On the integration of risk aversion and average‐performance optimization in reservoir control

The real‐time operation of a reservoir is a matter of trade‐off between the two criteria of risk aversion (to avoid dramatic failures) and average‐performance optimization (to yield the best long‐term average performance). A methodology taking into account both criteria is presented m this paper to derive “off‐line” infinite‐horizon control policies for a single multipurpose reservoir, where the management goals are water supply and flood control. According to this methodology, the reservoir control policy is derived in two steps: First, a (min‐max) risk aversion problem is formulated, whose solution is not unique, but rather a whole set of policies, all equivalent from the point of view of the risk‐aversion objectives. Second, a stochastic average‐performance optimization problem is solved, to select from the set previously obtained the best policy from the point of view of the average‐performance objectives. The methodology has several interesting features: the rnin‐max (or “guaranteed performance”) approach, which is particularly suited whenever “weak” users are affected by the consequences of the decision‐making process; the flexible definition of a “risk aversion degree,” by the selection of those inflow sequences which are particularly feared; and the two‐objective analysis which provides the manager with a whole set of alternatives among which he (she) will select the one that yields the desired trade‐off between the management goals.