Wide-Area EMT Studies · Large-Scale Simulation

Large-Scale EMT Simulation

The wide-area EMT mini-series asked first whether a model is adequate, then whether it has been validated. This closing guide asks the execution question: once the models are trusted, how do you build, initialise, reduce, couple and run a wide-area EMT case? Modelling one inverter is a bounded problem; building a whole system — and running it fast enough to be useful — is a discipline of its own, and the value of a wide-area case is that its detailed local area, reduced external system, initial operating point and simulation interface all represent the same physical event. It covers load-flow initialisation, boundary selection, network equivalencing, co-simulation, hybrid PDT–EMT and real-time simulation. It follows CIGRE Technical Brochure 881, with APS engineering interpretation, for EMTP® and other EMT tools.

Reading time ≈ 25 min · Build, initialise, boundaries, equivalence, co-simulate & real-time

The previous guide explained acceptance and validation — how an EMT model earns trust through SMIB tests, weak-grid checks, field validation, disturbance replay and hardware-in-the-loop or control-replica evidence. This page moves from individual-model trust to system-scale execution. Once the plant, converter, protection and network models are adequate and validated, the next challenge is building a large EMT case that starts from the correct operating point, represents the right part of the network, equivalents the rest without distorting the answer, and runs fast enough to be useful. The value of a wide-area EMT case is not only that it is detailed; it is that the detailed local area, the reduced external system, the initial operating point and the simulation interface all represent the same physical event.

The last of three wide-area EMT guides

This closes the mini-series that follows the nine-part IBG modelling series. The first guide asked is the model detailed enough? The second asked has it been tested and validated enough to trust? This guide asks the execution question: once the models are adequate and validated, how do we build, initialise, reduce, couple and run a wide-area EMT case? And model adequacy still matters here — a perfect network equivalent cannot rescue an inadequate local converter model, and a validated converter model can still give the wrong answer if the surrounding network is initialised or equivalented incorrectly.

By the end of this page, the reader should understand

what makes a large-scale EMT study different from a single-plant study, and the two ways to build a wide-area EMT model;
why the EMT initial condition must match the reference load-flow, and how synchronous machines and IBGs are initialised;
what operating zones and measurement points mean, and how boundary buses are selected;
when a network can be reduced or equivalented, and the difference between a static voltage source, a controlled dynamic source, a synchronous-machine equivalent and an FDNE;
why static equivalents can distort frequency response;
what co-simulation means, why splitting at a natural time delay improves convergence, and what hybrid PDT–EMT simulation is;
why EMT–EMT co-simulation helps with speed and IP security, and what real-time simulation, HIL and SIL mean.

What large-scale EMT simulation is

A large-scale EMT simulation is an electromagnetic-transient time-domain model that includes not just one converter or one plant, but a substantial part of the surrounding power system: generators, IBGs, HVDC, FACTS, transformers, lines, cables, protection, loads and network equivalents. The challenge is no longer only converter detail — it is matching detail, size, initialisation, boundary representation and simulation speed. Carrying the theme of the whole mini-series: on the adequacy page a model that runs is not necessarily adequate; on the validation page it is not necessarily validated; here, a large case that runs is not necessarily correctly initialised, equivalented or coupled.

Why large-scale EMT is difficult

EMT uses small time steps, and hundreds of components may be included;
many vendor models may be black-boxed, and each converter may impose its own time-step limit;
the model must start from a valid load-flow condition;
network reduction can change the frequency, voltage and impedance response;
co-simulation interfaces can introduce delay or instability; and
protection and control actions can change the topology during the run.

A note on IBG and IBR

The APS series uses IBG (inverter-based generation). This page uses IBR (inverter-based resource) where it follows CIGRE TB 881 terminology. The two are closely aligned, but IBR is slightly broader — it includes generation, storage and other converter-interfaced resources such as HVDC and FACTS. Read them as the same idea here.

Abbreviations used on this page

EMTElectromagnetic transient (time-domain) simulation

PDTPhasor-domain transient (the RMS / phasor method)

RMSRoot-mean-square (phasor) simulation

EMTP®Electromagnetic Transients Program (an EMT tool)

IBG / IBRInverter-based generation / resource

OEMOriginal equipment manufacturer (the vendor)

SGSynchronous generator (machine)

PoC / PCCPoint of connection / common coupling

SCRShort-circuit ratio (system strength)

X/RReactance-to-resistance ratio

FDNEFrequency-dependent network equivalent

SSCISub-synchronous control interaction

SMIBSingle machine – infinite bus

HVDCHigh-voltage direct current

FACTSFlexible AC Transmission System device

SVC / STATCOMStatic var compensator / static synchronous compensator

PLLPhase-locked loop

DERDistributed energy resource

HIL / SILHardware- / software-in-the-loop

RTS / DRTS(Digital) real-time simulator

SCADASupervisory control and data acquisition

NEMThe Australian National Electricity Market

FFTFast Fourier transform

DLLDynamic-link library (a compiled model interface)

SSNState-space nodal (solver)

Key idea

A large-scale EMT model is built by hand or by importing a phasor-domain (PDT) case and replacing the study-critical parts with true EMT models — and either way its post-initialisation steady state must match a reference load-flow, control states included, not just P and Q.
Boundary placement decides the answer. The retained EMT area must contain the mechanism being studied; the external equivalent must reproduce how the rest of the grid pushes back — and a static source, held too stiff, can falsely restore frequency and hide a real problem.
Co-simulation, hybrid PDT / EMT and real-time simulation make the intractable tractable — but each interface can add delay, instability or a phasor conversion that removes the very phenomenon under study, so the interface must be tested, not assumed.
A wide-area EMT case is trustworthy only when its models are adequate, its validation evidence is sufficient, its initial condition is correct, and its boundaries do not remove the phenomenon being studied.

Key terms used on this page

01Load-flow initialisation

Starting the time-domain run from a load-flow solution so it reaches steady state quickly, matching a reference operating point.

02Flat run

A no-disturbance run at the intended operating point; the model should stay quiet, with no drift or artificial oscillation, before any event.

03Boundary bus

The interface between the detailed retained EMT network and the external system equivalent.

04Retained area

The part of the network modelled explicitly in EMT; it must contain the mechanism being studied.

05External equivalent

The simplified representation of the rest of the system; it must reproduce how the wider grid pushes back at the boundary.

06Static voltage source

A fixed-magnitude, fixed-angle source behind an impedance; simple, but it does not re-adjust its flow after a disturbance.

07Controlled dynamic source

A source that adjusts voltage, angle or frequency by a control law or measured response — as good as its logic and data.

08Synchronous-machine equivalent

One or more equivalent machines (inertia, governor, exciter, impedance) representing the external grid’s electromechanical response.

09FDNE

Frequency-dependent network equivalent: represents the external impedance across a band of frequencies, not only at 50 / 60 Hz.

10Bergeron line

A travelling-wave line model whose natural propagation delay can be used to split EMT sub-systems.

11Co-simulation

Two or more simulation engines running together and exchanging signals at defined interfaces.

12Hybrid PDT–EMT

Keeping the local area in EMT and the wider grid in PDT / RMS, converting quantities at the boundary.

13Real-time simulation

Solving each time step within its own wall-clock duration, so the model runs at the speed of the real system.

14Negative load

Generation represented as a reduction in demand; useful for load-flow construction, but carrying no dynamic converter behaviour.

15HIL / SIL

Hardware- / software-in-the-loop: real control hardware, or its actual code, connected to a real-time simulated grid.

Section 1

Two ways to build the model

There are broadly two routes to a large-scale EMT model, and the choice shapes everything after it.

Method 1 — manual, one-to-one construction. Build the EMT model directly from equipment data, datasheets, control diagrams and vendor (OEM) models. It is accurate and gives the most complete representation — and for detailed work such as system restoration, temporary-overvoltage or harmonic studies it is the only adequate route — but it is slow and labour-intensive, so it suits smaller networks or special studies.
Method 2 — import from a PDT / RMS case. Start from an existing phasor-domain load-flow / dynamic case, convert the network into EMT form, replace the study-critical components with true EMT or vendor models, and check that the EMT steady state matches the reference load-flow. It is the common route for wide-area work, because a trusted PDT case already exists — but the import cannot invent detail the PDT model never held (a line arrives as a PI-section).

For wide-area studies the second method is usually the more practical, but it creates a major quality-assurance requirement: the imported EMT case must still represent the intended operating point.

Section 2

The load-flow match

After initialisation, the EMT model should sit close to the reference load-flow (a historical SCADA snapshot or a planning case) before any disturbance is applied. The bus voltages, angles, active and reactive power flows, generator outputs, transformer taps, shunt status and plant-controller references should not drift away from the planned case. If the model begins with artificial oscillations or mismatched P / Q flows, the simulated disturbance response may be contaminated before the event even starts.

Before running the disturbance, check…

the bus voltage magnitudes and angles; the active and reactive power flows; the frequency;
the synchronous-generator P / Q and terminal voltage; the IBG P / Q, voltage-control mode and current limits;
the transformer taps; the shunt / capacitor / reactor status; the HVDC / FACTS operating point; the load and DER values;
no unexpected controller drift; no protection flags active; no artificial oscillations.

Operating zones and measurement points

An operating zone is the part of the network or dispatch condition being represented for a specific study case — which generators, loads, network elements and conditions are in service. It matters because each zone may need different retained network detail, boundary equivalents and model validation. Measurement points are the buses or branches where the response will be checked — voltages, currents, active / reactive power, frequency, protection actions, converter signals or boundary flows. Choose them before the simulation, so the study is not judged only by whatever plots happen to look interesting afterwards.

One import hazard deserves a note. A negative load is a simplified way to represent generation as a reduction in demand; it can be useful for initial load-flow construction or distant background resources, but it cannot represent dynamic converter response, protection, ride-through, frequency response, current limit or reconnection — the very behaviours the frequency-stability and model-adequacy guides warned against. In EMT a negative load can even translate into a fictitious voltage source; replace it with a proper generator representation, or remove it.

Section 3

Initialising machines and inverters

A synchronous machine (SG) is normally initialised from the load-flow solution: the terminal voltage, active power, reactive power, rotor angle, excitation and governor states must all be consistent. If they are not, the machine can produce artificial oscillations before the disturbance. IBG initialisation is more delicate, because the converter controls, the PLL, the current references, the dc-link state, the plant controller and the protection timers must all start in a mutually consistent condition — and a black-box vendor model may require a specific initialisation sequence and settling time. Initialising a machine or a converter means setting its control states, not just its P and Q.

The isolated-voltage-source technique

In the isolated-voltage-source technique, the inverter or plant model is first connected to a controlled voltage source that reproduces the required terminal voltage, frequency and phase, and is allowed to settle at the target P / Q operating point before being connected into the wider EMT network. The purpose is to avoid the artificial transients caused by forcing a complex controller to start instantly inside a large network. Once the plant has settled on its own source, it is switched into the network at the same voltage and angle, so the transition is smooth.

Do not rush the settling period

Do not shorten the pre-disturbance settling period only to save time. Some plant controllers, measurement filters, PLLs and dc-link controllers need time to settle; if the disturbance is applied too early, the result may reflect an initialisation error rather than real system behaviour. Better results — settling below about 200 ms — come from building initial-condition rules into the controls, so the blocks are initialised backwards from the computed steady state.

Section 4

Boundary buses: the key decision

A boundary bus is the interface between the detailed retained EMT network and the external system equivalent; its location decides what is modelled explicitly and what is replaced by an equivalent. It is one of the most important engineering decisions in a large-scale EMT study. The retained area is the part of the network modelled explicitly in EMT; the external equivalent is the simplified representation of the rest. The retained area must contain the mechanism being studied; the equivalent must reproduce how the rest of the grid pushes back on it.

Where to place the boundary — and where not to

Choose the retained EMT area to include the plant or converter being studied, at its point of connection (PoC / PCC); nearby weak-grid buses; nearby HVDC, STATCOM, SVC, series capacitors and large cables; the protection and switching devices that may operate; the network paths that shape the voltage, frequency or impedance response; nearby converters that may interact; and the measurement points needed for validation.

Avoid placing the boundary too close to the disturbance; between strongly coupled converters; inside an important resonance path; where the equivalent would hide a protection action; or where the frequency or voltage response of the external system is itself important.

Section 5

Network equivalencing

Network equivalencing is not just deleting distant buses. It is replacing the external system with a model that reproduces the voltage, frequency and impedance behaviour seen at the boundary, over the frequency range and time scale relevant to the study. And the correct equivalent depends on that frequency range: a transient-stability voltage-recovery study may need very different external-system behaviour from an SSCI, harmonic-resonance or cable-switching study. Table 1 sets the options against each other.

Table 1 — External-network equivalencing options for a wide-area EMT study (after CIGRE TB 881, Section 4.1).
Equivalent	Best used for	Main strength	Main risk
Static voltage source	Strong external system, far from the event; simple local tests	Simplest, most common, cheap	Holds V / f too stiffly; false frequency / voltage recovery
Controlled dynamic source	Boundary voltage / flow that moves in the event	Holds pre-event P / Q; more realistic	Only as good as the control law and data
Synchronous-machine equivalent	External electromechanical / frequency response matters	Real inertia, governor and exciter dynamics	Poor for converter-rich external areas unless tuned
FDNE	Resonance, harmonics, SSCI, cable networks	Correct impedance across a frequency band	Must be fitted, passivity-checked and validated
Hybrid PDT–EMT boundary	Local EMT area, wider grid in PDT	Full external dynamics at low cost	Phasor conversion can remove a fast phenomenon
Full retained EMT area	The most accurate, safest option	No equivalencing error	Highest computational burden

A network-equivalencing workflow

1. Define the phenomenon and the frequency range.
2. Choose the retained EMT area, then 3. the boundary buses, then 4. the equivalent type.
5. Match the pre-disturbance load-flow.
6. Check the short-circuit level (SCR) and X/R at the boundary; 7. check impedance versus frequency if needed.
8. Run a benchmark disturbance; 9. compare against a fuller EMT, PDT or measured response.
10. Document the validity limits.

Section 6

Source and machine equivalents

Three of the equivalents deserve unpacking. A static voltage-source equivalent holds voltage and frequency too stiffly. It can be useful for simple local tests or a very strong external system, but it produces misleading results when the external system’s frequency, inertia, damping, voltage response or dynamic interaction matters. Its clearest failure: a static source can behave like an infinite system — after a disturbance it may force frequency or voltage back toward nominal even when the real external system would slow down, oscillate or remain depressed, hiding a frequency-security or voltage-recovery problem.

A controlled dynamic source reproduces some external behaviour by adjusting voltage magnitude, angle or frequency according to a control law or a measured response; it is more realistic than a fixed source, but only as good as the control logic and data behind it. A synchronous-machine equivalent represents the external grid with one or more equivalent machines carrying inertia, a governor, an exciter and an impedance; it reproduces frequency and electromechanical response far better than a fixed source, but it may not represent converter-rich external areas accurately unless it is tuned and validated.

Section 7

Frequency-dependent equivalents

An FDNE — frequency-dependent network equivalent — represents the external network impedance over a range of frequencies, not only at 50 / 60 Hz. This matters when resonance, harmonic interaction, sub-synchronous behaviour, cable networks or converter-control interaction are part of the study — exactly where a single power-frequency impedance is not enough. But an FDNE is not automatically correct: it must be fitted to the network data, checked for passivity and stability, and validated over the relevant frequency range. A poorly fitted equivalent can introduce artificial resonances that were never in the real system — the standing risk in every reduction.

Section 8

What the boundary choice actually does

A real example makes the abstraction concrete. Three EMT models of the Australian National Electricity Market (NEM) were compared on the same generator-loss event, differing only in how much was modelled in detail versus reduced to a static voltage source (Table 2).

Table 2 — The same generator-trip event in three NEM models with different static-equivalent boundaries (after CIGRE TB 881, Section 4.1.7). “HSM” is the field high-speed-monitor recording.
Model	Regions in detail	Static equivalent	Post-trip generator MW rise	Frequency settles at	Matches HSM?
2-state	Victoria, New South Wales	South Australia + Queensland	None	Restored to 50 Hz	No
3-state	Victoria, NSW, South Australia	Queensland	None	Restored to 50 Hz	No
4-state	All four mainland states	None	~30 MW (governor response)	~49.8 Hz	Yes

Only the full 4-state model reproduced the field recording: the monitored generator raised its output about 30 MW in a governor response, and the frequency settled near 49.8 Hz rather than fully recovering to 50 Hz. The reduced models got it wrong for a revealing reason. Their constant voltage-source equivalents poured a large amount of active power into the initial frequency deviation, which both restored the frequency artificially and stopped the real generator models from raising their own output to cover the loss. Modelling the whole system gives the highest accuracy for a frequency disturbance; a reduced model can still be reasonable, but only with an appropriate equivalent — and a static source is not it for frequency events. A well-tuned dynamic source would likely have done the job here too; the real decision is the effort of tuning it (and trusting it across all conditions) against the cost of running the full network in EMT.

Section 9

Co-simulation: splitting the problem

Co-simulation means two or more simulation engines run together and exchange signals at defined interfaces — one tool may simulate an EMT area while another simulates a PDT / RMS area, or two EMT simulators may each solve part of the system. It is used to reduce computation burden, reuse existing models, protect vendor / IP-sensitive models, combine tools with different strengths, enable real-time / HIL testing, and split a very large network into manageable parts. Its gains are real: the sub-systems run in parallel, and each can use its own solver, detail level and step size.

Co-simulation is not automatically stable

An interface can introduce delay, numerical instability, interpolation error, signal scaling or sign errors, inconsistent time steps, inaccurate boundary conditions, and hidden assumptions about phasor / waveform conversion — so the interface must be tested, not assumed. Convergence is guaranteed only when there is no algebraic loop between the sub-systems, and there is none when the split is made at a natural time delay. A natural time delay exists when a line, cable or communication path has a physical propagation delay between its two ends: place the interface across that delay and each side does not need the other side’s instantaneous value at the same moment, which makes the coupled simulation more stable and easier to converge — a Bergeron line model naturally contains exactly that travelling-wave delay. Where the system cannot be split at a natural delay, stability needs iterative coupling (heavy, not always supported) or a network equivalent at the interface, without introducing artificial resonances.

At every EMT / PDT or EMT / EMT boundary, check…

the voltage base, current base, sign convention, phase reference and frequency reference;
the time-step alignment, the interpolation method, the filtering / phasor extraction, and the delay;
the power balance and the stability / passivity;
the response to a small disturbance and to a large disturbance.

Section 10

Hybrid PDT–EMT simulation

Hybrid PDT–EMT simulation keeps the detailed local area in EMT and represents the wider grid in PDT / RMS, running in parallel. The interface converts EMT waveforms into phasors for the PDT side (typically by a fast Fourier transform, FFT) and converts PDT voltage / frequency / angle information back into time-domain boundary signals for the EMT side. It lets a study reuse existing, sometimes encrypted, PDT models for legacy plant whose control block diagrams are unknown, and it speeds the run up when each inverter’s time step is relatively large (tens of microseconds) — though where the models demand 1–2 µs steps the gain shrinks.

When hybrid PDT–EMT is — and is not — suitable

Suitable when the local area needs EMT detail; the wider grid is well represented by phasor dynamics; the interface is electrically far enough from the fast phenomenon; and the boundary signals can be converted without losing the mechanism. Not suitable when the fast EMT phenomenon crosses the interface; harmonic or resonance behaviour spans both sides; weak-grid interaction depends on external converter details; or the phasor conversion removes the key waveform behaviour. Hybrid simulation is not just a software trick — it is a way to retain EMT where the converter interaction happens while avoiding the cost of modelling the whole external grid in EMT.

A worked example — the Texas synthetic network

A hybrid case built on a 2000-bus, 432-generator synthetic model of the Texas system illustrates both the accuracy and the speed. The weak Panhandle region — high inverter penetration, no conventional units in service, 627 MW of wind — was simulated in EMT, the rest in balanced PDT, splitting the network at the four long lines (57, 122, 167 and 259 miles) that connect the region. For a three-phase 161 kV busbar fault, cleared after 100 ms with a 500 kV line trip and local load shedding, the full EMT run showed an oscillatory instability below 1 kHz — a power-electronics / network-impedance interaction that could trip protection and cascade. The hybrid co-simulation reproduced the same behaviour (small differences in peak and timing only), while the pure PDT run missed it entirely. The speed pay-off: the full EMT run was about \(148\times\) slower than balanced PDT, the co-simulation only \(18\times\) — roughly an \(8\times\) speed-up over full EMT, with equally good results.

Section 11

EMT–EMT co-simulation and confidentiality

EMT–EMT co-simulation splits one EMT problem into multiple EMT sub-systems. It can improve speed (parts on different processors in time-step synchronism), allow different time steps for the inverters (small) and the network (larger) — or accommodate a vendor model that only runs at a fixed step — bridge incompatible software environments, and place IP-sensitive models in secure environments. One operator’s real-time EMT simulator imports controller components from different vendors as dynamic-link libraries (DLLs) to preserve each manufacturer’s intellectual property, aiming to simulate the whole national network at close to three times real time.

IP-secure model sharing

In IP-secure EMT–EMT co-simulation, a vendor or project party runs its own black-box model in a separate environment and exchanges only boundary voltages / currents or agreed signals — preserving confidentiality while still allowing an interaction study. Two schemes are emerging: a single connection-point exposure, which behaves like a SMIB study except the “infinite bus” is a detailed representation of the full network (used to let a newly connecting generator tune against a fidelity it could not otherwise reach); and a multiple connection-point exposure, which moves the network model to the non-confidential side so the external user can run their own scenarios, provided the confidential models accept run-time P / Q / V dispatch without recompilation. Hosting others’ models this way is a software-as-a-service arrangement, so the tool and every dependency (compilers, operating system) must be properly licensed.

Section 12

Real-time simulation, HIL and SIL

Real-time simulation means the simulator must compute each time step within the same amount of real clock time: if the step is 50 microseconds, the simulator has less than 50 microseconds to solve the network, update the controls and exchange its inputs and outputs — overrun the step, even once, and the result is inaccurate or diverges. It takes a real-time simulator (RTS / DRTS): parallel multi-core processors, a real-time operating system and specialised software, with field-programmable gate arrays pushing steps into the hundreds of nanoseconds. A large EMT system of thousands of nodes is decoupled into sub-tasks across the cores to hold the constraint — on transmission networks the propagation delay of long lines gives the natural split (Bergeron decoupling); distribution systems and microgrids use other techniques, including decoupling at transformers via stub lines (trading accuracy for speed) or a state-space nodal (SSN) solver.

HIL and SIL

HIL — hardware-in-the-loop — connects real control or protection hardware to the simulated network: the simulator sends voltages and currents to the hardware, which sends control and protection actions back. SIL — software-in-the-loop — connects the actual or compiled control software to the simulated network without the physical hardware. Page 11 discussed HIL as validation evidence; this page explains what is needed at the system-simulation level to make HIL or real-time studies possible. HIL and SIL belong logically after model acceptance and validation, because they are ways to test model and controller behaviour under realistic system conditions.

One practical point governs the speed of the whole exercise: the smallest required time step often controls the whole simulation. A detailed converter model, a protection algorithm or a switching model may force a smaller step than the rest of the network needs — and co-simulation or sub-system splitting can sometimes isolate that small-step requirement so it does not slow everything else down.

Section 13

A large-scale EMT build, end to end

Putting the whole page into one repeatable procedure:

A large-scale EMT build workflow

1. Define the study question and phenomenon; 2. identify the required EMT area and the external network.
3. Select the retained area and the boundary buses; 4. choose the equivalent type for the external system.
5. Build or import the load-flow / PDT case; 6. replace the study-critical models with validated EMT models.
7. Initialise the synchronous machines, IBGs, HVDC / FACTS and controls; 8. check the steady-state match against the reference load-flow.
9. Run a flat-run / no-disturbance check; 10. run a benchmark disturbance.
11. Check the boundary response and the equivalent’s validity; 12. run the study cases.
13. Record the outputs, protection actions and interface signals; 14. document the assumptions and validity limits.

The signals worth plotting reveal both the physics and the numerics: the boundary bus voltage magnitude and angle, and the boundary active / reactive power; the system frequency; the retained-area bus voltages; the IBG active / reactive current and power; the PLL frequency / angle and the dc-link voltage where available; the HVDC / FACTS output; the synchronous-machine rotor angles and speeds; the protection trip / block / runback flags; the load and DER response; the interface exchanged signals in a co-simulation; the energy / power balance across the boundary; and any long-run drift.

Common mistakes

The traps that most often undermine a large-scale EMT study:

Eleven traps to avoid

Building a large EMT case before defining the study phenomenon.
Assuming an imported PDT / RMS case is automatically ready for EMT.
Applying the disturbance before the EMT model has settled.
Placing the boundary too close to the disturbance.
Using a static voltage source where the external-system frequency response matters.
Using a simple equivalent for harmonic, SSCI or cable-resonance studies.
Ignoring boundary sign conventions and base-value mismatches.
Treating co-simulation as automatically stable.
Forgetting that phasor conversion may remove the EMT phenomenon.
Running HIL / real-time without checking the time-step and I/O delays.
Documenting results without documenting the equivalent’s validity limits.

Key points

Build it right, not just big

Large-scale EMT is a system-building problem, not only a solver problem.
The EMT case must start from a steady state that matches the reference load-flow.
IBG, HVDC and FACTS initialisation must include control states, not only P and Q.
Boundary placement can decide whether the study mechanism is preserved or hidden.
Static voltage-source equivalents are simple but can give false frequency / voltage recovery.
FDNEs are useful when frequency-dependent external impedance matters.
Hybrid PDT–EMT is valuable when only the local area needs waveform detail.
Co-simulation can improve speed and IP protection but introduces interface risks.
Real-time / HIL studies require strict time-step and I/O timing discipline.
Every equivalent and interface must have a documented validity range.

The main takeaway: large-scale EMT accuracy depends on four linked decisions — what area is retained, how the rest of the system is equivalented, how the case is initialised, and how sub-systems exchange information. A detailed converter model inside a poorly initialised or poorly equivalented network can still give a wrong answer. The engineer must preserve the mechanism being studied at the boundary, not just make the case run.

This completes the wide-area EMT workflow: first decide whether the models are adequate, then test and validate them, then build the large-scale case with the right initialisation, boundaries, equivalents and simulation interfaces. This page is based on CIGRE Technical Brochure 881, with APS engineering interpretation.

References

CIGRE Technical Brochure 881 is the primary reference for this page; the CIGRE/CIRED brochure on inverter-based generation gives the wider modelling context; and Dommel’s 1969 paper is the origin of the travelling-wave line model used for co-simulation and real-time decoupling.

CIGRE Working Group, Electromagnetic Transient Simulation Models for Large-Scale System Impact Studies in Power Systems Having a High Penetration of Inverter-Connected Generation. CIGRE Technical Brochure 881.
CIGRE/CIRED Joint Working Group, Modelling of Inverter-Based Generation for Power System Dynamic Studies. CIGRE Technical Brochure.
H. W. Dommel, “Digital Computer Solution of Electromagnetic Transients in Single- and Multiphase Networks,” IEEE Transactions on Power Apparatus and Systems, vol. PAS-88, no. 4, pp. 388–399, 1969.
EMTP®, Electromagnetic Transients Program — Documentation, Network Equivalencing and Real-Time / Co-Simulation Notes. Powersys / EMTP®.

Part 7 Reading now

Large-Scale EMT Simulation

Building, initialising, reducing and coupling a wide-area EMT case — network equivalencing, co-simulation, hybrid PDT–EMT and real-time.

Series progress 7 of 12

Part 8

Modelling IBGs for Frequency Stability

Applying the models to frequency stability: inertia, ROCOF, synthetic inertia and fast frequency response.