[lukelowry]

Description

Adds solver events that take a branch in and out of service. Part 1 of #606, modern attempt of #391. This Branch model is designed specifically for ContingencyAnalysis , as introducing breaker and relay raises the same problem we faced with BusFault. Next PR will be a smooth breaker and relay.

@pelesh

(+ Added worked out examples with diagrams to demonstrate topology scenarios that may be encountered and how that impacts the Jacobian)

Proposed changes

• Branch::setEnabled(bool), Implemented using an admittance mask on the branch equations.
• SystemModel::getComponent(id) looks up a branch by its case-file id.
• Solver events enable / disable take a device id and are handled like fault events.
• Example DynamicSimulation/Toy/ThreeBusBasic: BR_1_2 opens at 1 s and recloses at 5 s.
• Tests: BranchTests::singularJacobian (pattern kept, values zero)

Checklist

• All tests pass.
• Code compiles cleanly with flags -Wall -Wpedantic -Wconversion -Wextra.
• The new code follows GridKit™ style guidelines.
• There are unit tests for the new code.
• The new code is documented.
• The feature branch is rebased with respect to the target branch.
• The CHANGELOG.md has been updated to reflect the changes.

Further comments

• Islanding is not handled yet

Repsonse

[pelesh]

Great work!

There is one major item that needs to be addressed and that is to identify cases when taking line out of service makes underlying linear system singular. That needs to be clearly identified in this PR, so that a remedy can be implemented subsequently.

My guess is that islanding would be one such instance. This can be easily checked with ten-gen example. There may be other instances where zeroing out branch Jacobian stamp may lead to singular system Jacobian.

@abirchfield and @shakedregev: please chime in.

The rest are nitpicking suggestions.

[lukelowry]

Addressed!

[lukelowry]

@pelesh Four examples added of some trivial and non trivial examples that we may feasibly run into, in Branch/README.md

[shakedregev]

Where do you set the diagonal on the removed rows to 1?

[lukelowry]

This implementation does not do such, because its not clear how to determine when that should be done (algo for algebraic connectivity, which we could implement but it would certainly be flaky)

[shakedregev]

If a row + column is all 0s, set diagonal to 1. So how are you dealing with this?

[lukelowry]

That is not the only path to singular Jacobian, for example: (Branch/README.md)

And an edge case that is feasible but not probable:

Applications like MATPOWER or PSSE have island-detection but that is not a catch-all. The goal is to be able to catch/detect these scenarios and communicate an expected failure to converge. Any assistance appreciated

[shakedregev]

Are the matrices symmetric? If they are, then any symmetric matrix has what's known as an LBL^T factorization. L is lower triangular and B has 1X1 or 2x2 blocks on the diagonal. If any of those blocks are singular, the matrix is singular.

What does red convey? A removed element?

[lukelowry]

No unfortunately because of the tap&phase shifting transformer model the admittance matrix is not guaranteed to be Hermitian

And yes red was my attempt to indicate "this element has just been removed"

[shakedregev]

If you just want to flag a problematic system, you can use rank revealing LU (I think KLU does this). If you factorize to L*U, by convention L has a diagonal of 1s. So you will get some 0s on the diagonals of U, say in place $U_{j,j}$. Simply set $U_{j,j}=1$, $U_{i<j,j}=0$, $L_{j,i}<j=0$ for all such $j$ s. This will solve an equivalent system to the original one, except it will be nonsingular. Whatever value you get for $x_j$ will just be meaningless.

[shakedregev]

You may have to use SPQR.

[shakedregev]

If all examples fall into 0s on the diagonal, or singular 2X2 blocks, you can simply check for these. You can fix the first by setting the diagonal to 1. For the 2X2, I think it depends on the domain knowledge of what this actually means, but maybe setting the off diagonals to 0 is sufficient.

[pelesh]

I think this can be merged as preliminary capability to take branches out of service. Follow up work should include:

• Better documentation and more examples on possible singular systems when a branch is taken out of service.
• Ability to access branches through a unique ID at O(1) complexity (see #370).
• Remediation of singular systems.

This branch needs rebasing before merging.

[lukelowry]

Rebased! I think its ready.