(1) You can transfer the ABM model to dynamic model based on their behavior. Then use optimal control strategies on the dynamic model. Like this paper did: Article Modeling individual fear factor with optimal control in a di...
(2) Consider the behaviors of agents as variables. If you have totally n behaviors, you will get a space with n*t dimension. With optimization methods like Neural Network, Practical swarm optimization or Heuristic. You can find a solution in your n*t dimension. This is our working paper, we will share it when it is published.
I used ABM in disaster recovery to find the optimal post-disaster recovery strategies to increase the community welfare and decrease the vulnerability of the built environment.