Leveraging Model Predictive Control Methods and Ensemble Forecasts to Inform Reservoir Release Decision-Making

Abstract Submission: Inflow forecasts are critical for improving reservoir operations, enabling a better balance of water supply, flood control, and environmental objectives. While deterministic forecasts predict expected inflow conditions, they fail to capture forecast uncertainty. Conditioning release decisions on deterministic forecasts without accounting for forecast uncertainty can result in an overestimation of forecast value to improve system outcomes. Uncertainty is often represented in the form of ensemble forecasts, where each ensemble represents a plausible trajectory of future inflow conditions. Advancements in model predictive control (MPC) methods have emerged as a powerful approach to incorporate ensemble forecasts into reservoir operations and can optimize release decisions across a forecast horizon to support the real-time control of reservoirs. This study explores the effectiveness of multi-scenario and tree-based MPC methods to leverage forecast information and its uncertainty over the forecast horizon to improve release decision-making in multi-objective water systems. Releases are optimized over a rolling forecast horizon using MPC methods conditioned on deterministic, ensemble, and perfect inflow forecasts. Results are compared against releases conditions on the current rule curve operations of the dam with deterministic and perfect inflow forecasts. The application of these methods in the context of Forecast Informed Reservoir Operations (FIRO) aims to increase total water storage in watersheds, as well as provide additional benefits for flood control, environmental, and recreational interests.

Learning Objectives/Expected Outcome (Optional) :