Question about Initialization Error Regarding Numeric Loop

[jinningwang]

Hi Hantao,

I encountered an initialization error when developing the AC8B model:
When generating the code by andes prep, it returns an error saying: cannot evaluate value ...

May I ask how to set up the initialization for such a model?

Regards,
Jinning

Note:
AC8B model diagram, where FEX is a piece-wise function.

[jinningwang]

Answer from Hantao,

The initialization error is caused by the numeric loop when initializing the model. A similar issue occurs in the EXAC1 model (in exac1.py).

Since the FEX is a piece-wise function, the explicit initialization cannot be derived directly. For such variables, we need to solve the initial values by iteration, just like how we solve the DAE. Therefore, we need to assign iteration equations v_iter and start values v_str for them.

In the EXAC1, three variables are involved in the iteration initial values issue, all of them need to be assigned v_iter and v_str:

        self.IN = Algeb(tex_name='I_N',
                        info='Input to FEX',
                        v_str='1',
                        v_iter='KC * XadIfd - INT_y * IN',
                        e_str='ue * (KC * XadIfd - INT_y * IN)',
                        diag_eps=True,
                        )

        self.FEX = Piecewise(u=self.IN,
                             points=(0, 0.433, 0.75, 1),
                             funs=('1', '1 - 0.577*IN', 'sqrt(0.75 - IN ** 2)', '1.732*(1 - IN)', 0),
                             info='Piecewise function FEX',
                             )
        self.FEX.y.v_str = '1'
        self.FEX.y.v_iter = self.FEX.y.e_str

        ... ...

        self.INT = Integrator(u='ue * (LA_y - VFE)',
                              T=self.TE,
                              K=1,
                              y0=0,
                              info='Integrator',
                              )
        self.INT.y.v_str = 0.1
        self.INT.y.v_iter = 'INT_y * FEX_y - vf0'

One more thing is that, in the IN.e_str, we avoid division (KC * IFD / VE - IN = 0) but use (KC * IFD - VE * IN = 0), because the first will result in numerical inaccuracies.

@cuihantao
Feel free to correct me if I misunderstand your design.

Regards,
Jinning