Wrong result in a realistic, PowerGrids-based example

[ceraolo]

The attached OM.png and Dymola.png show variable IPu as obtained using the models ScTT and ScTT1 from the enclosed Test.mo. Note that they require PowerGrids library.

The two results should be identical since ScTT1 just adds some sensors to ScTT.
The Dymola's results are indeed identical.
The OM results are quite different, and the one from ScTT1 is badly wrong:

• the value during the first 100 ms is totally different (and invalid)
• the short circuit current (immediately after 100 ms) is too low
• the current, after an initial reduction after the short circuit occurring at 200ms, rises again: it should continuously reduce

[casella]

Fixed Test.mo to actually use the embedded Support package.

[casella]

@ceraolo, I suspect that the two examples converge to different solutions of the same problem, one with high voltage and low current, the other with low voltage and high current. ScTT starts with the generator voltage at 0.18 p.u., which is most likely not the solution you are seeking.

The model is a bit messed up from the point of view of initialization: bus is on the medium-voltage side, but it has UNom = 380 kV and UStart = 1.05 times that. Conversely, NTHV is on the high-voltage side, but has UNom = 21 kV.

I would suggest you to set nominal and start values properly, possibly by first running a power-flow model as suggested in the User's Guide and using the results to set the proper start values for all components. Then, you will most likely get the same results with both models also in OpenModelica.

BTW, we could probably add some asserts the machine and load models that warn you if the initial voltage values are below 0.5 p.u., because that is most likely not a valid solution. What do you think?

[ceraolo]

Well, maybe it was not a good idea to combine different issues in the same ticket.
I'm going to close this one and open a new one on initialisation only.

Just a comment here:
The system up to 0.1s is linear, and therefore can have only one equilibrium point. Models ScTT and sScTT1 mathematically describe the same system since they differ only for sensors; therefore they cannot stay stably ad different points during the whole 0-0.1s period, whatever data we put as start values.

[casella]

Replying to ceraolo:

Well, maybe it was not a good idea to combine different issues in the same ticket.
I'm going to close this one and open a new one on initialisation only.

Just a comment here:
The system up to 0.1s is linear,

It is definitely not. Please run it with the transformational debugger. You will see there are several nonlinear equations in the intialization phase, including a system with 25 unknowns and three tearing variables, namely the generator angle, direct current and quadrature current.

The reason is that the default initialization option (Global steady state, fixed set points) computes the system-wide steady-state, thus including all the nonlinear machine equations.