I don't see any basically wrong with the concept. It's a matter of tuning the parameters, I'm thinking.

Why not experiment with random values that are more spread out (i.e. a greater range)? ...