Unscented Transform (UT)

The UT directly addresses the core challenge of Probabilistic Power Flow by propagating uncertainty from renewable generation and load through the nonlinear AC power flow equations. Unlike linearization, which fails under high variance, the UT accurately captures the non-Gaussian skewness of voltage magnitudes.

• Mechanism: Sigma points are selected around the forecasted mean of wind/solar injection and passed through the full Newton-Raphson solver.
• Result: Accurate estimates of voltage violation probability and line flow distributions with only ~2n+1 deterministic evaluations.

In Distribution System State Estimation, the UT serves as a powerful alternative to the Extended Kalman Filter (EKF). It handles the highly nonlinear relationship between limited real-time measurements and the system state without calculating Jacobian matrices.

• Sigma-Point Kalman Filter (SPKF): The UT forms the prediction step, propagating state uncertainty through the grid model.
• Advantage: Superior performance during sudden topology changes or when dealing with non-Gaussian pseudo-measurements derived from historical load profiles.

The UT is critical for tracking the sub-transient and transient dynamics of synchronous machines. The nonlinear differential-algebraic equations governing rotor angle and voltage are poorly served by linearization.

• Parameter Tracking: Sigma points capture the uncertainty in the mechanical torque input and field voltage.
• Stability Prediction: The UT provides a robust estimate of the generator's transient stability margin by accurately projecting the covariance of the rotor angle state through the swing equation.

The highly nonlinear relationship between a battery's open-circuit voltage and its State of Charge, combined with hysteresis effects, makes the UT the standard for robust SoC estimation in grid storage systems.

• Sigma-Point Filtering: The UT propagates the current integration uncertainty through the nonlinear voltage-SoC curve.
• Outcome: Drastically reduced drift compared to Coulomb counting alone, enabling precise control of distributed energy storage resources without frequent recalibration cycles.

Grid operators need accurate dynamic load models (e.g., ZIP + induction motor) for stability studies. The UT enables online parameter identification by treating the unknown model coefficients as stochastic states.

• Dual Estimation: The UT simultaneously estimates the fast-varying system state (voltage) and the slow-varying parameters (load composition).
• Benefit: Captures the time-varying nature of load composition without requiring intrusive customer-level metering, adapting the model as industrial motors cycle on and off.

The UT optimally fuses data from Phasor Measurement Units (PMUs) with traditional SCADA scans. The nonlinear transformation from polar to rectangular coordinates introduces biases that the UT naturally corrects.

• Unscented H-infinity Filter: A robust variant of the UT-based filter that minimizes the worst-case estimation error, protecting against PMU data anomalies and GPS timing attacks.
• Result: High-fidelity, time-synchronized state snapshots that are resilient to non-Gaussian measurement noise.

A nonlinear state estimator that linearizes the system dynamics and measurement functions using a first-order Taylor series expansion around the current state estimate. In power system state estimation, the EKF computes Jacobian matrices of the power flow equations. The Unscented Transform was developed to overcome the EKF's limitations: linearization errors in highly nonlinear systems, the need to derive Jacobians analytically, and poor covariance propagation when distributions become non-Gaussian through nonlinear transformations.

A deterministic sampling filter that uses spherical-radial cubature rules to compute Gaussian-weighted integrals numerically. The CKF selects 2n equally weighted points at the intersection of the unit sphere and coordinate axes. Compared to the Unscented Transform, the CKF offers improved numerical stability in high-dimensional systems (n > 20) because its sigma point weights remain positive, avoiding the negative central weight that can cause covariance matrix non-positive definiteness in the UT.

The direct application of the Unscented Transform to the recursive Kalman filtering framework. Rather than linearizing the state transition and measurement models, the UKF:

• Generates sigma points from the current state estimate and covariance
• Propagates each point through the nonlinear process model
• Reconstructs the predicted mean and covariance from the transformed points
• Updates the estimate using the unscented measurement prediction

This avoids Jacobian computation entirely and captures third-order accuracy for Gaussian inputs.

A matrix factorization of a symmetric positive-definite covariance matrix P into P = LLᵀ, where L is a lower triangular matrix. This decomposition is the computational backbone of sigma point generation in the Unscented Transform. The matrix square root of the scaled covariance is computed via Cholesky factorization, and sigma points are formed by adding and subtracting the columns of L from the mean vector. Numerical stability of the decomposition is critical—ill-conditioned covariance matrices require modified Cholesky or singular value decomposition alternatives.