State Estimation Extensions

These features build on the WLS core of State Estimation, share its result structure (SEResult) and follow its Write-back policy.

FACTS in state estimation (se_view)

State estimation is a snapshot: runse! never invokes the outer control loop, every controller is frozen at its current operating point. se_view reports the frozen view without mutating the net.

Links in state estimation

Bus links (impedance-less couplers, addLink!) are not part of the Ybus, so SE runs on the contracted net, like the power flow: every closed-link cluster is fused onto its representative bus, open links remain real separations. Link flow measurements are allocation inputs, never WLS rows.

On a fused cluster, VmMeas/VaMeas on a member remap exactly. PinjMeas/QinjMeas aggregate only as a whole: when every member carries an active injection measurement of the kind, one LINKAGG row with the summed value and root-sum-square sigma replaces them, otherwise all member injections of that kind are excluded with a warning (a partial sum would bias the fused balance). ShuntQMeas and bus-referenced ImagMeas follow their device to the representative; branch measurements keep their branch, and a branch shorted inside a cluster loses its flow measurements with a warning. calcLinkFlowsSE! is a weighted least-squares split per link component (nodal balance rows at weight 1, each link measurement at 1/sigma^2; without measurements it equals calcLinkFlowsKCL!) and runs after the write-back because it consumes the estimated branch flows and shunt powers.

Island-wise estimation

A net can split into several synchronous AC islands (multi-area MATPOWER cases, CGMES deliveries with stub islands). Like the power flow, runse! partitions the measurement set onto the islands and estimates every measured island with its own angle reference; unmeasured islands are skipped and reported.

Closed links fuse islands into one estimation group; an open link remains a real separation. An island with its own slack keeps it as reference; one without gets the MATPOWER-like promotion of the power flow (detect_ac_islands' reference bus), an unpowered stub without any candidate an angle pin on its first bus. Unestimated islands keep their start values in SEResult.voltages; validate_measurements and the sequential elimination work across islands with the caller's measurement indices. Independent chi-squares add, so the summed J/dof is a valid overall statistic, while the per-island verdict keeps a local problem visible.

Transformer tap estimation

A tap position that disagrees with the model (a stale SCADA position, a local operation) poisons the estimate around the transformer: the band test reports :high and the suspects cluster there although every telemetry row is healthy. Releasing the tap as an estimation state resolves this. The estimator solves with the tap as a continuous state, rounds it to the nearest mechanical step, and solves once more with the tap fixed; the reported result is that of the fixed run.

Use

Notes

• The regulator states start from the current branch position; a position contradicting the declared mode is an error. Fixation rounds regulator 1 on the ratio grid of the transformer and regulator 2 on the shift grid phase_step_deg; out-of-range positions are flagged and clamped. The ratio grid is the fraction grid t = ratio / (1 + n * tap_step) for nameplate data (MATPOWER mpc.sparlectra.tap_changers, SCF), the step ratios ratio + n * tap_step or the table points of a CGMES RatioTapChanger, and the model steps of a typed ratio model. SEResult.voltages/objectiveJ/dof are those of the fixed run.
• Freeze reasons: radial_no_voltage_pin for a bridge transformer (its removal cuts the net) whose cut-off side carries no active voltage measurement, not_observable for a tap state the measurements cannot pin (a :both release can freeze partially). A frozen transformer solves bitwise identical to no release and still appears in tapEstimates with its reason. Release and back-calculation are mutually exclusive per transformer.
• The Web UI's "estimate taps" first solves once with the taps frozen and then estimates them from that state (from a flat start many tap states crawl to the solution). The estimate is kept only if its fixed positions fit the measurements at least as well as the model positions of that first solve. If it is rejected, or the run does not converge with released taps (a set collectively too thin for them), every released tap is frozen back to its model position and the estimation repeats once; the run log says so and the result metadata carries se_tap_estimation_fallback = true with se_tap_estimation_fallback_reason not_converged or worse_fit: the tap positions in such a result are model values, not estimates.
• A band failure that appears only through the fixation is not bad data: when the continuous fit passes or undershoots the band and the fixed run fails it :high, tapFixation.offgrid_residual is set (summary note :offgrid_tap_residual). The true position then sits between steps, or the step table (tap_step, neutral position) is wrong.

Cascade model. The released tap enters as one or two WLS states with

\[t(r_1, r_2) = \frac{t_{base}}{(1 + r_1)\,(1 + r_2\, e^{j\alpha})}\]

in the calcSkewAngleTap/calcAdmittance convention of the importers; every prediction re-adds the branch at the cascade position, so at the initial position the released model reproduces the stamped power flow to machine precision.

Fixation and dof. The fixation raises dof: a released tap is one extra state, dof_before = m - (voltage states + taps), and the fixation removes it, dof_after = m - voltage states = dof_before + released taps. J moves the other way, e.g. J 58.5 (dof 41) -> J 62.5 (dof 42), both inside their band; only a fixation that jumps out of the band is a finding. On multi-island nets the sums aggregate like the island chi-squares.

Machine transformers. A generator step-up transformer's machine terminal is AVR-held and not independently observed, so a released tap would only absorb the voltage control; the Web UI mass release skips such transformers (a generator bus hanging on the transformer alone), an explicit setTapEstimation! stays possible. The back-calculation uses the estimated network-side voltage, the AVR setpoint, the dispatch P and the measured machine reactive power.

Guards. Tap states are outside the observability count (like the gated current rows); the bridge test and the per-column numerical test check the redundancy around the transformer (a meshed path or measurements on both sides) directly. A frozen regulator keeps its exact model value inside a written cascade.

Topology validation

A wrong service state (a breaker recorded closed that is open, or the reverse) poisons an estimation differently from bad telemetry: the errors cluster around one station and no elimination cures them. Three advisory stages validate the topology: findings and recommendations only, never a mutation of a status, a measurement or the model.

Stage-1 findings: a measured flow over an open element (:open_element_with_flow), a closed branch reading dead at both measured ends while the model expects a clear flow (:closed_element_without_flow, low severity), voltage measurements disagreeing across a closed link (:closed_link_voltage_mismatch), and the node balance at completely measured nodes (:kcl_violation; shunt term from the measured voltage, partially measured nodes skipped). Out-of-service elements enter only the status-contradiction check.

Stage 2 takes a station as a closed-link contraction cluster; stage-1 findings at the same station are noted as precheck_agreement, and with correlation columns on a fully correlated cluster carries correlated_group. A curable gross error never produces a topology finding, because the elimination succeeds first.

Stage 3 supports a hypothesis when the :high failure disappears under the toggle (inside or below the band); several supported hypotheses are flagged ambiguous. Candidates come from the stage-2 stations and the stage-1 status contradictions (candidates = :auto, capped by max_candidates, with per-candidate timing), or from an explicit list. Nothing is switched automatically.