Determining Inflow Forecast Horizon for Reservoir Operation

A critical study issue to incorporate imperfect forecast in real-time reservoir operation is determining the forecast horizon. In this study, properties for the longest forecast horizon (LFH) and the effective forecast horizon (EFH) are derived from a multistage, deterministic optimization model for the operation of a single water supply reservoir with a concave benefit function. The LFH addresses the question of how long a forecast is sufficient to make an optimal reservoir release decision for the current stage if the effect of forecast uncertainty is not considered. The EFH represents a forecast horizon with the information for decision making as much as allowed by uncertainty effect control, which is set as prescribed decision reliability quantified by the error bound (i.e., the largest difference between the optimal release decisions made under any two inflow scenarios). The properties of LFH and EFH are used to specify the criteria and design the procedures for determining EFH and LFH. A hypothetical but typical case study is used to demonstrate the criteria and procedures. Both theoretical analysis and the case study results show that LFH and EFH are affected by multiple factors such as the reservoir capacity, inflow variability, forecast uncertainty, maximum allowable error bound, and ending storage estimate accuracy. LFH is longer with a larger capacity, smaller inflow variability, and smaller forecast uncertainty, and EFH is longer with smaller forecast uncertainty, larger error bound, and more accurate ending storage estimates.