optimizing complex neural network cost function using fmincon

I am trying to fit a neural network output to some given data, more importantly , the height of the predicted data should not be lower than the given ones , so i pass the heights as inequality connstraints , when i choose the constrains to be just above the ground (>0), the output of the fmincon becomes reasonable , however when applying the exact height , it keeps diverging , is this solver (fmincon ) suitable for this kind of optimization ?