Observability Analysis

A power system is topologically observable if a spanning tree of the network graph can be formed where every branch is incident to at least one node with a known voltage phasor. This method relies purely on the incidence matrix and Kirchhoff's laws, ignoring branch impedances. It is computationally faster than numerical methods and is used for initial PMU placement screening. A system that fails topological observability may still be numerically observable if sufficient measurement redundancy exists.

Numerical observability is determined by the rank of the measurement Jacobian matrix. The system is observable if the gain matrix G = H^T * W * H is non-singular (full rank). Unlike topological methods, this accounts for actual branch impedances and measurement weights. Key indicators include:

• Condition number of the gain matrix
• Pivot analysis during triangular factorization
• Zero pivots indicate unobservable branches or islands

The Optimal PMU Placement (OPP) problem seeks to minimize the number of PMUs while achieving full system observability. This is typically formulated as an Integer Linear Programming (ILP) problem:

• Objective: Minimize sum of PMU installation costs
• Constraint: Each bus must be observed by at least one PMU (direct or via adjacent bus)
• Advanced formulations incorporate N-1 contingency constraints, ensuring observability persists after any single PMU or line failure

When a system is unobservable, the depth of unobservability quantifies the extent of the problem. It identifies unobservable branches and unobservable buses that cannot be estimated from available measurements. The system is decomposed into observable islands, and the metric reports:

• Number of unobservable branches
• Maximum distance from an observable bus to an unobservable one
• Pseudo-measurement requirements to restore observability

A critical measurement is one whose removal makes the system unobservable. Similarly, a critical set is a group of measurements where removing the entire set causes loss of observability, but removing any single member does not. Identifying these is essential for:

• Bad data detection vulnerability assessment
• Redundancy planning
• Understanding that critical measurements have zero residual sensitivity, making errors undetectable in state estimation

N-k observability analysis ensures the system remains observable after the loss of k PMUs or communication links. This is critical for Remedial Action Schemes (RAS) and wide-area control. The analysis evaluates:

• Single PMU loss (N-1): Most common requirement
• Communication channel failure: Loss of PDC data streams
• Substation blackout: Simultaneous loss of all PMUs at a site
• Solutions often require redundant PMU placement beyond the minimal set

The Observability Gramian is a symmetric, positive semi-definite matrix that quantifies the energy of the system output. If the Gramian is full rank, the system is completely observable. In PMU placement, the numerical conditioning of the Gramian directly impacts the accuracy of state estimation, with a small condition number indicating a robust sensor configuration.

A network topology where the number and location of PMUs are insufficient to solve for all bus voltage phasors. This occurs when pseudo-measurements cannot compensate for missing real-time data. Key characteristics include:

• Unobservable branches: Lines where current flow cannot be calculated
• Unobservable islands: Clusters of buses isolated from direct measurement
• Critical measurements: Single points whose loss collapses observability

Distinct from topological observability, this checks if the measurement Jacobian matrix is well-conditioned for numerical solution. A system may be topologically observable but numerically unobservable if PMU placements create ill-conditioned gain matrices. This leads to convergence failures in the Weighted Least Squares (WLS) estimator despite theoretical solvability.

The combinatorial problem of minimizing the number of PMUs while achieving full system observability. Formulated as an Integer Linear Programming (ILP) problem with constraints ensuring each bus is observed by at least one PMU. Advanced formulations incorporate:

• N-1 contingency: Maintaining observability after any single branch or PMU loss
• Zero-injection bus rules: Leveraging Kirchhoff's Current Law to merge observable islands
• Channel limits: Accounting for finite PMU measurement channels

Graph-theoretic criteria used to determine observability without solving equations. The fundamental rules are:

• Direct measurement rule: A bus with a PMU is directly observable
• Branch rule: If one terminal bus voltage is known and branch current is measured, the opposite terminal is observable
• Zero-injection rule: If all but one adjacent bus to a zero-injection node are observable, the remaining bus becomes observable via KCL

A metric quantifying the severity of unobservability by measuring the geodesic distance from unobservable buses to the nearest observable node. A high depth indicates large pockets of the network are invisible to operators. This metric guides incremental PMU deployment by prioritizing locations that minimize the maximum depth-of-unobservability across the system.