The optimal control problem associated with many water resource systems may be formulated as a nonlinear optimization problem involving a nonlinear objective function subject to both implicit system (state) constraints and implicit bound constrains. A general solution is proposed that uses a disaggregated or dual level approach in which the system state equations are removed from the control formulation and evaluated externally using mathematical simulation while the resulting optimization formulation is solved using a genetic optimization routine. Potential algorithm performance enhancement may be obtained by replacement of the simulation algorithm with a neural network representation. The resulting network may be trained on-line using real time data or it may be obtained using multiple off-line state predictions obtained from a simulation model. Potential applications of the approach are presented along with a discussion of potential problems.