Constrained differential dynamic programming and its application to multi-reservoir control

The authors describe and analyse a successive approximation dynamic programming technique believed to be the most efficient procedure available for solving multi-reservoir control problems. Two advantages are claimed for this approach: the memory and computational requirements do not grow exponentially with the decision or state variable, and discretization of state or control space is not required. The solution of a 10-reservoir example is presented as an illustration of the technique in operation.