How can I improve a poor data fit using lsqnonlin?

Hi everybody,

I try to fit a physical model to a set measured data using lsqnonlin with 'trust region'. The data I want the model to fit to is an array of (8,13,5). Where dimension 1(Size of a transistor) and dimension 3(Stress Voltage) make up the number of datasets. I posted exampledata in the uploaded file.I fit the model to the data using the Code below.The extraparameters VDStress and W are directly weighting parameters, directly linked to the each dataset. I use multistart also multistart.

For the best result I get an Exitflag = 1 but still, the result is extremely poor fitted. The bounds are set from 0 to inf.

Has anybody got a clue how to improve the poor fit?

Thanks in advance, Julian

   problem=createOptimProblem('lsqnonlin',...
                               'x0',x0,...
                               'objective',@nestedfun,...
                               'lb',lb,'ub',ub,...
                               'options',optimset('MaxFunEvals',FunEvals,...
                               'TolFun',FuncTol,'TolX',XTol,...
                               'MaxIter',Iter));

   [x,resnorm,residual,exitflag,output] = lsqnonlin(problem);
   % Nested function that computes the objective function     

     function F = nestedfun(x)
        A=x(1).*(IdsW).^x(2).*(IsubStress./IdsStress).^x(3);
        B=x(4).*VDStress.^x(4).*(IdsW).^x(5);
        F=(Stresstime.^x(6).*(A+B))-ydata;
     end

[Matt J]

You need to attach the optimProblem and ydata as well. You should also better explain how to extract IdsW, IsubStress, etc... from the Dat in your matlab file.