Struct PersistentEstimator 

pub struct PersistentEstimator { /* private fields */ }

A reusable estimator that keeps its factorization between solves.

estimate is the one-shot entry point: it builds the state layout, the constraint set and the solver’s symbolic factorization, uses them once, and throws all three away. For a single estimate that is the whole cost of doing business; for a sequence of them — a time series, a scenario sweep, a live feed refreshing the same telemetry — it is repeated work, and measurably so. Power flow solved the identical problem with PersistentSolver; this is its counterpart.

When a cached factorization stays valid

Four things are cached — the state layout, the constraint set, each measurement’s resolved MeasurementModel, and one symbolic factorization — and all four hold under the same condition, which is what makes it one condition to state rather than four.

The gain matrix’s sparsity pattern depends on the topology and on the measurement set’s structure — which quantities are measured where — but not on their values. So this stays valid while:

• the network is unchanged, and
• the same measurements exist, on the same targets, in the same order.

New readings on the existing sensors are the case this is built for and need no reset. Sigmas may change too: a weight scales G’s values, never its pattern. Adding, removing or reordering a measurement does invalidate it, as does anything that changes whether an angle is measured at all — that flips whether a reference is pinned, and with it the number of unknowns. Call reset then.

Nothing here detects a stale cache. The same is true of PersistentSolver, and for the same reason: the check would cost as much as the work it saves.

Implementations

impl PersistentEstimator

pub fn new(options: SeOptions) -> Self

pub fn reset(&mut self)

Discards every cached artifact. Call after changing the network or the structure of the measurement set.

pub fn estimate( &mut self, measurements: &[Measurement], buses: &mut [Bus], net: &SeNetwork, ) -> SeReport

Estimates, reusing whatever this estimator already holds.

The iterative-linear method has nothing to reuse between calls — it factorizes once per call already, and its matrix depends on the starting magnitudes, which move — so it delegates to the one-shot path. The saving here is Newton-Raphson’s, which is also the slower method and so the one that wanted it.