How to seed feasible population to GA ?

I am trying to solve a MILP problem by GA.

BUT the GA is not able to find feasible initial population on its own. If the Zeros is given as initial solution it is not able to converge.

I tried to feed the linprog solution as an initial solution but this just leads to deaddrop convergence of the GA and that too with constraint violation.

Help me I want gradual convergence of GA with the solution which never violates the constraints even though solution was sub optimal.

P.S. : I know that GA is not best method for MILP but this is just experimental.