Large power-systems for power-flow study 1

[smilingbuddha]

I have just been started on a research project which involves finding ways to speed up the solution time of the power-flow problem (see Wikipedia here) which is basically calculating voltages at each of the buses of a given power system.

I have a few data-sets like the standard IEEE power-flow test cases, mentioned here

However, these bus-systems are still very small. Is there any database (like the standard IEEE power-flow test cases mentioned above), but with a large number of buses. Large being on the order of 1000 to 10000 buses, or even larger.

If also there is some standard algorithm for generating a "fake" power system with an arbitrary number of buses, or standard technique to join say several IEEE 300 bus systems to make a larger network, that would also be helpful.

 

[QBplanner]

YOU CAN try to SUBMIT YOUR REQUEST TO WECC to see if they can share their base cases with you. If so, their basecases will meet your requirements.

 

[davidbeach]

Not sure that the problem hasn't been solved. I'm working with a OneLiner model of one corner of the WECC system with a mere 15000 buses and don't find it to run in an unreasonably slow pace on my desk top computer. And, no, it can not be shared.

 

[smilingbuddha]

Hello davidbeach,

Thanks for the reply.

Are you solving the power-flow problem on this network of 15000 buses?
If so, could you give me a rough estimate of the time it takes to do power-flow (i.e. finding voltages
at the non-slack buses) on your desktop.

Thanks

 

[davidbeach]

No power flow, just faults, but that involves solving for voltages and currents everywhere. Working on a breaker rating study that happens to require 448 faults to check the necessary conditions it runs the whole thing is 5 or 6 seconds.

 

[QBplanner]

you cab try to fill out the form and WECC may consider share some cases with you.
please see the following link:

http://www.wecc.biz/library/Pages/Powerflow Base Cases.aspx

"WECC Powerflow Base Cases are available to all members free of charge. Some Powerflow Base Cases are also available to non-members upon approval."
run a 15000 or more node power flow using pss/e or pslf takes less than 2 to 3 seconds. some cases take less than 1 second.

 

[bacon4life]

David, Are you running power flow or just faults in aspen?

Aspen fault calculation are near instantaneous, however solving WECC powerflow cases using PowerWorld often takes a couple of seconds per case on my computer. Not too long if you are doing just one, but our contingency file includes 8000 contingencies. FERC/NERC seems to be pushing for real-time contingency tool to able to resolve the whole set of contingencies every 2 minutes.

 

[davidbeach]

Just faults. Power flow is for the planners. So, yes, I'm dealing with a process that can be solved in a single pass (unless bypassing of series caps becomes a factor), but for power flow you're looking at a single case with multiple iterations instead of a huge number of single pass calculations. There've been afternoons where I know I've run over a million different fault cases. If I test the ratings of every breaker I have modeled I produce a huge number of cases, and then working through many different system configurations, it all adds up.

There's no doubt that there are still incremental improvements in computational speed to be wrung out, but large scale power system analysis has long been one of the cutting edge fields in applied mathematics. In a 15k bus system there are theoretically 225E6 entries in the Z (or Y) matrix; and probably 224.9E6 of them are zero. I'm reasonably sure that no modern system analysis program, fault or load flow, spends any time to speak of looking at the zeros.

I'm reasonably certain that FERC/NERC are going to ultimately drive us to knowing the system state rather than estimating it (synchrophasors everywhere) and if you know the system state, the contingency analysis will be easier; someone will figure out a means to determine which contingencies are irrelevant for any given state and greatly reduce the number to be evaluated at any give time. If the OP wants to really get out in front, on the bleeding edge, it would be the math to determine which contingencies matter at any moment rather than trying to wring out another incremental improvement in sparse matrix analysis.

 

[smilingbuddha]

Thanks so much for the link QBPlanner, I will definetely to get one of these cases.

I had a couple of more questions.

I am pretty new to the field of power-flow studies so please
Please forgive me if they seem trivial, I am pretty new to the field of power-flow analysis myself.

1. In your experience what is the number of buses in the typical largest network on which power-flow is routinely performed?

2. Given ,say, a real n-bus system, how would one join an identical copy of this system to get a 2n bus system. Is it even reasonable to do so.

3. Say I have an algorithm to generate a sparse graph ie a graph where each vertex has the number of outgoing edges bounded by a small integer constant.
This graph can be used to represent the buses in a power-system. (Or better to say a power-system is a sprase graph)
Is there a way to assign reasonable real and reactive powers, basically obtaining a table like the IEEE 30 bus system here:

http://www.ee.washington.edu/research/pstca/pf30/ieee30cdf.txt

Thanks.

 

[HornTootinEE]

It seems to me you're looking for a method that works faster than the established power flow solution methods, which would be independent of the number of bus connections; what I mean is, you could test your methods on the IEEE 30-bus model and somewhat understand how it will behave as it gets larger. My old power book from school covers a couple of mathematical methods for solving power flows that were/are used in traditional power-flow software. So I think beyond "the software" you'll have to find a faster mathematical method-or a better way to implement in your source code.

Interesting project either way.

 

[QBplanner]

Based on my limited experience,

1. In your experience what is the number of buses in the typical largest network on which power-flow is routinely performed?

(I run a power flow case with appro. 25000 buses but my interest is only on about 1000 buses.)

2. Given ,say, a real n-bus system, how would one join an identical copy of this system to get a 2n bus system. Is it even reasonable to do so.
{I have encoutered above scenario only once in the past 20 years. It was because our area network was interconnected to the neighboring system with the new inter-ties. So our power flow cases have to include the neighboring system}

3. Say I have an algorithm to generate a sparse graph ie a graph where each vertex has the number of outgoing edges bounded by a small integer constant.

{I am not sure I fully understand your third question. I am guessing that you are trying to apply the sparse matrix methodology into the power flow program to increase the computation speed?
I am not aware that the PSS/E PSLF using sparse martix but in one of the EMTP typs of software they implemented the sparse matrix to get rid of the zeros first to improve the speed. Frankly speaking, off line power flow studies nowadays, engineers don't really care too much about the speed. However, in really time simulation, speed does play an important role. As far as I know most of the real time tools used in the control centers are using lousy algorithms to trade of the speed other than improve the matrix computation methodology.}
Hope it helps.

 

[dpc]

FWIW, I believe Aspen OneLiner is one of the few (two?) commercial programs that uses a current injection method for power flow simulations. It should be very fast. You might want to do a search for papers and research by William Tinney - he's probably already analyzed every method you are going to think of to evaluate.

 

[QBplanner]

I am not sur ethat Aspen base cases have the load, if no load, what power flow can it solve?

 

[dpc]

Aspen can do power flow (

http://www.aspeninc.com/aspen/index.php?option=com_content&view=article&id=88&Itemid=67)

If certain users choose not to enter load data and use it to solve power flow, that is not related to the capabilities of the software. There is a world outside of WECC.

 

[HornTootinEE]

dpc: Don't tell WECC You said that :O

 

[QBplanner]

thanks dpc!
I checked the website you posted. Looks like Aspen has a powerful power flow tool. Too bad we did not purchase it. Otherwise I would like to try it and compare it with pss/e and pslf.
By the way, this gets nothing to do with WECC.

 

[QBplanner]

but my personal ingonance

 

[dpc]

With the computational power of computers at present, the biggest issue with doing power flow calculations is having accurate power data in the model, not the solution time.

 

[smilingbuddha]

Hi dpc,

By accurate data, do you mean the accuracy of the non-linear solver itself, or the accuracy of the measurements
fed to the non-linear solver?

Because issues regarding non-convergence of the newton solver, or
occasional convergence to a bad solution do arise.

 

[dpc]

Oh, I'm very familiar with convergence issues. No, I was talking about the accuracy of the end-use load data - regardless of the source of the data. The practicing engineer is more concerned about the validity of the end result than the efficiency of the solution engine.