Sections 1 to 11 cover Phase 1: the series-compensated line, the synchronous machine, the multi-mass shaft, and the two hand calculations (slip and damping torque) that the dynamic model is checked against. Sections 12 to 18 cover the induction generator, the energy accounting, the doubly-fed machine (current loop only, then full controls), the impedance formulation, the meshed network and the phase-coordinate model added in Phase 2. Code references are to src/sso_sciml/.
- Base frequency
f0 = 60 Hz,wb = 2π f0rad/s. - Voltages, currents, fluxes, impedances and torque are in per unit on the generator rating.
- Time is in seconds. Because reactances are per-unit values at base frequency, every flux equation carries a factor
wb:dψ/dt = wb · (voltage terms). - Speed is per unit.
dw_i = ω_i − 1is the speed deviation of massi. - Angles are electrical radians.
Two rotating frames are used.
- Synchronous frame (D, Q), rotating at
wb. The infinite bus voltage lies on the D axis:V_b = v_b + j0. The series capacitor voltage is a state in this frame. - Rotor frame (d, q), fixed to the generator rotor. The q axis leads the d axis by 90°.
δis the angle of the q axis ahead of the D axis.
A complex quantity moves from the synchronous frame to the rotor frame by
(f_d + j f_q) = (f_D + j f_Q) · e^{−j(δ − π/2)}
which gives, for the voltage behind the line with A = v_cD + v_b and B = v_cQ:
v_d = A sin δ − B cos δ
v_q = A cos δ + B sin δ
and for the line current going the other way:
i_D = i_d sin δ + i_q cos δ
i_Q = −i_d cos δ + i_q sin δ
A stator-frame oscillation at frequency f_e appears in either rotating frame at f0 − f_e. A sub-synchronous resonance at 39 Hz is a 21 Hz mode in the model.
Generator convention on the stator: positive current flows out of the machine. Air-gap torque T_e is positive when it decelerates the rotor.
Two rotor circuits per axis: field fd and one damper 1d on the d axis, two dampers 1q, 2q on the q axis.
Flux linkages (d axis; the q axis has the same form with X_aq, 1q, 2q):
ψ_d = −X_d i_d + X_ad i_fd + X_ad i_1d
ψ_fd = −X_ad i_d + (X_ad + X_fd) i_fd + X_ad i_1d
ψ_1d = −X_ad i_d + X_ad i_fd + (X_ad + X_1d) i_1d
Voltage equations, stator transients kept:
(1/wb) dψ_d/dt = v_d + R_a i_d + ω ψ_q
(1/wb) dψ_q/dt = v_q + R_a i_q − ω ψ_d
(1/wb) dψ_fd/dt = v_fd − R_fd i_fd
(1/wb) dψ_1d/dt = −R_1d i_1d (same for 1q, 2q)
Torque: T_e = ψ_d i_q − ψ_q i_d.
The dψ_d/dt and dψ_q/dt terms are the ones a positive-sequence phasor program sets to zero. Without them the network has no dynamics and the sub-synchronous mode does not exist.
machine.axis_from_standard uses the classical relations:
X_ad = X_d − X_l
X_fd = X_ad (X_d' − X_l) / (X_ad − (X_d' − X_l))
X_1d = (X_d'' − X_l) X_ad X_fd / (X_ad X_fd − (X_d'' − X_l)(X_ad + X_fd))
R_fd = (X_ad + X_fd) / (wb T_d0')
R_1d = (X_1d + X_ad X_fd / (X_ad + X_fd)) / (wb T_d0'')
These reproduce X_d, X_d', X_d'' exactly and the time constants approximately (assumption A4).
With the rotor circuits shorted and a rotor-frame complex frequency s (per unit):
X_d(s) = X_d − s X_ad² · 1ᵀ (s L_rr + R_rr)⁻¹ 1
X_d(0) = X_d, X_d(∞) = X_d''. Implemented in AxisCircuit.operational_reactance.
Series R_E, X_E, X_C between the generator terminal and the infinite bus. X_E is the transformer, line and system reactance together.
The line inductance carries the stator current, so it is not an independent state. It is merged into the stator flux:
ψ_d' = ψ_d − X_E i_d ψ_q' = ψ_q − X_E i_q
The stator equations keep their form with ψ → ψ', R_a → R_a + R_E, X_d → X_d + X_E, and the terminal voltage replaced by the voltage behind the line, V_c + V_b. The torque expression is unchanged because the X_E terms cancel.
Capacitor, synchronous frame:
dv_cD/dt = wb (X_C i_D + v_cQ)
dv_cQ/dt = wb (X_C i_Q − v_cD)
A series R-L-C on its own has, in the rotating frame, two mode pairs at f0 − f_er and f0 + f_er, where f_er = f0 √(X_C / X), both decaying at R wb / (2X).
Six masses in a chain: HP, IP, LPA, LPB, GEN, EXC.
dδ_i/dt = wb · dw_i
2 H_i d(dw_i)/dt = T_i − D_i dw_i + K_{i−1}(δ_{i−1} − δ_i) − K_i(δ_i − δ_{i+1})
T_i is the turbine torque share on the four turbine masses, −T_e on the generator, zero on the exciter. K is in per-unit torque per electrical radian.
Natural frequencies come from the eigenvalues of M^{-1/2} K M^{-1/2} with M = diag(2H_i / wb) (shaft.torsional_modes).
Twenty states: ψ_d', ψ_q', ψ_fd, ψ_1d, ψ_1q, ψ_2q, v_cD, v_cQ, δ_1..6, dw_1..6. Defined in physics/fbm.py.
For a given terminal power, power factor and voltage (FBM.equilibrium):
I = conj(S / V_t).V_b = V_t − (R_E + j(X_E − X_C)) I. Rotate everything soV_bis on the D axis.E_Q = V_t + (R_a + jX_q) Ilocates the q axis:δ = arg(E_Q).- Transform
IandV_tto the rotor frame.ψ_d = v_q + R_a i_q,ψ_q = −(v_d + R_a i_d),i_fd = (ψ_d + X_d i_d) / X_ad. Damper currents are zero. v_fd = R_fd i_fd,T_m = T_e,V_c = −jX_C I.- Shaft angles from the torque each section carries.
Implemented in physics/slip.py. It uses only the equivalent circuit.
A stator current at frequency ω_e (per unit, below 1) is seen from a rotor turning at ω_r at the frequency ω_e − ω_r, which is negative. The slip is
s = (ω_e − ω_r) / ω_e
In the rotor frame the stator voltage is v = −(s_r + jω_r) X(s_r) i with s_r = j(ω_e − ω_r), so the stator-frame impedance of the machine is
Z_m(jω_e) = R_a + jω_e · X(j(ω_e − ω_r)), X = (X_d(·) + X_q(·)) / 2
Because the rotor-frame frequency is negative, the imaginary part of X changes sign and the real part of Z_m is negative. That is the negative resistance.
Loop impedance: Z(jω) = R_E + jωX_E − jX_C/ω + Z_m(jω).
- Resonance
ω_er:Im Z = 0. - Net resistance:
R_net = Re Z(jω_er). Self-excitation whenR_net < 0. - Hand growth rate:
σ = −wb · R_net / (d Im Z / dω). For a plain R-L-C this is−R wb / (2X). - Circuit root: Newton iteration on
Z(s) = 0in the complex plane.
Averaging X_d and X_q drops the coupling between sub- and super-synchronous components (assumption A5).
Implemented in physics/torsional.py.
Impose a generator angle oscillation Δδ(s) with Δω = sΔδ / wb on the electrical subsystem and read the torque:
K_e(s) = ΔT_e / Δδ = c (sI − A_ee)⁻¹ (b_δ + (s/wb) b_ω)
A_ee is the 8×8 electrical block of the state matrix; b_δ, b_ω are its coupling columns; c = ∂T_e/∂x_e.
Electrical damping coefficient: D_e(ω) = Im K_e(jω) · wb / ω.
For torsional mode n with shape φ and modal inertia H_n = Σ H_i φ_i², first-order perturbation gives
σ_n = −D_e(ω_n) φ_gen² / (4 H_n)
This holds when the electrical resonance and the torsional mode are weakly coupled (assumption A6).
- Linearisation:
torch.func.jacrevon the right-hand side, float64. - Eigenvalues and participation factors: SciPy on the CPU. Participation of state
kin modeiis|w_ki v_ki|, normalised per mode. - Time integration: classical fourth-order Runge-Kutta, fixed step, batched. References: SciPy Radau and DOP853.
- Mode estimation from a signal: matrix pencil.
Implemented in physics/induction.py. Single-cage rotor, round, so the whole model is written in the synchronous frame as complex quantities f = f_D + j f_Q. No rotor angle is needed.
Generator convention on the stator (current i positive out of the machine), rotor current i_r positive into the rotor. With X_s = X_ls + X_m, X_r = X_lr + X_m:
ψ_s = −X_s i + X_m i_r
ψ_r = −X_m i + X_r i_r
Voltage equations, rotor shorted, rotor speed ω_r:
(1/wb) dψ_s/dt = v_s + R_s i − j ψ_s
(1/wb) dψ_r/dt = −R_r i_r − j (1 − ω_r) ψ_r
The line is merged into the stator flux as in section 5: ψ' = ψ_s − X_E i, R = R_s + R_E, and the stator voltage becomes v_b + v_c:
(1/wb) dψ'/dt = v_b + v_c + R i − j ψ'
(1/wb) dv_c/dt = X_C i − j v_c
Currents from fluxes, in closed form, with X_ss = X_s + X_E and Δ = X_m² − X_ss X_r:
i = (X_r ψ' − X_m ψ_r) / Δ
i_r = (X_m ψ' − X_ss ψ_r) / Δ
Torque and motion (single mass, constant mechanical torque):
T_e = Im(conj(ψ') i) = ψ'_D i_Q − ψ'_Q i_D
2H dω_r/dt = T_m − T_e
State vector: ψ'_D, ψ'_Q, ψ_rD, ψ_rQ, v_cD, v_cQ, ω_r.
Terminal voltage, needed for the energy accounting, includes the inductive drop of the line:
v_s = v_b + v_c + R_E i + X_E ((1/wb) di/dt + j i)
with di/dt taken from the right-hand side through the flux map above.
At slip s0 = 1 − ω_r the loop is the usual equivalent circuit:
Z = R_E + R_s + j(X_E − X_C + X_ls) + jX_m ∥ (R_r/s0 + jX_lr)
i = −v_b / Z, i_r = (v_b / Z) · jX_m / (R_r/s0 + j(X_m + X_lr))
Fluxes follow from the flux map, v_c = −jX_C i, T_m = T_e. For a given power into the infinite bus, s0 is found on the stable side of pull-out (InductionGenerator.equilibrium).
Stator impedance at stator-frame complex frequency s (per unit):
Z_m(s) = R_s + s X_ls + (s X_m) ∥ (R_r / slip + s X_lr), slip = (s − jω_r) / s
At s = jω with ω < ω_r the slip is real and negative, so the rotor branch carries negative resistance. The loop analysis of section 9 is then applied unchanged (slip.analyse_loop).
With ω_r held constant the electrical model is linear with complex coefficients, and its eigenvalues λ are exactly the roots of Z(s) = 0 through s = λ/wb + j. The boundary R_net = 0 is therefore exact for the electrical subsystem, where for the salient synchronous machine it was an approximation.
Implemented in physics/energy.py.
Let Δi and Δv be the deviations of stator current and terminal voltage from the steady state. The power the machine delivers through the oscillation itself is
p_osc = Re(Δv · conj(Δi))
This is second order in the perturbation. The first-order cross terms with the steady state average out over a cycle and are left out by construction.
If the machine presents impedance Z_m to the oscillation then Δv = −Z_m Δi in generator convention, and
p_osc = −Re(Z_m) |Δi|²
so the resistance implied by a waveform is R_eff = −∫ p_osc dt / ∫ |Δi|² dt. Energy lost in the line over the same window is R_E ∫ |Δi|² dt. The oscillation is sustained when the two are equal.
For an oscillation that grows or decays at rate σ, Z_m must be evaluated at the complex frequency σ/wb + jω, not at jω. The difference is the rate of change of stored magnetic energy. It was 13 % of the resistance at a decay rate of 2.4 1/s (docs/steps/02-induction-generator.md).
Implemented in physics/dfig.py. The machine and line equations are those of section 12, with a rotor voltage v_r from the converter:
(1/wb) dψ_r/dt = v_r − R_r i_r − j (1 − ω_r) ψ_r
The controller acts on rotor current in the synchronous frame:
v_r = v_ff + Kp (i_ref − i_r) + Ki x
dx/dt = i_ref − i_r
Kp is in per unit, Ki in per unit per second, x in per unit seconds. v_ff is a constant feed-forward, used only to hold the steady state when Ki = 0.
At this stage the frame is ideal, i_ref is constant, there is no grid-side converter, and ω_r is a parameter. State vector: ψ'_D, ψ'_Q, ψ_rD, ψ_rQ, v_cD, v_cQ, x_D, x_Q. The model is linear in the state.
For stator power P + jQ into the infinite bus: i = (P − jQ) / v_b, v_c = −jX_C i, ψ' = −j (v_b + v_c + R i), i_r = (ψ' + X_ss i) / X_m, ψ_r from the flux map, v_r = R_r i_r + j (1 − ω_r) ψ_r. Then i_ref = i_r and x = v_r / Ki.
Pole cancellation on the rotor transient reactance σX_r = X_r − X_m² / X_s, for a closed-loop bandwidth α in rad/s:
Kp = α σX_r / wb Ki = α R_r
so Ki / Kp = R_r wb / σX_r, 10.9 1/s for the base machine.
For a perturbation, Δv_r = −(Kp + Ki / S) Δi_r with S = wb (s − j) the synchronous-frame frequency in 1/s and s the per-unit stator-frame frequency. The rotor equation becomes that of a shorted rotor with resistance
R_eff(s) = R_r + Kp + Ki / (wb (s − j))
and the machine impedance is that of section 12 with R_r → R_eff (slip.dfig_impedance). This is exact for the model above.
Kpis added rotor resistance. At negative slip it adds negative resistance at the stator.- At
s = jωtheKiterm is−j Ki / (wb (1 − ω)): imaginary, so after division by slip it is reactive. It shifts the resonance and does not add resistance to first order. It grows without bound asω → 1, which creates extra zeros of loop reactance near synchronous frequency. The resonance search therefore takes the lowest crossing from capacitive to inductive.
The hand approximation "Kp as rotor resistance" drops the Ki term. Its error against the solver is 2 % at the tuned ratio and grows with Ki / Kp (docs/steps/03-dfig-current-control.md).
Implemented in energy.flows. With Δ denoting deviations from the steady state and motor-convention stator current −Δi:
to network = ∫ Re(Δv_s conj(Δi)) dt
from shaft = ω_r ∫ Im(conj(Δψ') Δi) dt
from converter = ∫ Re(Δv_r conj(Δi_r)) dt
copper = R_s ∫ |Δi|² dt + R_r ∫ |Δi_r|² dt
stored = [ Re(−conj(Δi) Δψ_s + conj(Δi_r) Δψ_r) / (2 wb) ] end − start
from shaft + from converter = to network + copper + stored
Each term is second order, a product of two deviations. The balance follows from the voltage equations because the inductance matrix is real and symmetric.
Implemented in physics/dfig_full.py. Complex quantities are in the synchronous frame unless marked with a superscript c, which means the phase-locked-loop frame: f^c = f · e^{−jθ}.
Three inductive branches meet at the terminal: the stator, the grid-side filter (R_f, X_f, current i_g from converter to terminal) and the line, which carries i_L = i + i_g. The terminal voltage is eliminated by using two loop fluxes as states:
λ1 = ψ_s − X_E i_L (stator plus line)
λ2 = X_f i_g + X_E i_L (grid-side filter plus line)
(1/wb) dλ1/dt = v_b + v_c + R_s i + R_E i_L − j λ1
(1/wb) dλ2/dt = v_g − v_b − v_c − R_f i_g − R_E i_L − j λ2
(1/wb) dv_c/dt = X_C i_L − j v_c
With a = X_E / (X_f + X_E), the combination λ1 + a λ2 plays the part of ψ' in section 12 with X_E replaced by X_E X_f / (X_f + X_E). That gives i and i_r in closed form, then i_g = (λ2 − X_E i) / (X_f + X_E).
Terminal voltage, from the grid-side branch:
v_t = v_g − R_f i_g − X_f ((1/wb) di_g/dt + j i_g)
The current derivatives come from the flux derivatives through the same linear map. Every controller acts on states, filtered quantities or currents, so the right-hand side has no algebraic loop.
dθ/dt = Kp_pll · v_tq^c + Ki_pll · x_pll
dx_pll/dt = v_tq^c
v_tq^c is the q component of terminal voltage in the loop's own frame; it is zero when the frame is aligned. Gains from a natural frequency ωn and damping ζ: Kp_pll = 2ζωn, Ki_pll = ωn².
P_ref = K_opt ω_r³
dP_f/dt = (P_t − P_f) / T_f (same for Q_f)
i_rd,ref = Kp_o (P_ref − P_f) + Ki_o x_P dx_P/dt = P_ref − P_f
i_rq,ref = −(Kp_o (Q_ref − Q_f) + Ki_o x_Q) dx_Q/dt = Q_ref − Q_f
v_r^c = Kp (i_r,ref − i_r^c) + Ki x_r dx_r/dt = i_r,ref − i_r^c
P_t + jQ_t = v_t · conj(i_L). With the terminal voltage on the d axis, d-axis rotor current sets active power and q-axis rotor current sets reactive power; the minus sign is because more reactive power needs less q-axis rotor current.
i_gd,ref = −(Kp_dc (v_dc,ref − v_dc) + Ki_dc x_dc) dx_dc/dt = v_dc,ref − v_dc
i_gq,ref = 0
v_g^c = ff · v_f^c + Kp_g (i_g,ref − i_g^c) + Ki_g x_g
dv_f^c/dt = (v_t^c − v_f^c) / T_ff
C_dc v_dc dv_dc/dt = P_rsc − P_gsc
P_rsc = −Re(v_r conj(i_r)) P_gsc = Re(v_g conj(i_g))
ff is 1 with terminal-voltage feed-forward and 0 without. Converter voltages are taken as ideal: the DC voltage does not limit or scale them.
Without feed-forward, a perturbation sees Δv_g^c = −(Kp_g + Ki_g/S) Δi_g^c: the converter is a resistance Kp_g behind X_f, connected across the terminals. With fast feed-forward the terminal voltage perturbation is cancelled and the converter approaches a current source.
2H dω_r/dt = P_m / ω_r − T_e, T_e = Im(conj(ψ_s) i), ψ_s = λ1 + X_E i_L
For a rotor speed ω_0: P = K_opt ω_0³. Solve (v_b + (R_E + j(X_E − X_C)) i_L) conj(i_L) = P + jQ_ref for i_L; then v_t and θ = arg(v_t). The grid-side current is i_g = i_gd e^{jθ} with i_gd chosen so that P_rsc = P_gsc. The rest follows as in section 14, and P_m = T_e ω_0.
λ1 (2), ψ_r (2), v_c (2), λ2 (2), ω_r, θ, x_pll, x_r (2), P_f, Q_f, x_P, x_Q, x_g (2), v_dc, x_dc, v_f (2): 23 states.
Implemented in physics/circuit.py. It gives the modes of the current-loop model of section 14 without a state vector.
Every element is a ratio of polynomials in the per-unit stator-frame frequency s:
series branch: Z = R + sX + X_C / s
parallel: Z_a ∥ Z_b = Z_a Z_b / (Z_a + Z_b)
machine: Z_m(s) = R_s + sX_ls + (sX_m) ∥ (R_eff(s) · s / (s − jω_r) + sX_lr)
R_eff(s) = R_r + Kp + (Ki / wb) / (s − j)
The modes are the zeros of the total impedance seen around the loop:
Z_m(s) / n + Z_net(s) = 0, λ = wb (s − j)
n is the fraction of units online: n identical machines in parallel on a fixed base have impedance Z_m / n.
Networks used:
radial: Z_net = R_E + sX_E + X_C / s
with plant impedance: Z_net = (R_E + sX_E + X_C / s) + (R_plant + sX_plant)
with a parallel path: Z_net = (R_E + sX_E + X_C / s) ∥ (ρR_E + sρX_E)
Each root s gives one eigenvalue of the synchronous-frame model; the state-space model has that eigenvalue and its conjugate.
Building the total impedance by cross-multiplication leaves common factors in numerator and denominator (at s = 0, s = j and s = jω_r). Each denominator root removes the numerator root nearest to it when the two are within 1e-5. A true mode can sit near a pole when the gain is low, so the test is on distance between roots and not on the size of the denominator.
Implemented in physics/dfig_meshed.py. Controls and machine as in section 15. Network:
terminals — (R_t, X_t) — plant bus —+— compensated line (R_E, X_E, X_C) —+— infinite bus
+— parallel path (R_p, X_p) ————————+
i_L = i + i_g flows through the plant impedance. i_p flows in the parallel path and i_c = i_L − i_p in the compensated line. The plant bus voltage is eliminated with a third loop flux:
λ1 = ψ_s − X_t i_L − X_E i_c
λ2 = X_f i_g + X_t i_L + X_E i_c
μ = X_p i_p − X_E i_c
(1/wb) dλ1/dt = v_b + v_c + R_s i + R_t i_L + R_E i_c − j λ1
(1/wb) dλ2/dt = v_g − v_b − v_c − R_f i_g − R_t i_L − R_E i_c − j λ2
(1/wb) dμ/dt = v_c + R_E i_c − R_p i_p − j μ
(1/wb) dv_c/dt = X_C i_c − j v_c
Fluxes and currents are related by
[λ1] [−(X_s+X_t+X_E) −(X_t+X_E) X_m X_E ] [i ]
[λ2] = [ X_t+X_E X_f+X_t+X_E 0 −X_E ] [i_g]
[ψ_r] [ −X_m 0 X_r 0 ] [i_r]
[μ ] [ −X_E −X_E 0 X_p+X_E ] [i_p]
solved numerically at each evaluation. Stator flux for the torque: ψ_s = λ1 + X_t i_L + X_E i_c.
Steady state: both branches end at the infinite bus, so the network seen from the terminals is v_b behind Z = R_t + jX_t + Z_c Z_p / (Z_c + Z_p) with Z_c = R_E + j(X_E − X_C) and Z_p = R_p + jX_p. Then i_c = i_L Z_p / (Z_c + Z_p), and the rest follows section 15.
State vector: the 23 states of section 15 plus μ_D, μ_Q.
Implemented in physics/dfig_abc.py as a check on section 14. Each quantity is a vector of phase a, b, c values in the stationary frame. With θ = wb t and phase offsets φ = (0, 2π/3, −2π/3):
x_phase,k = x_D cos(θ − φ_k) − x_Q sin(θ − φ_k)
x_D = (2/3) Σ_k x_phase,k cos(θ − φ_k)
x_Q = −(2/3) Σ_k x_phase,k sin(θ − φ_k)
The rotating-frame terms −jψ of section 14 disappear from the stator and capacitor equations. The rotor flux, referred to the stationary frame, keeps a speed term:
(1/wb) dψ_s/dt = v_b cos(θ − φ) + v_c + (R_s + R_E) i
(1/wb) dψ_r/dt = v_r − R_r i_r + ω_r J ψ_r
(1/wb) dv_c/dt = X_C i
J = (1/√3) [[0, −1, 1], [1, 0, −1], [−1, 1, 0]] is a 90° rotation of the space vector written in phase values. Currents follow from fluxes phase by phase with the same reactances as section 12.
The controller works in the synchronous frame: it turns i_r into D and Q with the angle θ, applies v_r = v_ff + Kp (i_ref − i_r) + Ki x with dx/dt = i_ref − i_r, and turns v_r back into phase values. The system is therefore linear with periodic coefficients, dx/dt = A(t) x + b(t).
Integration is by the trapezoidal rule:
(I − h/2 A_{n+1}) x_{n+1} = (I + h/2 A_n) x_n + h/2 (b_n + b_{n+1})
A mode at frequency f in the rotating frame appears at f0 − f in a phase quantity.
No machine. A voltage-source converter behind a filter reactor R_f + jX_f meets the network of section 17 at its terminal. Current i is positive out of the converter. R_s = R_f + R_t, X_s = X_f + X_t.
Network, with i = i_l + i_p (line and parallel path):
(X_s/wb) di/dt + (X_E/wb) di_l/dt = v_g − v_b − (R_s + jX_s) i − (R_E + jX_E) i_l − v_c
(X_s/wb) di/dt + (X_P/wb) di_p/dt = v_g − v_b − (R_s + jX_s) i − (R_P + jX_P) i_p
(1/wb) dv_c/dt = X_C i_l − j v_c
v_t = v_g − (R_f + jX_f) i − (X_f/wb) di/dt
Controls, in the control frame (lower case, turned by the angle θ):
u = −[Kp_dc (v_dc,ref − v_dc) + Ki_dc x_dc] dx_dc/dt = v_dc,ref − v_dc
C_dc v_dc dv_dc/dt = P_src − Re(v_g i*)
i_ref = u + j i_q,ref with a phase-locked loop
i_ref = u v_f without one: the current follows the voltage
v_g = k_ff v_f + Kp (i_ref − i) + Ki x dx/dt = i_ref − i
dθ/dt = Kp_pll v_t,q + Ki_pll x_pll dx_pll/dt = v_t,q
T_f dv_f/dt = v_t − v_f (− j wb T_f v_f without a phase-locked loop)
P_src is constant: positive for a source, negative for a load. With a phase-locked loop i_q,ref is the constant that holds |v_t| at 1 pu in steady state. Without one θ = 0 and the filter acts on the 60 Hz waveform, so its equation carries the rotation term.
Admittance. The converter alone, driven by v_t, has 10 states (8 without a phase-locked loop). With A, B its Jacobians with respect to state and v_t, and C picking the current:
Y(s) = −C (sI − A)^-1 B current into the converter per terminal voltage
Network impedance. With z(σ) = R_t + σX_t + (R_E + σX_E + X_C/σ) ∥ (R_P + σX_P):
z± = z(s/wb ± j) Z(s) = [[z_d, −z_q], [z_q, z_d]]
z_d = (z+ + z−)/2 z_q = (z+ − z−)/(2j)
Whole system. det(I + Y(s) Z(s)) = 0 at every mode. The number of unstable modes is the number of unstable poles of Y and Z apart, less the number of anticlockwise turns of det(I + Y(jω) Z(jω)) about zero as ω runs from −∞ to ∞.
Passivity. Y is passive at ω if the smallest eigenvalue of (Y + Y^H)/2 is not negative. Z of this network is passive at every frequency. If at every frequency either Y is passive or the largest singular value of Y Z is below 1, no eigenvalue of Y Z reaches the real axis at or to the left of −1, and the system is stable.
Positive-sequence part. m_p = (m_dd + m_qq)/2 + j (m_qd − m_dq)/2 of a 2 × 2 matrix. It leaves out the coupling to the mirror frequency.
Loop eigenvalues (step 18). Let λ_k(ω) be the eigenvalues of Y(jω) Z(jω). If plant and network are each stable alone, the system is stable if no λ_k is real and at or below −1 at any frequency. Two sufficient tests follow. Small gain: |λ_k| < 1 wherever Y is not passive (implied by the singular-value test above). Gain margin with a guard φ: |λ_k| < 1 wherever |arg(−λ_k)| < φ. With φ = 0 and a continuous scan this is the condition itself; a finite guard covers the gaps between scan points.
Doubly-fed plant as an admittance. The plant of section 15 on a line of zero impedance is the plant on an ideal voltage. The capacitor voltage then enters the flux equations exactly as a change of terminal voltage does, so the columns of the state matrix for vc_D, vc_Q are B, those two states are dropped, and C gives minus the line current. The state is set from the steady state of the case: stator flux, X_f i_g and v_t − V_b in place of the two loop fluxes and the capacitor voltage.
Rational fit (step 19). D(s) = det(I + Y(s) Z(s)) is a ratio of polynomials. From samples D(jω_k) the AAA algorithm builds r(s) = Σ w_j f_j/(s − s_j) / Σ w_j/(s − s_j) on a subset s_j of the scan points, adding points until |r − D| is below a tolerance at all of them. The roots of its numerator are the modes of the system; the largest real part is the growth rate. On noisy data the tolerance is set to the noise level.
Least-squares fit and closing on the network (step 20). Each entry of the admittance is fitted as Y_ij(s) = Σ_m c_ij,m φ_m(s) + d_ij with partial fractions φ_m on poles shared by the four entries. The poles come from the vector-fitting iteration: solve Σ c φ + d − Y Σ c̃ φ = Y in the least-squares sense with errors weighted by 1/|Y|, replace the poles by the zeros of 1 + Σ c̃ φ, repeat. Each pole p then has a 2 × 2 residue R = U S V^H, and each singular value is one state: dx/dt = p x + s v^H v_t, current u x. A state is kept if s / (|jω_k − p| ‖Y(jω_k)‖) exceeds three times the noise level at some scan point ω_k. The kept plant dx/dt = A x + B v_t, i = C x + D v_t is closed on the network with unknowns x, line current i_c, path current i_p, capacitor voltage v_c, plant-bus voltage v_m and v_t:
(X_E/wb) di_c/dt = v_m − (R_E + X_E J) i_c − v_c dv_c/dt = wb (X_C i_c − J v_c)
(X_P/wb) di_p/dt = v_m − (R_P + X_P J) i_p
0 = i_c + i_p + C x + D v_t
(X_t/wb) d(i_c + i_p)/dt = v_t − v_m − (R_t + X_t J)(i_c + i_p)
J = [[0, −1], [1, 0]]. The modes are the finite generalised eigenvalues of this set.