Least Absolute Value (LAV)

LAV inherently assigns zero weight to outlier measurements, eliminating the need for separate bad data detection and removal cycles. While WLS requires a Chi-Square test and iterative re-weighting, LAV's L1-norm geometry ensures that gross errors are completely rejected in a single estimation run.

• Mechanism: The objective function minimizes the sum of absolute residuals, not squared residuals
• Result: Outliers do not skew the estimate; they are treated as irrelevant
• Contrast: WLS amplifies outliers due to squaring, requiring post-hoc statistical tests

LAV is highly sensitive to leverage points—measurements located at the extremities of the factor space. A single bad leverage point can force the LAV estimator to pass through it, corrupting the entire state estimate.

• Definition: A leverage point has a large influence on its own fitted value
• Risk: LAV can break down if a gross error occurs at a critical injection measurement
• Mitigation: Strategic PMU placement or pre-screening with projection statistics

LAV has a breakdown point of 1/n, meaning a single strategically placed outlier can cause unbounded error. This contrasts with high-breakdown estimators like the Least Median of Squares (LMS).

• Breakdown Point: The fraction of contamination an estimator can tolerate before producing arbitrarily large errors
• LAV Limit: Fails if contamination exceeds 1/n in leverage positions
• Practical Implication: Requires careful measurement redundancy and placement

LAV requires solving an LP at each iteration, which is computationally heavier than WLS's sparse matrix factorization. However, modern interior-point solvers and the elimination of bad data post-processing narrow this gap.

• WLS: One sparse Cholesky factorization per iteration
• LAV: Full Simplex or interior-point LP solve per iteration
• Modern Context: GPU-accelerated LP solvers make LAV viable for real-time distribution systems

While LAV applies linear weighting to all residuals, the Huber M-Estimator applies quadratic weighting to small residuals and linear weighting to large ones. This hybrid approach retains Gaussian efficiency for clean data while bounding the influence of outliers.

• LAV: Pure L1-norm, treats all residuals linearly
• Huber: L2-norm for |r| ≤ k, L1-norm for |r| > k
• Trade-off: Huber requires tuning the threshold parameter k; LAV is parameter-free

The classical statistical benchmark that minimizes the sum of weighted squared residuals. Unlike LAV, WLS is highly sensitive to outliers because squaring amplifies large residuals. It requires a separate Bad Data Detection step (Chi-Square test) and iterative re-weighting to handle gross errors. WLS remains the industry default due to its computational efficiency and differentiability, but it lacks the automatic bad data rejection property inherent to LAV.

A robust compromise between WLS and LAV. The Huber function applies quadratic weighting to small residuals (like WLS for Gaussian noise) and linear weighting to large residuals (like LAV for outliers). This is controlled by a tuning constant. It provides high statistical efficiency under normal conditions while bounding the influence of bad data, avoiding the computational complexity of pure LAV linear programming.

A post-estimation diagnostic suite required by non-robust estimators like WLS. The Normalized Residual Test flags measurements where the residual exceeds a statistical threshold. The Chi-Square Test evaluates the global fit. LAV largely eliminates the need for these routines because it automatically assigns zero weight to outlier measurements during the optimization, effectively performing bad data identification and suppression in a single step.

The mathematical optimization framework used to solve the LAV problem. Because the absolute value objective is non-differentiable at zero, it cannot be solved with gradient-based methods like Newton-Raphson. Instead, the problem is reformulated as a linear program by introducing slack variables and solved using the Simplex method or interior-point algorithms. This LP formulation is the computational cost driver that historically limited LAV adoption.

The prerequisite check performed before any state estimation. It determines whether a unique solution exists given the available measurements and network topology. LAV inherits the same observability requirements as WLS: the Gain Matrix must be non-singular. However, LAV's robustness can mask critical measurement losses—if a measurement is an outlier, LAV ignores it, potentially creating an unobservable condition that must be detected algorithmically.

A dynamic extension that incorporates time-series predictions of load and generation as pseudo-measurements to bridge static snapshots. When combined with LAV, the robust criterion can automatically reject a faulty forecast that deviates significantly from reality, preventing a bad prediction from corrupting the estimate. This synergy is powerful for distribution grids with high renewable penetration and volatile net load.