How can I linearize an already compiled model ?

I would like to improve the performance of my script which use linmod to linearize a simulink model in Matlab 2016b. The linmod function is used many times with different state/input vectors so I think compiling the model only once could improve performance.

The structure of my scipt would be :

feval(model,[],[],[],'lincompile');
for i = 1:numel(x)
  out = linmod(model,x{i},u{i});
end
feval(model,[],[],[],'term');

But unfortunalty, it only works with the 'v5' linarization algorithm. With the default block by block algorithm, an error is raised when changing model parameters :

Error using dlinmod>local_push_context (line 293)
Cannot change the 'AnalyticLinearization' parameter while the model 'MyModel' is running

Error in dlinmod (line 181)
have = local_push_context(models, want);

Error in linmod (line 59)
[varargout{1:max(1,nargout)}] = dlinmod(model, Ts, varargin{:}, Args);

Is there any way to make linmod work with a compiled model ? Do you have any other idea to improve performance of many linearizations of the same model ?

Answer 1

[Julien Prodigue]

I have found a solution for this.

dlinmod sets some models parameters before computing the linearization and those parameters cannot be changed while the model is compiled : you juste have to set those parameters before compiling. Thus, the set_param wont raise an error if the parameter to set is already set to the same value.

    set_param(model,'AnalyticLinearization','on');
    set_param(model,'BufferReuse','off');
    set_param(model,'SimulationMode','normal');
    set_param(model,'RTWInlineParameters','on');
    set_param(model,'InitInArrayFormatMsg','None');
    feval(model,[],[],[],'lincompile');
    for i = 1:numel(x)
      out = linmod(model,x{i},u{i});
    end
    feval(model,[],[],[],'term');