State Estimation

The canonical algorithm underlying most modern state estimators. WLS minimizes the sum of weighted squared residuals between measured values and the values predicted by the network model. Higher-confidence measurements (e.g., from Phasor Measurement Units) are assigned larger weights, while lower-accuracy SCADA readings receive smaller weights. The objective function is:

min J(x) = Σ wᵢ(zᵢ - hᵢ(x))²

• zᵢ represents the raw measurement vector
• hᵢ(x) is the nonlinear function relating the state vector to the measurement
• wᵢ is the weight inversely proportional to measurement error variance
• The solution iteratively solves the normal equations using Newton's method

Gross errors in measurements—caused by sensor drift, communication noise, or transducer failures—must be identified and eliminated before corrupting the state estimate. The primary mechanism is normalized residual testing:

• After convergence, the measurement residual vector r = z - h(x̂) is computed
• Residuals are normalized by their standard deviations: rᵢᴺ = |rᵢ| / √Ωᵢᵢ
• The largest normalized residual exceeding a statistical threshold (typically 3.0) is flagged as bad data
• The suspect measurement is removed and the estimator re-runs iteratively
• Chi-squared (χ²) test on the objective function detects the presence of bad data globally

Before state estimation can execute, the physical connectivity of the grid must be translated from a detailed node-breaker model into a simplified bus-branch model. The topology processor:

• Reads the real-time status of every circuit breaker and disconnect switch from SCADA
• Merges electrically connected nodes into logical buses
• Identifies energized islands and detects topological errors (e.g., mismatched breaker statuses)
• Updates the admittance matrix Y_bus used in the measurement function h(x)
• Topology errors are a leading cause of state estimator divergence and must be resolved through generalized state estimation that includes breaker status as a variable

Two distinct paradigms govern how the estimator treats time:

Static State Estimation (traditional):

• Solves a single snapshot of the grid independently at each measurement scan
• No memory of prior states; each solution starts fresh
• Susceptible to momentary anomalies and convergence issues

Dynamic State Estimation (emerging):

• Uses Kalman filtering to predict the next state based on a dynamic model of generator and load evolution
• Fuses the prediction with new measurements for a statistically optimal update
• Provides inherent bad data rejection and smoother state trajectories
• Enabled by high-speed PMU data and requires accurate dynamic models

When multiple interacting bad data points are present, classical WLS can fail. Robust estimators are designed to resist the influence of outliers without iterative deletion:

• Least Absolute Value (LAV): Minimizes the sum of absolute residuals instead of squares, automatically rejecting outliers but at higher computational cost
• Huber M-Estimator: Applies quadratic weighting to small residuals and linear weighting to large residuals, blending WLS efficiency with LAV robustness
• Least Median of Squares (LMS): Minimizes the median squared residual, capable of handling up to 50% contamination but requiring combinatorial search
• These methods are critical during cascading failures when multiple sensors may report erroneous data simultaneously

A recursive mathematical algorithm that estimates the dynamic state of a system from a stream of noisy, asynchronous measurements. It operates in a two-step cycle: prediction using a physics-based model, and update using new sensor data. In power grids, variants like the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) track voltage magnitude and angle dynamics during transient events.

• Handles non-linear power flow equations
• Provides real-time uncertainty covariance
• Essential for dynamic state tracking with PMU data

Statistical techniques that identify and reject grossly erroneous measurements before they corrupt the state estimator's solution. The most common method is the Largest Normalized Residual (LNR) test, which flags measurements whose estimated error exceeds a statistical threshold.

• Detects sensor malfunctions and communication errors
• Uses Chi-squared distribution testing
• Prevents cascading misestimation across the network

A topological assessment that determines whether the available measurement set is sufficient to uniquely estimate the voltage magnitude and angle at every bus. If a network is unobservable, it contains observable islands separated by unmonitored branches.

• Numerical method based on gain matrix rank
• Identifies critical measurement points
• Guides optimal PMU and sensor placement strategies

The computational integration of data from disparate measurement sources—SCADA, PMUs, smart meters—to produce a more accurate and reliable grid state estimate than any single source. Fusion algorithms reconcile different sampling rates and latency characteristics.

• Combines fast PMU data with wide-area SCADA
• Weighted by sensor precision and trustworthiness
• Enables robust estimation during partial sensor failure

A steady-state optimization technique that minimally adjusts raw process measurements to satisfy known physical conservation laws, such as Kirchhoff's current and voltage laws. This provides a consistent, physically plausible dataset for model calibration and operator displays.

• Uses weighted least squares minimization
• Enforces zero net injection at each bus
• Eliminates minor sensor drift and calibration errors

The rigorous mathematical characterization of confidence bounds around a state estimate. It distinguishes between aleatoric uncertainty (irreducible sensor noise) and epistemic uncertainty (model gaps or missing topology information).

• Provides error covariance matrices
• Enables risk-aware operational decisions
• Critical for probabilistic contingency analysis