7.2 Eigenvalue analysis and participation factors

Eigenvalues and Eigenvectors of Power Systems

Eigenvalue analysis and participation factors are the primary tools for assessing small-signal stability. They let you identify which oscillatory modes in a power system are well-damped, which are dangerously close to instability, and which specific machines or controllers are driving that behavior. This section covers how to compute and interpret eigenvalues, eigenvectors, participation factors, and their sensitivities to system parameters.

Computing Eigenvalues and Eigenvectors

A linearized power system model takes the form $\dot{x} = Ax$, where $A$ is the state matrix and $x$ is the vector of state variables (rotor angles, speeds, flux linkages, etc.). The eigenvalues and eigenvectors of $A$ fully characterize the system's dynamic response to small disturbances.

Eigenvalues are found by solving the characteristic equation:

$\det(A - \lambda I) = 0$

where $\lambda$ represents the eigenvalues and $I$ is the identity matrix. For a system with $n$ state variables, this yields $n$ eigenvalues (real or complex conjugate pairs).

Eigenvectors satisfy the relationship $Av = \lambda v$, where $v$ is a non-zero vector. In other words, multiplying the state matrix by an eigenvector simply scales that vector by the corresponding eigenvalue.

For realistic power systems (hundreds or thousands of states), you won't solve the characteristic polynomial directly. Instead, numerical methods handle the computation:

• QR algorithm: computes all eigenvalues of the full state matrix; suitable for smaller systems
• Arnoldi iteration (or related sparse methods): selectively computes a subset of eigenvalues, typically those nearest the imaginary axis, which are the most stability-critical; essential for large-scale systems

Interpreting Eigenvectors

Eigenvalues come in two flavors: right eigenvectors and left eigenvectors, and they carry different physical meaning.

• The right eigenvector $v_i$ for mode $i$ describes the mode shape, meaning the relative activity and phasing of each state variable when that mode is excited. If two generators have large, in-phase entries in the same right eigenvector, they swing together in that mode.
• The left eigenvector $w_i$ for mode $i$ describes how each state variable contributes to the excitation of that mode. It tells you which initial conditions most strongly excite mode $i$.

Together, left and right eigenvectors form the basis for participation factor analysis (covered below).

Eigenvalue Significance for Stability

Eigenvalue Properties and Stability

Each eigenvalue $\lambda = \sigma + j\omega$ encodes two pieces of information about its corresponding mode:

• Real part ($\sigma$): determines the damping of the mode
  • $\sigma < 0$: the mode decays over time (stable)
  • $\sigma > 0$: the mode grows over time (unstable)
  • $\sigma = 0$: sustained oscillations with no decay or growth
• Imaginary part ($\omega$): determines the frequency of oscillation, with the oscillation frequency in Hz given by $f = \omega / (2\pi)$

The damping ratio $\zeta$ quantifies how quickly oscillations decay relative to their frequency:

$\zeta = \frac{-\sigma}{\sqrt{\sigma^2 + \omega^2}}$

• $\zeta = 0$: undamped (purely oscillatory, no decay)
• $0 < \zeta < 1$: underdamped (oscillations that gradually decay)
• $\zeta = 1$: critically damped (fastest return to equilibrium without oscillation)
• $\zeta > 1$: overdamped (slow return to equilibrium, no oscillation)

In practice, power system operators often require $\zeta \geq 0.05$ (5%) for all electromechanical modes. Modes with damping ratios below this threshold are flagged as poorly damped and may require corrective action.

Time Constants and Stability Margins

The time constant of a mode tells you how long it takes for the oscillation envelope to decay to about 37% of its initial value:

$\tau = \frac{-1}{\sigma}$

A large $\tau$ (corresponding to a small $|\sigma|$) means the mode decays slowly. Eigenvalues that sit close to the imaginary axis in the complex plane are the most concerning because they represent modes with long time constants, poor damping, and the potential to sustain oscillations or drift toward instability under slightly changed conditions.

Participation Factors for Mode Analysis

Calculating Participation Factors

Eigenvectors alone can be misleading because they depend on the scaling and units of state variables. Participation factors solve this problem by combining left and right eigenvector information into a dimensionless, unit-independent measure.

The participation factor of the $k$-th state variable in the $i$-th mode is:

$p_{ki} = v_{ki} \cdot w_{ik}$

where $v_{ki}$ is the $k$-th entry of the $i$-th right eigenvector and $w_{ik}$ is the $k$-th entry of the $i$-th left eigenvector.

Participation factors are normalized so that:

$\sum_k p_{ki} = 1$

for each mode $i$. This means you can directly compare the relative involvement of different state variables in a given mode.

Interpreting Participation Factors

A high participation factor $p_{ki}$ means state variable $k$ is strongly associated with mode $i$. In practical terms:

• If the rotor speed state of Generator 5 has a participation factor of 0.45 in a 0.8 Hz inter-area mode, that generator is a dominant participant in that oscillation.
• Participation factors let you trace a problematic mode back to specific generators, exciters, or controllers, which is exactly what you need to know when deciding where to install a power system stabilizer (PSS) or tune an existing controller.
• A mode where participation is spread across many generators in different areas is typically an inter-area mode, while a mode dominated by one or two nearby machines is a local mode.

Eigenvalue Sensitivity to Parameters

Calculating Eigenvalue Sensitivities

Once you've identified a critical mode, the next question is: how sensitive is that eigenvalue to changes in system parameters? The sensitivity of eigenvalue $\lambda_i$ with respect to a parameter $\alpha$ (such as a gain, time constant, or loading level) is:

$\frac{\partial \lambda_i}{\partial \alpha} = \frac{w_i^T \left(\frac{\partial A}{\partial \alpha}\right) v_i}{w_i^T v_i}$

where $w_i$ and $v_i$ are the left and right eigenvectors of mode $i$, and $\frac{\partial A}{\partial \alpha}$ is the derivative of the state matrix with respect to the parameter.

Computing this requires:

1. Identify the parameter $\alpha$ of interest (e.g., PSS gain, exciter time constant, line reactance).
2. Compute $\frac{\partial A}{\partial \alpha}$, either analytically or by numerical perturbation of the state matrix.
3. Use the already-computed left and right eigenvectors to evaluate the formula above.

The result is a complex number. Its real part tells you how the parameter affects damping, and its imaginary part tells you how it affects oscillation frequency.

Applying Eigenvalue Sensitivities

Sensitivity analysis has several direct applications:

• Controller tuning: If $\lambda_i$ is highly sensitive to a PSS gain, small adjustments to that gain can significantly improve the damping of mode $i$. Conversely, if sensitivity is low, tuning that parameter won't help much.
• Identifying influential parameters: Ranking parameters by their eigenvalue sensitivity highlights which ones most affect stability. This focuses engineering effort where it matters.
• Robustness assessment: By evaluating sensitivities across a range of operating conditions (light load, heavy load, contingency scenarios), you can determine whether a mode remains well-damped or whether certain operating points push eigenvalues toward the imaginary axis.
• Placement of FACTS devices: Sensitivity analysis can guide where to install thyristor-controlled series compensators, SVCs, or other FACTS devices by identifying locations where parameter changes most effectively shift critical eigenvalues to the left in the complex plane.