Support disconnected DC network islands#55
Merged
Conversation
samtalki
force-pushed
the
ck/dc-island-support-followup
branch
from
490f41c to
7ee4b3c
Compare
Benchmark Results (Julia v1)Time benchmarks
Memory benchmarks
|
There was a problem hiding this comment.
Pull request overview
This PR adds first-class support for disconnected (multi-island) DC network topologies across the DC power flow + DC OPF stack, including per-island reference bus selection, KKT/sensitivity generalization to multiple reference constraints, caching adjustments to keep KKT builders lightweight, targeted documentation updates, and expanded test coverage (including APF interop behavior).
Changes:
- Introduces energized-island topology caching in
DCNetworkand exposesreference_buses(net)to provide one deterministic reference bus per energized island (preserving the configuredref_busfor its island). - Generalizes DC OPF modeling + KKT dimensions/indices/residuals/Jacobians to use
n_refreference constraints (one per energized island), with cache invalidation updates to handle topology-driven dimension changes. - Adds multi-island regression tests, updates docs to describe multi-island behavior and nonsmooth sensitivity boundaries, and rejects APF conversion for disconnected cases.
Reviewed changes
Copilot reviewed 20 out of 20 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| test/test_dc_islands.jl | Adds focused multi-island DC tests (islands, isolated buses, bridge opening, cache refresh, regressions). |
| test/test_apf_integration.jl | Adds test asserting APF conversion rejects multi-island networks. |
| test/runtests.jl | Registers the new DC islands test file in the test suite. |
| src/types/show.jl | Improves printing for multi-island reference buses and empty flow vectors. |
| src/types/dc_opf_problem.jl | Builds one reference constraint per energized island; caches _n_ref; clears work on topology invalidation. |
| src/types/dc_network.jl | Adds energized-topology cache, reference bus selection per island, and reduced B factorization for multi-island cases. |
| src/sens/topology.jl | Updates sensitivity documentation to match per-island reference elimination. |
| src/sens/susceptance.jl | Switches KKT layout sizing to the new multi-ref layout helper. |
| src/sens/lmp.jl | Updates LMP decomposition semantics to be uniform per island and uses non-ref set for multiple refs. |
| src/sens/flowlimit.jl | Updates KKT layout sizing to the new multi-ref layout helper. |
| src/sens/cost.jl | Updates KKT layout sizing to the new multi-ref layout helper. |
| src/prob/kkt_dc_opf.jl | Generalizes KKT dims/indices/residual/Jacobian construction to n_ref reference constraints. |
| src/prob/dc_opf.jl | Extracts a vector of reference-constraint duals (eta_ref) aligned with reference buses. |
| src/PowerDiff.jl | Exports reference_buses. |
| ext/PowerDiffAPFExt.jl | Rejects APF conversion when multiple energized islands exist; uses effective reference bus. |
| docs/src/math/dc-power-flow.md | Documents per-island reference elimination and topology cache/thread-safety contract. |
| docs/src/math/dc-opf.md | Documents multi-island reference constraints, KKT dimension changes, and nonsmooth topology boundaries. |
| docs/src/getting-started.md | Updates DC power flow description for per-island reference elimination. |
| docs/src/api.md | Adds reference_buses to the API list. |
| docs/src/advanced.md | Documents reference_buses and topology caching/thread-safety + rebuild requirements. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
samtalki
reviewed
Jun 15, 2026
samtalki
left a comment
samtalki
left a comment
Member
There was a problem hiding this comment.
@cameronkhanpour made an integration pass and reviewed the result. Apologies in advance if my review comments point to anything that was my doing.
I fixed:
- stale island tests that still used removed DCNetwork(::Dict) paths
- unsupported dcline fixture handling in those tests
- empty range display crashes for zero branch DC networks
- related docs/comment wording cleanup
- Reject non-binary switching states in to_apf_network so PowerDiff's energized-island count (b*sw != 0) cannot disagree with APF's on/off branch binarization (sw > 0.5); add a regression test. - Annotate the DCOPFProblem placeholder constructor call so the positional arguments (empty variable/constraint fields, normalized demand, initial _n_ref, optimizer, silent) are self-explanatory. - Make the documented theta-stationarity consistent with the implementation: define E_ref as the n x n_ref reference-selection matrix and include the Diag(sw) gate on the angle-difference dual term across lmp.jl, susceptance.jl, kkt_dc_opf.jl, and dc-opf.md; carry the same gate in calc_congestion_component (a no-op for energized branches where sw == 1). - Clarify that the per-island reference choice is deterministic (configured ref_bus for its island, lowest sequential index otherwise) and that the energy component is exactly uniform per island only in the unregularized limit, in lmp.jl and the math/advanced docs. - Complete the DCNetwork field table in advanced.md (demand, pg_init, topology_cache).
klamike
reviewed
Jun 15, 2026
klamike
reviewed
Jun 15, 2026
Locks in the fix from the prior commit: calc_congestion_component must gate the angle-difference dual term by sw to match the KKT theta-stationarity A' Diag(sw) (gamma_ub - gamma_lb). The test uses a fractionally switched branch with a binding angle limit and asserts the gated result differs from the old ungated form, which it does not for fully closed networks (sw == 1).
Explain that the cache's from_bus/to_bus are the branch endpoints already encoded by the incidence matrix A, materialized as dense Int vectors so the connectivity refresh reads them in O(1). Annotate _refresh_topology_cache!: union-find over energized branches builds the island partition, then one reference bus is chosen per island (lowest index, except the configured ref_bus which is forced for its own island).
Replace the hard rejection of non-binary sw with an allow_fractional kwarg (default false). Fractional switching states are the basis of switching sensitivity, so rejecting them outright was too strict; the only constraint is that APF cannot represent partial switching. With allow_fractional=true the network is converted under APF's own sw > 0.5 binarization. Base the single-island requirement on that same binarization rule (closed and nonzero susceptance) rather than PD's b*sw != 0 energized check, so the slack-bus assumption matches the topology APF builds even for fractional sw. For binary sw the two rules coincide, so the default path is unchanged. Add a union-find connectivity helper and tests for both the opt-in conversion and the still-rejected disconnected-under-binarization case.
Corrected wording in the description of 'topology_cache'.
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 21 out of 21 changed files in this pull request and generated 3 comments.
Comments suppressed due to low confidence (1)
src/types/dc_opf_problem.jl:274
- Phase angle difference constraints are currently gated only by
sw, but the new energized-island logic (andreference_buses) treats a branch as energized iffb[e] * sw[e] != 0. If a branch is de-energized by settingb[e] = 0whilesw[e] == 1, it will correctly split islands / add a reference bus, but these angle-difference constraints will still couple the endpoint angles as if the branch were present. This makes the OPF feasible set inconsistent with the energized-topology definition used elsewhere (and with the comment “Open lines should not constrain angle differences”). Consider gating angle-difference constraints by the same energized predicate (or an equivalentsw_eff), and then updating the KKT residual/Jacobian and LMP congestion decomposition to match.
# Open lines should not constrain angle differences.
phase_diff_lb = @constraint(model, network.sw .* (network.A * va) .>= network.sw .* network.angmin)
phase_diff_ub = @constraint(model, network.sw .* (network.A * va) .<= network.sw .* network.angmax)
samtalki
approved these changes
Jun 15, 2026
samtalki
left a comment
samtalki
left a comment
Member
There was a problem hiding this comment.
Nice work @cameronkhanpour , and thanks @klamike
Address Copilot review: the file-header decomposition comment in lmp.jl, the inline comment in calc_congestion_component, and the LMP decomposition equation in dc-opf.md still showed the angle-difference dual term ungated. Add the Diag(sw) gate so they match the implementation and the theta-stationarity, and reword the inline note to cover fully closed, fractional, and open branches rather than implying energized always means sw == 1.
This was referenced Jun 15, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Stacked follow-up to #54.
This moves the disconnected-DC-topology work that was previously bundled into #54 into its own PR, so #54 can remain focused on the negative net demand fix for issue #53.
Changes
ref_busfor its island.AcceleratedDCPowerFlowsconversion clearly because APF exposes one slack bus.Review Notes
This is intentionally draft because review on #54 raised design questions that are specific to this broader feature:
DCNetworkand the resulting thread-safety contract_n_refas a cached invariant of a builtDCOPFProblemThose are now isolated here instead of gating the issue #53 fix.
Validation
git diff --check ck/dcopf-solve...HEADPkg.test("PowerDiff")SITE_BUILD=true include("docs/make.jl")