Value of information in reservoir optimization

Publisher Website
Google Scholar
Abstract
Numerical experiments with operation policy optimization for a reservoir fed with a random log normal input demonstrated a negligible sensitivity of results to the assumed form of input distribution, a near optimality of the policy ‘draft equals input mean’ for any reservoir size and any size of input sample used for the policy development, and a similarity between the effects of hydrologic and economic uncertainties. In particular, it was shown that the so‐called standard policy, ‘draft equals target,’ arises from any optimal policy as either hydrologic or economic uncertainties grow beyond limits. While there was a clear hierarchy of optimal policies governed by the amount of information underlying the optimization if operation results were assessed vis‐à‐vis expected values and the true imput population, the hierarchy was highly distorted when it was judged from the performance of individual policies on a 10‐year sample from the true input population, representing an actual future operation period. It was also found that the degree of optimality of a given policy did not depend on the degree to which the underlying input model and its parameters agreed with the true input population, but rather on the similarity of the features of the ‘historic’ sample underlying a given policy and the ‘future’ 10‐year sample on which the policy performance was tested.

This publication has 11 references indexed in Scilit: