Skip to content

Appendices ​

These appendices collect the supporting theory and practical checklists that the chapters refer to, kept separate so the main narrative stays focused. Each one is short enough to read in a sitting and is written to be consulted again when a study misbehaves.

A. Numerical integration and stability ​

Every reactive element and every control block in an EMT run is a differential equation turned into a recurrence. Which recurrence is chosen decides how accurate the run is and whether it can blow up. The cleanest way to compare the rules is on the scalar test equation

dydt=λy,Reλ<0,

whose exact solution decays as eλt. One step of size h=Δt multiplies the state by an amplification factor R(hλ):

RuleUpdateR(hλ)Order
Forward Euleryn=yn−1+hλyn−11+hλ1
Backward Euleryn=yn−1+hλyn11−hλ1
Trapezoidalyn=yn−1+h2λ(yn+yn−1)1+hλ/21−hλ/22

A rule is stable for a given hλ when |R|≤1.

  • Forward Euler is stable only inside a disc: for real λ=−1/T it needs h<2T. This is the bound Chapter 17 derives for control blocks, and the reason a time constant shorter than half a step diverges.
  • Backward Euler and trapezoidal are both A-stable: |R|≤1 for every λ in the left half-plane, whatever the step. That is why the network can be stepped at a fixed Δt even when it contains time constants far shorter than the step — a stiff circuit.

A-stability is not the whole story. As hλ→−∞ (a very fast mode relative to the step), backward Euler's factor tends to 0: the mode is killed in one step. Trapezoidal's factor tends to −1: the mode is neither damped nor grown, it flips sign every step. That limit is the origin of trapezoidal chatter in Chapter 6 — a current interruption excites exactly such a mode, and the rule faithfully preserves it as a Nyquist-rate oscillation. The standard remedy, critical damping adjustment, switches to backward Euler for a brief window after the discontinuity so the spurious mode dies, then returns to trapezoidal for its accuracy.

Accuracy. The local error per step is O(h2) for the two Euler rules and O(h3) for trapezoidal; summed over the t/h steps of a run, the global error is O(h) and O(h2) respectively. Halving the step therefore halves a first-order rule's error but quarters the trapezoidal error — the scaling Chapter 2 observes when it iterates the inductor companion model by hand. The trapezoidal rule is also symmetric: it introduces no artificial damping on an undamped oscillator, only a small phase (frequency) error, which is the property a transient study wants.

B. The per-unit system ​

Power engineering normalizes quantities to per unit (pu) so that equipment of very different ratings can be compared and connected through transformers without carrying turns ratios around. Choose two independent bases — a three-phase power Sbase (MVA) and a line-to-line voltage Vbase (kV) — and the rest follow:

Zbase=Vbase2Sbase [Ω],Ibase=Sbase3Vbase [kA],Ybase=1Zbase.

A quantity in pu is its physical value divided by its base. For example, at Sbase=100 MVA and Vbase=230 kV, Zbase=529 Ω, so a 52.9 Ω reactance is 0.1 pu.

Change of base. Equipment data is quoted on the equipment's own rating. To move an impedance from an old base to a new one,

Zpu,new=Zpu,oldSbase,newSbase,old(Vbase,oldVbase,new)2.

Why transformers disappear. If each side of a transformer takes its rated voltage as Vbase and both sides share Sbase, the turns ratio is exactly 1 pu and the transformer reduces to its leakage impedance — the property Chapter 13 relies on when it reads the short-circuit test.

In NumaSim. Most numeric fields have a unit selector, and choosing pu adds a Units tab with the component's S_base (MVA), V_base (kV) and Frequency (Hz) (see Parameters & units). Each component has its own bases, so a machine can be entered on its own rating and a line on the system base, side by side. Synchronous machines, induction machines and transformers take their impedance data in pu on their own S rated and rated voltage, exactly as datasheets quote it.

Common pitfalls.

  • Single-phase versus three-phase bases. Using S1ϕ=S3ϕ/3 with VLN=VLL/3 gives the same Zbase, so either is consistent — mixing one of each is not.
  • Percent versus per unit. Nameplates often quote impedance in percent: 10 % is 0.1 pu.
  • Peak versus RMS. Bases are RMS. An EMT waveform's peak phase voltage at 1 pu is 2/3Vbase, not Vbase.
  • A forgotten base change. A generator's 0.2 pu reactance on 50 MVA is 0.4 pu on a 100 MVA system base. Using the number unconverted halves the fault current contribution you expect.

C. Choosing the time step ​

A practical procedure, in the order the constraints usually bind:

  1. List the fastest phenomena you must resolve. A switching frequency fsw, a ringing frequency f0=1/(2πLC), the travel time τ of the shortest travelling-wave line, the fastest control time constant.
  2. Resolve the fastest period. Aim for at least 20 and preferably 50–200 steps per period of the fastest oscillation you care about. For switching, Chapter 6 uses the count N=1/(fswΔt); N=200 is comfortable, and a few tens is the floor before the waveform visibly degrades.
  3. Respect the line ceiling. A travelling-wave line needs Δt≤τ; with a longer step NumaSim represents the line as a nominal π instead (Chapter 9). The delay is rounded to N=round(τ/Δt) steps, a relative error of up to 1/2N, so tens of steps per transit keeps the travel time honest.
  4. Respect the control bound. Every control time constant T must satisfy Δt<2T (Appendix A). A constant much shorter than the step usually means nothing at that resolution; set it to zero rather than shrinking the step to accommodate it.
  5. Check convergence. Halve Δt and re-run. With the trapezoidal network the error should drop by about four; if the waveform you care about does not move, the larger step was adequate.
  6. Budget run time. The number of steps is duration /Δt. When one part of the circuit needs a much finer step than the rest, run that part in a subsystem at a multiple of the outer rate instead of slowing the whole network.

Symptoms of a step that is too large: ringing at the wrong frequency or amplitude, switching events that land late, a travelling-wave line silently modelled as a π, a control block whose output alternates sign and grows. Symptoms of a step that is too small: none in the waveform — only a run that takes far longer than it needs to.

D. Verifying an EMT model ​

A checklist for trusting a result before you act on it:

  1. Hand-check the steady state. Before looking at the transient, compare the settled values against something you can compute: Ohm's law, a voltage divider, τ=L/R or RC, a second-order ω0 and ζ, a load-flow solution. Module 1 works this way throughout.
  2. Check the start. A run should begin either from rest by design or in equilibrium. An unexplained surge in the first cycles usually means a source stepped on at t=0, a machine without a load flow, or a controller without a matching operating point (Chapters 15 and 18). Ramp the sources or initialize from a load flow.
  3. Check that nothing floats. Every electrical island needs a path to ground; the log warns about floating subnetworks.
  4. Check the energy balance. Power in should equal power out plus losses plus the rate of change of stored energy. A source delivering power into a purely reactive network over a full cycle is a modelling error.
  5. Refine the step. Halve Δt and confirm the answer converges (Appendix C).
  6. Compare against an independent method. A phasor calculation for the steady state, a closed-form transient for a simple subcircuit, or a published benchmark for a standard system such as the IEEE 9-bus or two-area cases.
  7. Read the log. Warnings about algebraic loops, fractional line delays and parameter ranges are the simulator telling you where its assumptions were stretched.
  8. Change one thing at a time. When a result surprises you, vary a single parameter and predict the direction of the change before you run. A model you can predict is a model you understand.

References ​

  • H. W. Dommel, Electromagnetic Transients Program (EMTP) Theory Book, Bonneville Power Administration — appendices on ordinary differential equation solution, re-initialization at discontinuities, and per-unit quantities.
  • J. Arrillaga and N. R. Watson, Power Systems Electromagnetic Transients Simulation, IET Power and Energy Series 39 — numerical integration and the analysis of continuous and discrete systems.
  • J. C. Butcher, Numerical Methods for Ordinary Differential Equations, Wiley — stability regions and A-stability.
  • J. D. Glover, T. J. Overbye and M. S. Sarma, Power System Analysis and Design, Cengage — the per-unit system and change of base.

Previous: Module 6 — System studies · Back to the course syllabus.