n3.nabble.com/Initial-conditions-for-Power-TAC-simulation-tp3396641p3410187.html

To start a new topic under Power TAC Developers, email [hidden email]
To unsubscribe from Power TAC Developers, click here.



Disclaimer
De informatie verzonden in dit e-mail bericht inclusief de bijlage(n) is vertrouwelijk en is uitsluitend bestemd voor de geadresseerde van dit bericht. Lees verder: http://www.eur.nl/email-disclaimer.

The information in this e-mail message is confidential and may be legally privileged. Read more: http://www.eur.nl/english/email-disclaimer.

Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

chris.flath
In reply to this post by grampajohn
grampajohn wrote
It seems that neither answer results in overall simplification for everybody, but I am inclined to have the clock start at the beginning of the initialization period, so brokers would see that two weeks have already passed when they log in.
I don't think it is a problem for the brokers to not start at t=0 but rather at t=14*24 and as it keeps consistency over the whole competition time frame it makes a lot sense to do it this way.
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
In reply to this post by Prashant Reddy
I have written up a detailed description of the start-of-game process that should help put this discussion in context and provide a blueprint for implementation. Please take a look and see if it makes sense, and if I've forgotten anything.

Thanks much.

John
Reply | Threaded
Open this post in threaded view
|

Re: Initial conditions for Power TAC simulation

grampajohn
Administrator
I want to revisit the bootstrap data discussion briefly. As implemented, we are recording the net consumption of the customer models; we are not separating production from consumption. Possibly that's because we don't have any models that produce power right now and I've been lazy. But I actually think it's pretty important to record consumption and production separately. In other words, the customer bootstrap data should be broken down by tariff subscription (customer-tariff pair) rather than by customer. I guess we really need a customer model that produces some power to see how this sim is going to work. How hard would it be to add some solar panels to the existing bottom-up model?

Suggestions? Volunteers?

Thanks -

John