Approximating networks for the solution of T-stage stochastic optimal control problems