Shiri, Thanks for sharing your information.
1, I agree that 10 K bus system would be a reasonably large system and that's what I can see in the papers. 2, I am interested in the term of 'real-time' in the industrial world. I guess the one minute window for power flow is too large if we further consider contingency analysis or other types of analyzes. 3, Thank you for your explanation. I agree with the fact that the number of iterations cannot be estimated without the start point information. But I am not sure about the how the number 90% comes out. Thank you! Best Regards, Yi On Tue, Jun 4, 2013 at 1:24 AM, Shri <[email protected]> wrote: > > On Jun 2, 2013, at 6:34 PM, Yi Liang wrote: > > > Dear all, > > > > I am wondering if the performance of current power flow analysis tool is > good enough in terms of speed. > It really depends on your definition of "good enough" and the kind of > analyses that you want to do. > > 1, What is the size of a power system that is large in reality? 30,000 > bus -system? > I think 10K+ bus transmission network can be categorized as large systems > for power system problems. > > What is the size of the United States power system? > Western Interconnection bulk power grid model is about 15K buses and > Eastern Interconnection about 40K+. U.S. grid I assume would be around > 100k+ buses. > > > > 2, In the real industrial world, is there any requirements for the > speed? what is it? > Yes and it depends on the kind of analyses that you'd like to run. I don't > have a good idea about what is the industrial speed requirement. Maybe some > of MATPOWER's industrial users can answer this but I assume the minimum > requirement would be to have a power flow solution in under a minute? > > > > 3, If we use MATPOWER to solve the power flow analysis problems and the > OPF problems, how long will it take? I don't have that kind of test cases > in hand, so I cannot test them. > MATPOWER's big test cases (~3k buses) take about 0.1 seconds for full AC > power flow and about 5 seconds for optimal power flow on my machine. These > numbers are based on the default > data files provided with MATPOWER that include a "close to solution" > initial guess. I've run cases around 20K buses and MATPOWER's full AC power > flow takes about 1 second. In general, the time taken for these solvers > depends on a number of factors such as initial guess, network sparsity > pattern, loading conditions, constraints, hardware, cache, compilers. > > Can anyone theoretically estimate the time complexity? As far as I know, > it depends on the iterations it takes for newton's method( for example, if > we use newton's method) > For large systems, the most dominant operation is LU factorization and it > consumes about 90%+ of the total solve time. So you could possibly do a > crude theoretical estimate of the time for solving an AC power flow = time > required for LU factorization X number of Newton iterations. For a given > network topology the time for LU factorization would be fixed but > unfortunately the # of Newton iterations vary depending on the loading > conditions, initial guess, and possibly other factors. So determining the > solve time theoretically is difficult. > > Shri > > > > > > Thank you! > > > > Best Regards, > > Yi > > > > -- Yi Liang, Master Candidate Room 403, Coordinated Science Laboratory Department of Electrical and Computer Engineering University of Illinois at Urbana Champaign 1308 W Main Street, Urbana, IL, 61801-2307
