Ensemble Forecasts with Blocked K-Fold Cross-Validation in Multi-Objective Water Systems Control

In this paper, we contribute the Parallel Ensemble foreCAst coNtrol (PECAN) algorithm to enhance multi-objective water systems control through the integration of Ensemble Forecast and data-driven control techniques. This integration allows evolving parallel system simulations for each forecast ensemble member to maximize the benefit provided by the probabilistic forecasts. To avoid potential overfitting and ensure the generalization capabilities of the designed solutions, we also implement a Blocked K-Fold cross-validation. Testing on the Lake Como water reservoir system shows that PECAN improves the controller performance by 8.2% with respect to traditional methods relying solely on forecast ensemble averages and by 26.8% over approaches that do not use any forecast. These results highlight the benefits of ensemble-based techniques for controlling water systems under highly variable hydroclimatic conditions.