Comprehensive guide to grid protection and real-time EMT simulation

Key Takeaways

• EMT studies are the default check wherever converter controls, protection or weak grid conditions decide the outcome, with phasor tools kept for system-wide screening.
• Timestep follows from the fastest event you need to resolve, and that same choice decides which parts of a model stay on CPU and which move to FPGA.
• Software selection should favour a model that carries from offline study to real-time hardware testing without being rebuilt.

EMT simulation shows you how a grid behaves in the first milliseconds after a disturbance, which is where inverter-heavy networks now succeed or fail.

Synchronous machines used to set the pace of a disturbance, and phasor tools captured that pace well enough for planning. U.S. developers expect solar to supply more than half of the 64 GW of generating capacity planned for 2025, with battery storage taking much of the rest. Those plants connect through power electronics whose controls act faster than a phasor model can represent.

You should treat EMT studies as the default check for any connection where converter controls, protection or weak grid conditions shape the outcome. Phasor studies still earn their place for system-wide screening. What matters is knowing where they stop being trustworthy, then choosing a timestep, a solver and a real-time simulation path that match the fastest event you need to see.

What EMT simulation solves at the waveform level

EMT simulation, short for electromagnetic transient simulation, solves the instantaneous voltages and currents of every phase in a network at steps of microseconds. It keeps the differential equations of lines, transformers, machines and converters intact, so you see the actual waveform, including harmonics, imbalance, switching events and the fast control responses that shape them.

Picture a capacitor bank energized on a 230 kV bus. The voltage doesn’t just step to a new level. It rings at several hundred hertz, overshoots the nominal peak, and decays over a few cycles. An EMT model reproduces that oscillation point by point, so you can read the peak stress on nearby equipment from the trace.

That detail carries a cost. Each component needs physical parameters such as line geometry, transformer saturation curves and converter control code, and the solver works through hundreds of steps per cycle. You’re trading compute time for a result that holds at the timescale where protection and inverter controls respond.

How EMT simulation differs from phasor simulation for grid studies

The main difference between EMT simulation and phasor simulation is what each one assumes about the waveform. Phasor tools represent voltages and currents as magnitudes and angles at the fundamental frequency and step in milliseconds. EMT tools make no such assumption and compute every instant, which captures sub-cycle events that phasor models average away.

Consider a solar plant behind a long radial line. A phasor study shows it riding through a nearby fault, because the model sees only the positive sequence voltage dip. The EMT study of the same fault shows the phase-locked loop losing its reference during the phase jump and the inverter tripping on a sub-cycle overvoltage. Only one of those results describes what the plant will do.

When a transient study needs EMT detail

You need EMT studies when the outcome depends on events faster than one cycle or on the internal controls of a converter. That covers most inverter connections in weak areas, any study of switching or lightning stress, and any case where protection acts on the waveform itself.

The clearest warning came from Texas. A single fault near Odessa in June 2022 led to the loss of 1,711 MW of solar PV generation from plants that should have stayed connected. Investigators reported that positive sequence models couldn’t confirm ride-through performance ahead of operation, and that common modelling practice left out protections such as AC overcurrent and DC bus protection.

Several conditions should move a study to EMT without much debate.

• Short-circuit ratio at the connection point is low
• Series-compensated lines or HVDC links sit electrically close
• Several converter plants share one substation
• Grid-forming controls or black start duty are specified
• Protection relies on sub-cycle or travelling wave elements

None of these means abandoning phasor work. Screen the whole system with it, then study the area that fails these checks in EMT, with the rest of the network reduced to an equivalent.

What timestep an EMT simulation needs for accurate results

An EMT simulation needs a timestep at least 10 to 20 times shorter than the period of the fastest phenomenon you want to resolve. Conventional grid transients are commonly run at about 50 microseconds. Converters with kilohertz switching need a few microseconds or less, and sub-microsecond steps apply once switching reaches tens of kilohertz.

The arithmetic is quick to check on your own model. A 50 microsecond step gives 333 points across a 60 Hz cycle, which is ample for a fault study. Apply that step to a converter switching at 5 kHz and you get 4 points per 200 microsecond switching period, far too few to place the switching edges. Line length matters as well, since a travelling wave covers roughly 15 km in 50 microseconds and a shorter overhead line can’t be modelled with its propagation delay.

Smaller isn’t automatically better. Halving the step doubles the compute load, and real-time work requires every step to finish on time. Pick the largest step that resolves the event you care about, then confirm it with one run at half the step. If the two traces agree, you’ve found your number.

“If the two traces agree, you’ve found your number.”

Where CPU solvers stop for fast switching converters

CPU solvers handle grid-scale EMT models well at steps of a few tens of microseconds, and they stop being practical when converter switching frequency calls for steps near or below one microsecond in real time. At that point, the switching portion of the model moves to an FPGA, while the surrounding network stays on the CPU.

A wind plant study shows the split. The transmission network, collector system and machine models run in HYPERSIM as the EMT solver, spread across parallel CPU cores. A 50 microsecond step is enough for everything on that side.

Detailed converter models change the picture. For a model built in Simulink, OPAL-RT uses ARTEMiS for CPU-side electrical solving, which keeps switched circuits stable at a fixed step. Once switching frequency pushes past what a CPU step can resolve, the converter moves to eHS, an FPGA-based solver, and the two sides exchange signals every CPU step.

The boundary between them deserves care. Place it at a point with natural decoupling, such as a transformer or a long cable, so the one-step delay doesn’t disturb the result.

How to compare EMT simulation software for your studies

EMT simulation software falls into two groups, offline tools that run as fast as the model allows and real-time tools that hold a fixed step against the clock. Compare options on converter model library depth, ability to accept vendor control code, parallel performance on large networks, and a path from desktop study to hardware testing.

A transmission planner preparing an interconnection study and a test engineer validating a plant controller will weigh those criteria differently. The planner cares most about running hundreds of fault cases and importing black-box inverter models from manufacturers. The engineer needs an EMT simulator that closes the loop with a physical controller through analog and digital I/O, so the same network model has to run in real time. Ask each vendor how much of a model survives that move, because rebuilding a validated network in a second tool is where errors creep in.

Transient studies earn trust when the model matches the grid

A transient study is only as reliable as the match between its modelling detail and the equipment on the network. Grids built around converters respond in microseconds and milliseconds, so the study has to work at that scale wherever controls and protection decide the result.

That isn’t an argument for running everything in EMT. It’s an argument for being deliberate. Screen broadly with phasor tools, mark the regions where their assumptions break, and spend your detailed modelling hours there. Choose the timestep from the physics and check it once.

Teams that work this way find problems at a desk instead of at commissioning. We see the same pattern at OPAL-RT when a study model moves onto a real-time simulator and meets the actual controller for the first time. The surprises are smaller when the offline work was honest about its limits. Good transient studies don’t remove uncertainty, but they tell you exactly where it sits.

“Good transient studies don’t remove uncertainty, but they tell you exactly where it sits.”