search

sharnett / jabr-power-flow

A Python implementation of a conic programming formulation to the power flow problem on tree networks, as described by Rabih Jabr.
MIT License
7 stars 4 forks source link

a generator isn't quite the same as a negative load in MATPOWER #5

Closed sharnett closed 10 years ago

sharnett commented 10 years ago

Since the voltages are fixed at the generators, but free at the loads. I think. Verify this, then probably convert MATPOWER cases to the Jabr/distribution network style.

sharnett commented 10 years ago

Couple solutions:

  1. (better) add more fixed voltage constraints to Jabr formulation, beyond just u1 = v1^2/sqrt(2)
  2. take original Jabr solution generator voltages, add those to MATPOWER, solve, compare
sharnett commented 10 years ago

I'm looking at case_ieee30_tree.m, where the Jabr method gives a different answer from Newton.

Proposed solution #1 seems to result in an infeasible model. Making them inequality constraints doesn't help. Removing the load flow constraints on the extra generators (so they're all slack buses) makes it feasible again, but the answer is still different from Newton.

I guess I'll try solution #2.

sharnett commented 10 years ago

I think I'll try #3 first: alter MATPOWER case file to remove generators, make all but one bus a load.

sharnett commented 10 years ago

Actually, generators are 'PV' buses -- the reactive power isn't fixed, just the real power and voltage magnitude. But removing only the reactive flow constraint at the extra generators didn't seem to do the trick. Needs further investigation.

sharnett commented 10 years ago

Is it possible that there are multiple solutions? This is no longer a distribution network.

sharnett commented 10 years ago

Look into trying different starting points for both Jabr and MATPOWER.

sharnett commented 10 years ago

Using the Jabr solution as a starting point for Newton didn't change Newton's answer on case_ieee_30. Can try using the Newton solution as a starting point for Jabr, but that probably won't do it since Jabr is a convex problem. Worth trying anyway.

If that fails, just convert everything to a distribution network, letting voltages vary freely at all buses but one.

sharnett commented 10 years ago

GOT IT. There were some extra settings in the (shunt susceptances and tap ratios) that I needed to set to zero.

sharnett commented 10 years ago

Verified on case_ieee30 and case118_v2. I think we might finally be done here.