One row per modelling assumption: what it affects, how it was tested, and what the test showed. "Not tested" is a valid entry and means the risk is open.
| ID | Assumption | Affects | Test | Result |
|---|---|---|---|---|
| A1 | Balanced three-phase conditions; dq model with no zero sequence | Everything. Unbalanced faults cannot be studied | Step 07 wrote the current-loop model in phase coordinates and ran a balanced event | The D-Q equations hold for balanced conditions: growth rates within 0.00014 1/s, zero sequence below 2e-9 pu. Unbalanced conditions still not studied |
| A2 | Stator and network flux transients are kept | Existence of the sub-synchronous mode | test_series_rlc_modes_appear_at_f0_minus_and_plus_fer: the network block gives modes at f0 ± f_er with decay R wb / 2X. Step 06 removed the transients from the linearised System B2 |
Holds to 0.02 Hz and 0.2 %. Without them the sub-synchronous mode does not exist: 28 unstable cases of 54 all reported stable, growth up to 28 1/s missed. Required |
| A3 | Line is a lumped series R-L-C; no shunt capacitance, no frequency dependence of R | Resonant frequency and network damping | Not tested. A distributed-line comparison belongs in Phase 2 | Open |
| A4 | Machine circuit parameters from the classical conversion of standard parameters | Rotor resistances, and through them the negative resistance and all damping | Reactances checked at the limits X(0) and X(∞). Time constants not compared with the exact conversion |
Reactances exact. Time-constant error not quantified |
| A5 | Slip calculation averages the d and q operational reactances | Hand estimate of resonance and growth rate | Compared with the locked-shaft eigenvalue of the full electrical model over X_C = 0.02 to 0.62 pu |
Growth rate within 2.6 %, frequency within 0.09 Hz |
| A6 | Damping-torque estimate is first-order in the coupling between electrical and torsional modes | Torsional growth-rate estimate | Compared with full eigenvalues over the same sweep | Within 0.04 1/s where growth is below 0.2 1/s. Overestimates the mode 1 peak by 86 % (7.34 against 3.95 1/s). Peak location agrees |
| A7 | No mechanical damping (D_i = 0) |
Torsional growth rates are worst-case; small positive values appear at every compensation level | By construction. Real units have small positive modal damping | Stated, not tested. Stability margins from this model are pessimistic |
| A8 | Constant field voltage and constant mechanical torque (no AVR, no governor) | Low-frequency swing mode; negligible above 10 Hz | Not tested | Open; expected small for torsional frequencies |
| A9 | No magnetic saturation | Operating-point reactances | Not tested | Open |
| A10 | Fixed-step RK4 at 0.1 to 0.2 ms | Time-domain accuracy | Convergence study against DOP853 at tight tolerance; observed order 4.00 | Error 1.5e-8 at 0.125 ms over 0.5 s. System B2 growth rate (step 06): error 3e-6 1/s at 0.5 ms, 8e-5 at 1 ms, 0.014 at 3 ms; unstable above 4.0 ms |
| A11 | Benchmark data entered from memory of secondary sources | Every number for System A | Shaft data reproduce the five published torsional frequencies to 0.005 Hz, which pins inertias and stiffnesses jointly. Network and generator data have no such check | Shaft: confirmed indirectly. Network resistance split and generator data: to be checked against the 1977 paper |
| A12 | Induction machine data entered from memory of secondary sources (2 MW class, single cage) | Every number for System B1, most of all the boundary through R_r |
Boundary mapped against R_r from 0.004 to 0.030 pu, so the sensitivity is known: 97 % compensation at 0.0055 pu, 73 % at 0.010 pu, 44 % at 0.020 pu |
Sensitivity quantified. Source values to be checked |
| A13 | Single-cage rotor with constant R_r |
Rotor resistance at slip frequency. Double-cage and deep-bar rotors have higher resistance at high slip frequency, which would lower the boundary | Not tested | Open. Direction of the error known: the model is optimistic for such rotors |
| A14 | No shunt capacitors at the plant terminals, no collector system | Operating range and reactive balance. Below about 25 % compensation the line cannot carry 0.9 pu | Observed as a pull-out limit in the steady-state solver | Sweeps start at 30 % for System B1. Step 05 added a plant-side series impedance of 0.010 + j0.15 pu to System B2: the boundary at a 10 Hz loop and rated speed moves from 21 % to 42 % compensation. Shunt capacitors still not modelled |
| A15 | Single lumped mass, constant mechanical torque | Coupling between rotor motion and the network mode | Compared with the speed held: growth rate moves by at most 0.24 1/s, boundary by at most 0.9 points of compensation | Small. A two-mass drive train is not modelled |
| A16 | The plant is one aggregated machine | Everything for Systems B1 and B2 | Step 05 varied the fraction of identical units online from 5 % to 100 % | Not monotonic: worst at 35 to 65 % online, where growth is up to 2.6 1/s above the full plant and the boundary is up to 13 points lower. Unequal units and unequal wind not tested |
| A17 | Slip calculation evaluates the machine at the real frequency jω |
Energy accounting and R_machine for oscillations that are not steady |
Compared with the waveform: 1.4 % for a sustained oscillation, 13 % at −2.4 1/s, 9.5 % at +6.9 1/s. At the complex frequency: at most 2.1 %. With speed held as well: 2e-9 | Holds near the boundary. Away from it, use the complex frequency |
| A18 | Ideal synchronous frame for the rotor current controller: no phase-locked loop, no measurement or switching delay | Controller behaviour near the sub-synchronous frequency; any phase lag changes how much of the gain acts as resistance | Step 04 added the loop and swept it from 1 to 60 Hz at 50 % compensation | Shifts the growth rate by −5.4 1/s at 0.75 pu speed and +1.4 1/s at 1.20 pu at the default 10 Hz. Delays are still not modelled |
| A19 | Constant rotor current references: no outer power or voltage loops; rotor speed is a parameter | Low-frequency behaviour and the dependence on operating point. With these choices the model is linear and its eigenvalues do not depend on dispatch | Step 04 freed the rotor and added outer loops | At most +0.5 1/s at the default 1 Hz outer loop. Not negligible when the outer loop is fast: +13 1/s at 20 Hz and 0.75 pu speed |
| A20 | Quick growth-rate estimate (net resistance over reactance slope) is first order in net resistance | The σ_hand feature |
Compared with the solver against gain: 9.1 against 8.3 1/s at Kp = 0.01, 57.6 against 35.6 1/s at Kp = 0.10 |
Holds for growth below about 10 1/s. The circuit root has no such limit |
| A21 | Controller gain treated as plain rotor resistance (Ki dropped) in the hand calculation |
Hand boundary for System B2 | Largest stable gain compared with the solver over 18 compensation levels and 3 speeds | 1.9 % at the tuned Ki / Kp of 10.9 1/s, 19 % at ten times that, 400 % at a hundred times. Always optimistic |
| A22 | No grid-side converter, no DC link, plant connected radially with nothing else on the bus | Absolute stability limits of System B2 | Step 04 added the grid-side converter and DC link. With voltage feed-forward they move the growth rate at rated speed by 0.6 1/s | The gap to practice is not explained by them. Step 05 tested the network: plant impedance plus one parallel path as strong as the line raises the largest stable current-loop bandwidth at rated speed from 3.5 to 19.2 Hz. Limits remain relative only |
| A23 | Grid-side converter feeds the terminal voltage forward through a 2 ms filter | How much the grid-side converter damps the sub-synchronous mode | Compared with no feed-forward, and swept the filter from 0.5 to 50 ms | Largest effect after the current-loop gain. No feed-forward: −7 1/s and up to 32 points of compensation at the boundary. Slowing the filter to 50 ms: −5 to −7 1/s. Actual plant practice not confirmed |
| A24 | Converters are ideal voltage sources: no switching, no modulation limit, DC voltage does not scale the output | Coupling between the DC link and the oscillation | Not tested | Open |
| A25 | Power follows rotor speed cubed and mechanical power is constant | Operating point at each rotor speed (0.244 pu at 0.75, 1 pu at 1.20) and slow speed dynamics | Not tested against a turbine model | Open; affects low-frequency modes more than the sub-synchronous one |
| A26 | No current, voltage or rate limits | Large-signal behaviour. An unstable case grows without bound | Compensation step of 5 points: growth rate from the waveform within 0.005 1/s of the eigenvalue while the swing is below 0.6 pu | Valid for small oscillations only. Sustained oscillation amplitude cannot be predicted |
| A27 | Plant-side impedance of 0.010 + j0.15 pu and parallel paths with the line's X/R are representative values | Size of the network effects in step 05 | Swept: network resistance 0.5 to 8 times, parallel path 0.25 to 32 times the line impedance | Sensitivity known. Values not tied to a real plant |
| A28 | Network variations are studied with the current-loop model in impedance form | Parallel path and units-online results | Radial case agrees with the state-space model to 2.3e-8. Full controls agree with the current-loop model within 4 points of compensation on the radial case (step 04) | Plant impedance and parallel path since confirmed in the time-domain model: 0.0005 1/s with the added loops off, within 2.3 1/s with full controls. Units online not confirmed in the time domain |
| A29 | Dropping network transients is tested on the linearised model, not with a nonlinear positive-sequence model | The comparison in step 06 | Not tested further | Sufficient for whether the mode exists. Says nothing about large-signal positive-sequence behaviour |
| A30 | The generic converter (System C) is an averaged model: no switching, no control delay, no current or voltage limits, and a single filter reactor with no filter capacitor | Behaviour above a few hundred hertz; large-signal behaviour | Step 28 adds a delay and a filter capacitor: an instability at the filter resonance in a quarter to a third of cases, and the verdict below 58 Hz changed in 1.6 % of cases with a phase-locked loop | Delay and capacitor lifted in step 28 as a separate study; datasets and steps 17 to 27 are without them. Limits still out |
| A37 | The control delay is a second-order Padé approximant per axis, with the frame rotation compensated; the network above 58 Hz is still series R and X with no shunt capacitance | The high-frequency results of step 28 | Step 30 adds shunt capacitance: a delay alone then makes 44.5 % and 49.1 % of cases unstable. Not tested against an exact delay | The inductive network is not safe above 58 Hz. Numbers for the longest delays are open |
| A38 | Shunt capacitance of the network is one lossless capacitor on the plant bus, 0.01 to 0.30 pu, drawn evenly on a logarithmic scale | The shares of unstable cases above 58 Hz in step 30 | Not tested against a distributed line or a cable with losses. The range is a choice, not data | Shows the mechanism and that an inductive network model misses it. The shares are for this choice |
| A39 | Protection of the doubly-fed plant: rotor voltage limit 0.4 pu, rotor current reference limit 1.5 pu, anti-windup, a DC chopper from 1.10 to 1.15 pu; trips read off the run at a DC voltage outside 0.75 to 1.25 pu, rotor current above 2 pu, terminal voltage outside 0.8 to 1.2 pu for 0.1 s, speed outside 0.65 to 1.35 pu. No crowbar, and the plant is not disconnected | Which trip comes first and when (step 33) | Rotor current trip at 1.5, 2 and 3 pu and left out. Round figures, not a manufacturer's | That an unstable case trips and a stable one does not holds at every level tried. Which trip is first, and how well the modes give its time, depend on the levels |
| A31 | The DC side of the converter is a constant power source or sink | The negative conductance of a load and the DC-voltage-loop modes | Not tested. A real source or load behind the DC link has dynamics of its own | Worst case for a load: no relief from the load itself |
| A32 | A converter without a phase-locked loop is the three-phase average of supplies whose current follows the voltage waveform; its voltage measurement filter is ten times faster than that of a converter with a phase-locked loop | The follow family of dataset C1 |
With the slower filter the current lags the voltage by tens of degrees and some converters are unstable on an ideal voltage; with the faster one none is (step 17) | Representative, not tied to a product. Single-phase effects at twice the line frequency are left out |
| A33 | With a phase-locked loop the reactive current reference is the constant that holds the terminal voltage at 1 pu; there is no AC voltage loop | Weak-grid results; a voltage loop adds a mode and changes the admittance at low frequency | Not tested | Open |
| A34 | One aggregate converter at one terminal | Interaction between converters of different kinds on the same network | Step 27: a source and a load on one bus, 65,536 cases. Unit-by-unit screening misses 6 % to 16 % of the unstable pairs | Lifted for two converters on a common bus (step 27) and for a doubly-fed plant beside a converter, at one terminal and a transformer apart (step 32: the reference screen clears no unstable case in 32,767 pairs). Open for more than two units and for a collector network |
| A35 | Scans are taken from the linearised model on a logarithmic grid of 4,001 points from −2,000 to 2,000 Hz; a coarse scan is that grid thinned, and scan error is independent complex noise on each entry | How fine a real scan must be, and what noise does (step 18) | Thinned to 1,001, 251, 101 and 51 points; noise of 5 % and 20 %. A linear grid and systematic error not tested | Spacing figures hold as a percentage of frequency |
| A36 | In the noise study the network impedance is exact and the noise level of the scan is known to the fit | Margins quoted against scan error in step 20 | Network error studied in step 22 with the plant exact: 0.08 1/s per percent in the sub-synchronous band. Scan noise and network error together not tested | Partly closed. The noise level still has to be estimated in practice |