I am working on a resource allocation problem using an SPEA 2 evolutionary algorithm. The problem involves decision variables where each variable has a different domain e.g. $E_i \le d_i$ where $E_i$ is the allocation to a user and $d_i$ is individual demands. The problem involves a linear constraint such that $\sum(E_i) = total resource$. The probability of the creation of feasible off-springs after crossover and mutation operators is extremely low. So, we need a repair operator for this purpose. I need guidance for the selection of suitable repair operators and how should I apply that, I mean should we repair all solutions or some percentage.
So far I designed an operator where the of an off-spring generated is repaired as follow: 1) apply bound 2) Determine constraint violation i.e. leftover resource or over-consumed resource 3) Find unmet demand $d_i$ 4) Divide remaining resource proportionally among user, so update $E_i$
This operator creates a feasible solution but I am not sure if my approach is reasonable and what percentage of solution I need to repair.
I would appreciate the guidance, comments, or any literature reference.