Limitations of deterministic optimization applied to reservoir operations

Deterministic optimization can produce suboptimal reservoir-control policie s by failing to incorporate adequately the impact of low-probability events . Resulting operating policies may not efficiently balance the costs of rat ioning, minor flooding, or other short-term impacts with the severe impacts of extreme flood or drought. This occurs when deterministic optimization i s applied to systems that are not "certainty equivalent." This paper demons trates this by contrasting control policies developed using deterministic o ptimization of inflow forecasts with control policies using stochastic opti mization of probabilistic inflows. For a range of hypothetical reservoir-co ntrol problems, it is observed that deterministic optimization results in c osts that are greater on average and that are much greater for extreme even ts, particularly for reservoir systems with limited storage capacity and fo r objectives described by nonquadratic functions. This is true even when fo recasts are relatively accurate, such as when streamflows are highly autoco rrelated.