PF Convergance
Hi everyone, So, I was having trouble with PF convergance and then I looked into Faq#5 Why does MATPOWER power flow not converge?http://www.pserc.cornell.edu//matpower/#pfconvergence Now in that FAQ, I am having trouble with point number 5. When I try to run the CPF from 0 load and generation to the target case, I get following errors and warnings: step 1 : lambda =NaN, corrector did not converge in 10 iterations NaN's cannot be converted to logicals. Error in printpf (line 175) nzld = find((bus(:, PD) | bus(:, QD)) bus(:, BUS_TYPE) ~= NONE); Error in runcpf (line 392) printpf(results, 1, mpopt); Error in Wildpoldsried_Extraction (line 236) results = runcpf(mpcbase, mpctarget); Warnings: Warning: Matrix is singular to working precision. In newtonpf at 89 In runcpf at 203 In Wildpoldsried_Extraction at 236 Warning: Matrix is singular to working precision. In cpf_predictor at 71 In runcpf at 249 In Wildpoldsried_Extraction at 236 Kindly let me know if you guys have any idea on how to address these issues. Kind Regards, Waqas
Re: convergence problem in runpf.
Jose, Would it be fine with you if the steps you’ve mentioned below are added to MATPOWER FAQ#5 http://www.pserc.cornell.edu//matpower/#pfconvergence Many a times, useful and detailed suggestions, such as what you’ve enumerated, get lost in email exchanges and someone trying to pull up this information has to resort to digging it out of the archive. It’ll be good to have your steps up on the FAQ. Thanks, Shri From: Jose Luis Marin mari...@gridquant.commailto:mari...@gridquant.com Reply-To: MATPOWER discussion forum matpowe...@list.cornell.edumailto:matpowe...@list.cornell.edu Date: Wednesday, August 12, 2015 at 2:42 AM To: MATPOWER discussion forum matpowe...@list.cornell.edumailto:matpowe...@list.cornell.edu Subject: Re: convergence problem in runpf. Mirish, I couldn't help notice that you're building this model from scratch (well, from a database) and you mentioned To make the problem simple I used all buses as PQ buses except one slack bus. This actually makes it harder to converge, unless you have *very* accurate data on what the reactive injections Q (on generator buses) should be. May I suggest a different, incremental approach: 1. Start by keeping all generator buses you can as PV, instead of PQ. They will help holding up the voltage profile. After all, a PV node is a slack bus in what regards the reactive power injection. 2. For the loads, start by zeroing out PD (real power demand), but keeping QD (reactive demand) 3. For generators, set the scheduled PG to zero 4. For lines transformers, zero out the resistance R 5. The resulting network will be a purely reactive power model. Now run a powerflow. If this doesn't have a feasible powerflow solution, it is because some branches have an X parameter that is too large (or equivalently, some load QD is too large). Ramp down the profile of QD until you see convergence. 6. Look at the resulting Q flows across branches, and try to detect anomalously large values (i.e. clear outliers). They will help you uncover values of X that may be wrong (too large). Also, keep an eye on negative X coming from equivalents such as 3-winding transformers; they may also be wrong. 7. Once you get that working, ramp up the values of PD on loads and PG on generators (keeping an eye on the swing's resulting PG, in order to redistribute big excesses). 8. Finally ramp up the resistance on lines. The whole idea is based on the fact that, for transmission networks (lines with RX), the reactive flows are like the backbone on which real power flows can sort of ride on. Get a healthy backbone first, and then you can start transporting real power. Hope it helps, -- Jose L. Marin Gridquant España SL Grupo AIA On Wed, Aug 12, 2015 at 2:36 AM, Mirish Thakur mirishtha...@gmail.commailto:mirishtha...@gmail.com wrote: Dear Mr.Shree, Thank you very much for your help. As per your suggestion and FAQ I tried to find out the problems. The results I got- 1) Fast-decoupled power flow did not converge in 30 iterations. 2) By following http://www.pserc.cornell.edu/matpower/#pfconvergence I tried to runcpf to get good initial guess and i got results like step 1 : lambda = 0.084, corrector did not converge in 10 iterations. Where lambda is 1 and for reducing steady state loading limitation I reduced demand less than 60 % which also failed to converge the power flow. 3) Also I tried to run an optimal power flow according to Dr. Ray's explanation given in following link- https://www.mail-archive.com/search?l=matpower-l@cornell.eduq=subject:%22Re%5C%3A+%5C%5BMatpower%5C%5D+3500+bus+simulation%22o=newest but got the results like- MATPOWER Version 5.1, 20-Mar-2015 -- AC Optimal Power Flow MATLAB Interior Point Solver -- MIPS, Version 1.2, 20-Mar-2015 (using built-in linear solver) itobjective step size feascond gradcond compcond costcond - 0 1200199.7 2.41677 0.71 536.7620 1 946197.39 15.531 1.3682 1.75871 525.914 0.209885 2 954529.91 15.405 0.766107 0.203773 297.341 0.00871422 3 954849.8 12.849 0.7277120.0545952 258.471 0.00033166 4 954629.03 13035 0.69114 0.107402 258.048 0.000228815 5 954614.88 33406 0.692682 0.255673 257.828 1.46744e-05 6 954525.69 14111 0.579613 0.143897 256.765 9.24569e-05 7 954539.42 61648 0.581139 0.501345 255.994 1.42362e-05 8 954518.93 22452 0.573652 0.478609 255.465 2.12443e-05 9 954494.92 8540.4 0.556318 0.403754 254.653 2.48944e-05 10 954523.58 20366 0.556265 0.570707 254.104 2.97206e-05 11 954522.07 6142.4 0.554989 0.647881 256.561 1.57288e-06 12 954573.42
Re: convergence problem in runpf.
Dear Jose and Shree,
Thank you very much for your guidance. I followed the step as you suggested
and runpf converged in 8 iterations. Actually there were three PQ buses
which had excessive Q demand and when I reduced the Q demand for those
buses then it converged properly.
Nice regards
Mirish Thakur
KIT University.
On Wed, Aug 12, 2015 at 9:42 AM, Jose Luis Marin mari...@gridquant.com
wrote:
Mirish,
I couldn't help notice that you're building this model from scratch (well,
from a database) and you mentioned ***To make the problem simple I used
all buses as PQ buses except one slack bus*. This actually makes it
harder to converge, unless you have *very* accurate data on what the
reactive injections Q (on generator buses) should be.
May I suggest a different, incremental approach:
1. Start by keeping all generator buses you can as PV, instead of PQ.
They will help holding up the voltage profile. After all, a PV node is a
slack bus in what regards the reactive power injection.
2. For the loads, start by zeroing out PD (real power demand), but
keeping QD (reactive demand)
3. For generators, set the scheduled PG to zero
4. For lines transformers, zero out the resistance R
5. The resulting network will be a purely reactive power model. Now
run a powerflow. If this doesn't have a feasible powerflow solution, it is
because some branches have an X parameter that is too large (or
equivalently, some load QD is too large). Ramp down the profile of QD
until you see convergence.
6. Look at the resulting Q flows across branches, and try to detect
anomalously large values (i.e. clear outliers). They will help you uncover
values of X that may be wrong (too large). Also, keep an eye on negative X
coming from equivalents such as 3-winding transformers; they may also be
wrong.
7. Once you get that working, ramp up the values of PD on loads and PG
on generators (keeping an eye on the swing's resulting PG, in order to
redistribute big excesses).
8. Finally ramp up the resistance on lines.
The whole idea is based on the fact that, for transmission networks (lines
with RX), the reactive flows are like the backbone on which real power
flows can sort of ride on. Get a healthy backbone first, and then you
can start transporting real power.
Hope it helps,
--
Jose L. Marin
Gridquant España SL
Grupo AIA
On Wed, Aug 12, 2015 at 2:36 AM, Mirish Thakur mirishtha...@gmail.com
wrote:
Dear Mr.Shree,
Thank you very much for your help. As per your suggestion and FAQ I tried
to find out the problems.
The results I got-
1) Fast-decoupled power flow did not converge in 30 iterations.
2) By following http://www.pserc.cornell.edu/matpower/#pfconvergence
I tried to runcpf to get good initial guess and i got results like
step 1 : lambda = 0.084, corrector did not converge in 10 iterations.
Where lambda is 1 and for reducing steady state loading limitation I
reduced demand less than 60 % which also failed to converge the power flow.
3) Also I tried to run an optimal power flow according to Dr. Ray's
explanation given in following link-
*https://www.mail-archive.com/search?l=matpower-l@cornell.eduq=subject:%22Re%5C%3A+%5C%5BMatpower%5C%5D+3500+bus+simulation%22o=newest
https://www.mail-archive.com/search?l=matpower-l@cornell.eduq=subject:%22Re%5C%3A+%5C%5BMatpower%5C%5D+3500+bus+simulation%22o=newest*
but got the results like-
MATPOWER Version 5.1, 20-Mar-2015 -- AC Optimal Power Flow
MATLAB Interior Point Solver -- MIPS, Version 1.2, 20-Mar-2015
(using built-in linear solver)
itobjective step size feascond gradcond compcond
costcond
-
0 1200199.7 2.41677 0.71 536.762
0
1 946197.39 15.531 1.3682 1.75871 525.914
0.209885
2 954529.91 15.405 0.766107 0.203773 297.341
0.00871422
3 954849.8 12.849 0.7277120.0545952 258.471
0.00033166
4 954629.03 13035 0.69114 0.107402 258.048
0.000228815
5 954614.88 33406 0.692682 0.255673 257.828
1.46744e-05
6 954525.69 14111 0.579613 0.143897 256.765
9.24569e-05
7 954539.42 61648 0.581139 0.501345 255.994
1.42362e-05
8 954518.93 22452 0.573652 0.478609 255.465
2.12443e-05
9 954494.92 8540.4 0.556318 0.403754 254.653
2.48944e-05
10 954523.58 20366 0.556265 0.570707 254.104
2.97206e-05
11 954522.07 6142.4 0.554989 0.647881 256.561
1.57288e-06
12 954573.42 6192.9 0.513972 0.716706 253.604
5.32434e-05
13 954575.97 5912.1 0.509457 0.699751 252.612
2.64406e-06
14 954576.23 16534
