Lesson 4 of 8

State Estimation & Observability

SCADA delivers thousands of measurements to a control centre every few seconds — but operators never act on the raw numbers. Some meters are inaccurate, some are missing, a few are simply wrong, and voltage angles are not measured at all. State estimation is the step that turns that imperfect stream into one consistent picture of the grid. Every contingency study, optimal power flow and operator display downstream runs on its output.

What "the state" means

The state of a transmission network is the set of voltage magnitudes and angles at every bus. For N buses that is 2N − 1 unknowns, since one angle is the reference. Once the state is known, every line flow, injection and loss can be calculated from the network model — so estimating the state is equivalent to knowing the whole operating point.

What the estimator works with

Analogue measurements: Active and reactive power flows on lines and transformers, bus injections and voltage magnitudes, refreshed every few seconds by RTUs and IEDs.
Status signals: Open/closed positions of breakers and disconnectors. A topology processor turns them into the bus–branch model the estimator actually solves.
Synchrophasors: Where PMUs are installed, directly measured voltage and current phasors — including the angles SCADA cannot provide.
Pseudo-measurements: Forecast or historical values used where real meters are missing, given a low weight because they are uncertain.

A well-designed system has redundancy: noticeably more measurements than unknowns. That redundancy is what lets the estimator average out noise and catch errors.

Weighted least squares

Each measurement z_i is modelled as a known function h_i(x) of the state x plus an error with standard deviation σ_i. The estimator looks for the state that explains all measurements best, trusting accurate meters more than poor ones:

J(x) = Σ (z_i − h_i(x))²σ_i²

Weighted least-squares objective

Because h(x) contains the same nonlinear power-flow equations as a load flow, J(x) is minimised iteratively, much like Newton–Raphson, with a gain matrix G = HᵀR⁻¹H built from the measurement Jacobian H and the error covariance R. This formulation was introduced to power systems by Fred Schweppe in 1970 and remains the industry standard.

Observability

Before solving, the estimator must check that the measurement set can determine the state at all. If a part of the network has too few measurements, or they are placed so that they do not constrain some voltages, that area is unobservable — no algorithm can recover what was never measured. The fix is either more meters, PMUs at well-chosen buses, or pseudo-measurements, accepting that the estimate there is less certain.

Bad data: catching the meter that lies

Redundancy also makes errors visible. After solving, each residual z_i − h_i(x̂) is compared with what noise alone would explain. A chi-square test on J flags that something is wrong; the largest normalised residual test points to the measurement that is most likely responsible, which is removed before the estimator runs again. Typical culprits are a reversed CT polarity, a stale value or a telemetry fault.

The limits of detection

Bad data can only be detected where there is redundancy. A critical measurement — one without which the network would become unobservable — always has a residual of zero, so an error in it is invisible. Topology errors, such as a breaker reported closed that is actually open, are even harder: they distort the model itself rather than a single number.

A cybersecurity angle

In 2009 researchers showed that an attacker who knows the network model can craft false measurements that shift the estimated state while passing every residual test. Such false data injection attacks are one reason measurement integrity and PMU coverage feature so prominently in grid cybersecurity.

Where state estimation sits in the control centre

The chain runs from SCADA to the topology processor, through the state estimator, to the applications operators rely on: contingency analysis for N-1 security, optimal power flow and the displays on the control-room wall. It typically runs every few seconds to minutes. Two developments are changing it: PMUs measure angles directly, so with enough of them the problem becomes linear and can run at the PMU reporting rate; and smart-meter data is making state estimation possible in distribution networks, where it was long unaffordable.

Key takeaways

The state is the set of bus voltage magnitudes and angles; with it, every flow in the network follows.
State estimation combines redundant, noisy SCADA, status and PMU measurements into one consistent operating point.
Weighted least squares trusts accurate meters more and is solved iteratively with the power-flow equations.
Observability analysis checks whether the measurements can determine the state at all.
Redundancy allows bad data to be detected and removed — but not in critical measurements, and topology errors are harder still.