One commonly used technique is to combine the GA with a local optimization technique specific to the problem being solved. This can be done within the IML GA framework by incorporating a local optimization into the objective function evaluation: return a locally optimized objective value, and optionally replace the original solution passed into the module with the optimized solution.
Always replacing the original solution with the locally optimized one will cause faster convergence, but it is also more likely to converge prematurely to a local optimum. One way to reduce this possibility is to not replace the original solution in every case, but replace it with some probability p. For some problems, values of p from 5 to 15 percent have been shown to significantly improve convergence, while avoiding premature convergence to a local optimum (Michalewicz, 1996) .