State Estimation

The estimator requires a redundant set of measurements—more sensor inputs than the minimum needed to solve the system. This overdetermination allows the algorithm to perform statistical bad data detection and identification. By analyzing the normalized measurement residuals (the difference between the raw measurement and the calculated estimate), the system can flag gross errors from failed current transformers (CTs) or potential transformers (PTs). The largest normalized residual test is the standard method for iteratively identifying and removing corrupted data points, ensuring a single faulty sensor does not corrupt the entire operational picture.

The industry-standard computational engine for static state estimation is the Weighted Least Squares (WLS) algorithm. It minimizes the sum of the squares of the weighted deviations between the measured values and the estimated true values. The weighting matrix is typically the inverse of the measurement error covariance matrix, giving higher confidence to precise sensors like phasor measurement units (PMUs) and lower confidence to older SCADA transducers. The iterative Gauss-Newton method is used to solve the non-linear power flow equations until the state vector converges to a stable solution.

Before estimation can begin, the network must be deemed observable. A power system is algebraically observable if the set of available measurements allows the unique determination of all bus voltage phasors. Topological observability analysis checks if a spanning tree of the network graph can be formed using only measured branches. If the system is unobservable, pseudo-measurements—forecasted loads or generation based on historical data—must be injected to make the system solvable. Observable islands are identified and processed independently.

• Static State Estimation (SSE): Solves for the system state at a single snapshot in time using only the current measurement scan. It assumes a quasi-steady-state operating condition.
• Dynamic State Estimation (DSE): Utilizes a system model (like the Kalman filter) to predict the state evolution over time. It fuses the prediction with new measurements to provide a filtered, time-coherent trajectory. DSE is critical for tracking fast electromechanical transients and provides a natural buffer against momentary sensor dropouts.

The integration of synchrophasor data fundamentally improves estimation accuracy. Unlike traditional SCADA scans that arrive every 2-4 seconds, PMUs stream time-synchronized, complex voltage and current phasors at 30-60 samples per second. Linear State Estimation (LSE) leverages PMU data exclusively to formulate a linear measurement model, eliminating the need for iterative non-linear solvers. This enables sub-second solution times, transforming the estimator from a steady-state monitoring tool into a real-time dynamic monitoring platform.

A state estimator must distinguish between measurement errors and topology errors. A topology error occurs when the digital status of a circuit breaker or switch in the network model does not match its physical position. This creates a gross structural mismatch. Generalized state estimation extends the state vector to include breaker status variables, allowing the algorithm to automatically detect and correct topology errors by treating suspect breakers as unknown parameters to be estimated, rather than fixed inputs.

The core mathematical engine of classical state estimation. The Weighted Least Squares (WLS) algorithm minimizes the sum of squared differences between measured values and calculated values, weighted by the inverse of each measurement's error variance.

• Assigns higher trust to precision instruments (low variance) and lower trust to noisy sensors.
• Iteratively solves the non-convex AC power flow equations until convergence.
• Bad data detection relies on analyzing the normalized residuals after the WLS solution converges.

The topological pre-check that determines if a state estimator can solve for all unknown bus voltages. A network is algebraically observable if the measurement Jacobian matrix has full rank.

• Identifies observable islands: isolated sections of the grid with sufficient measurements to solve independently.
• Detects critical measurements: individual sensors whose loss renders the system unobservable.
• Guides meter placement strategies to eliminate unobservable branches and ensure redundancy.

The statistical post-processing that identifies gross measurement errors before they corrupt the state estimate. The Chi-squared test on the sum of weighted squared residuals flags the presence of bad data.

• Largest normalized residual test identifies the specific measurement most likely to be erroneous.
• Leverage points: measurements that heavily influence the estimate but have small residuals, making them difficult to detect.
• Modern approaches use robust estimators like Least Absolute Value (LAV) that are inherently less sensitive to outliers.

The pre-processing stage that translates breaker and switch statuses into the bus-branch model used by the state estimator. Errors here are the most common cause of state estimation divergence.

• Maps physical substation connectivity (node-breaker model) to the computational bus-branch model.
• Topology errors: incorrect breaker statuses that create false branches or merge buses that are physically separate.
• Generalized state estimation extends the algorithm to estimate breaker statuses alongside analog measurements.

An advanced formulation that incorporates the time evolution of system states using Kalman filtering techniques. Unlike static estimation, it predicts the next state based on a dynamic model before processing measurements.

• The Extended Kalman Filter (EKF) handles the non-linear generator swing equations for rotor angle and speed estimation.
• Provides predictive capability: forecasts state a few seconds into the future for early warning of instability.
• Enables anomaly detection by comparing predicted states against actual measurements to identify sudden changes.