Shifting & computational complexity

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

Shifting & computational complexity

chris.flath
I am pulling Antonios' last post up here for a greater audience.
----
Hello to everyone

I have started creating some basic Mappings that we allow us to do this kind of shifting but I don't know if this is such a good idea. They are in the controllable-load branch of the household customer.

Right now I am working on the shifting mechanisms of the villages and their households. The problem is the scale on which we wish to work on this.

For example, it it surely not realistic to have all the village shift to a certain timeslot in order to pay less, but the computational cost in having shifting mechanisms on each and every household on each customer maybe be truly discouraging. Think that every timeslot (or every time we publish new Tariffs) we will have to do this all over again, for each tariff and each customer. How are we gonna handle that vast amount of computations?

Any insights on that?

Antonios
Reply | Threaded
Open this post in threaded view
|

Re: Shifting & computational complexity

grampajohn
Administrator
The only hope is to do it statistically. So if 10% of the customers in a model have a TOU tariff, and if 50% of those with a TOU tariff will actually shift, then 5% of the customers in the model will shift. It's not clear that you would have to split the model into a separate set for the 95% that do not shift and the 5% who do - instead, you could simply skew the distribution of the random-number draws that determine when the various loads start and stop.

Does this make sense?

John
Reply | Threaded
Open this post in threaded view
|

Re: Shifting & computational complexity

achryso
Hello there.

I have nearly finished daily shifting evaluation algorithm. Now the household models are taking the tariffs in question and they optimize their consumption for the testing day in order to find the best suited and less costly tariff.

I have some implementation questions about the shifting.

1) The changes I do for the day in question are not saved during the evaluation. That would be really difficult and unnecessary. The question is after I have chosen the tariff that is optimal for my consumption, when should the daily shifting of the customer take place.

Andreas suggested I use a time window same with the one of the tariff selection pattern(for example if I choose a tariff each day, then when i choose tariff I will optimize Customers schedule for the time window until the next tariff publication-selection).

Now that the tariff publication is 3 hours this is problematic, changing schedule every little while is too costly.

So what you think should be the best time to reschedule according to the current tariff?

2) I will now implement shifting in a more wide time window, let's say three days for some appliances that are more consuming. Would that be useful or daily shifting is alright? And if we agree that I should do it, what would be the optimal time window in your opinion?

Looking forward for you insights over this.

Cheers, Antonios
Reply | Threaded
Open this post in threaded view
|

Re: Shifting & computational complexity

Mathijs de Weerdt
Hi Antonios,

Sorry for dropping into this conversation a bit late (and not answering your questions ;)).

I'm interested in improving the computational efficiency of the shifting algorithm. Are you using a statistical method as John suggested, or have you implemented shifting in more detail? If so, are you using an approximation algorithm?

Cheers,
  Mathijs
Reply | Threaded
Open this post in threaded view
|

Re: Shifting & computational complexity

achryso
For now, I have been working in appliance level, in the most detailed form that this can be implemented.
I mean that I take each house that belongs to a certain Village Customer and move its appliances to work in the most cost efficient timeslot and then aggregate the results in order to have the village final consumption for the tariff evaluation.

This is much more detailed from the statistical model suggested by John but you can always create and implement something in a more aggregated level.

If you like the implementation is in my trunk on the github.

https://github.com/chrysopoulos18/powertac-household-customer