08 — Operating Point, DAE Linearization and Model Reduction

To connect OpenHPL with classical control theory, we linearize the nonlinear DAE at a steady operating point.

1. Equilibrium

For

$$F(\dot x,x,z,u)=0,$$

an equilibrium $(x^\star,z^\star,u^\star)$ satisfies

$$F(0,x^\star,z^\star,u^\star)=0.$$

This means constant reservoir/surge conditions, steady penstock pressure and flow, constant turbine power, constant speed and balanced electrical power.

2. Perturbations

$$x=x^\star+\Delta x,\quad z=z^\star+\Delta z,\quad u=u^\star+\Delta u.$$

3. Descriptor linearization

Linearizing the residual gives

$$E\Delta\dot x=A_x\Delta x+A_z\Delta z+B\Delta u,$$ $$0=G_x\Delta x+G_z\Delta z+G_u\Delta u.$$

If $G_z$ is nonsingular locally,

$$\Delta z=-G_z^{-1}G_x\Delta x-G_z^{-1}G_u\Delta u.$$

Substitution gives

$$E\Delta\dot x=A_r\Delta x+B_r\Delta u,$$

where

$$A_r=A_x-A_zG_z^{-1}G_x,$$ $$B_r=B-A_zG_z^{-1}G_u.$$

4. State-space model

If $E$ is nonsingular,

$$\Delta\dot x=A\Delta x+B\Delta u,$$ $$\Delta y=C\Delta x+D\Delta u.$$

5. Transfer function

$$G(s)=C(sI-A)^{-1}B+D.$$

For FCR, two useful mappings are

$$G_{P_my_g}(s)=\frac{\Delta P_m(s)}{\Delta y_g(s)},$$

and the closed-loop frequency response

$$G_{fP_L}(s)=\frac{\Delta f(s)}{\Delta P_L(s)}.$$

6. Why reduction is necessary

A PenstockKP model with $N$ cells introduces roughly $2N$ pressure/flow states. Most classical governor studies do not need every high-frequency hydraulic mode. Model reduction should therefore retain the modes that materially influence FCR bandwidth.

7. Modal view

For the linearized matrix $A$, eigenvalues

$$\lambda_i=\sigma_i+j\omega_i$$

identify water-column, surge-tank, governor and electromechanical modes. Modes far above the control bandwidth can often be truncated if their input-output participation is small.

8. Reduced model

A reduced model should satisfy approximately

$$G_r(j\omega)\approx G(j\omega)$$

over the frequency range relevant to FCR. The goal is not to reproduce every water-hammer detail, but to preserve the hydraulic dynamics that affect stability, droop response and delivered power.

9. Validation

The reduced linear model should be compared against the original nonlinear OpenHPL model for small disturbances around the same operating point:

$$\Delta y_{nonlinear}(t)\approx\Delta y_{linear}(t).$$

Only then should the linear model be used for Bode, Nyquist, root-locus or controller design.