Two-level converter causing computation spike every 3.3 ms

I am currently running a real time simulator based on a simulink model. In this model I use the Two-level-converter block available in simulink. The problem I encounter is that every 3.333ms it uses large computational time.

This can be observed in the added picture where you should compare the computation time when using the two-level converter and when using a controlled voltage source (the converter controller is never changed.)

Any idea of what is causing this? (3.333 ms is exactly 6 times in a 50 Hz cycle, 50 Hz is the power system frequency)

Any idea on how to reduce the spikes in computation time?

[Peter O]

Sounds like it's related to the commutation of the power system between phase legs if it's six times per cycle, assuming it's a three phase system.

Are you modeling it with the Switching devices mode? If so, it's modeling device turn on and off during those intervals, which is a bit more complex than continuous conduction. If you probe the system around that point do you see any chattering in the currents? How big is the load it's feeding?

[Karl Fredrik Wirgenes]

Thanks for the feedback.

I forgot to mention that the I am running matlab 2019B. An the converter is set to be an average model (Uref controlled), which uses a switching function (not a switching device(?)). You seem to be correct that it is related to the commutation as it occours for each zero crossing of the inverter current.

chattering in the currents?

Not at all, almost a perfect sinusoid.

How big is the load it's feeding?

A large microgrid running at about 7 MW with a power factor of 0.9 (perfectly balanced system)

[Joel Van Sickel]

Can you share your model? Are you using simscape (blue) or specialized power systems (black)? You can also run the profiler in Simulink to see how long a model runs. Can you recreate this problem on desktop? In general, even if something strange is happening with commutation, a fixed step solver should take the same amount of time. The exception to this rule is if somehow a setting it set allowing the solver to iterate. The other difference would be if you are logging significantly more data with the new block, but that would not likely line up with your system frequency.