Uncertainty-informed multi-reservoir flood control optimization: A probabilistic forecasting and stochastic decision-making framework

Abstract

Forecast uncertainty is a critical factor influencing the risk associated with reservoir flood control operations. Traditional assumptions of Gaussian-distributed forecast residuals often fail to capture their variability, leading to challenges in risk assessment and decision-making. To address this issue, this study developed a stochastic inflow scenario generation algorithm that accounts for heteroscedastic residuals and coupled it with an NSGA-III-based reservoir optimization model. Furthermore, a decision-making approach integrating the Fuzzy Analytic Hierarchy Process (FAHP) and the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method was employed to perform multi-objective decision-making, taking into account the fuzziness of decision preferences. The proposed framework was tested on a parallel reservoir system located in the lower reaches of the Yellow River Basin, China. The key findings are as follows: (1) Forecast residual behavior exhibits significant variability across different flow magnitudes. (2) The conventional assumption of Gaussian-distributed forecast residuals underestimates flood uncertainty and its associated risks, particularly during extreme flood events. (3) Incorporating forecast uncertainty into reservoir flood control operations enhances risk mitigation in multi-reservoir systems.