Balancing Forecast Accuracy and Switching Costs in Online Optimization of Energy Management Systems

TL;DR

This work studies how switching costs influence the integration of forecasting and online optimization in energy management systems. It formalizes a Fixed Horizon Control framework with rolling-origin forecasts and introduces forecast-stability metrics, notably the novel Scenario Distribution Change (SDC) for probabilistic forecasts. The paper provides deterministic and stochastic performance bounds showing a trade-off between switching costs and forecast error accumulation, and demonstrates that forecast stability can reduce the adverse impact of switching costs, sometimes favoring longer commitment periods. Empirical validation on CityLearn battery scheduling confirms that equal forecast/plan update frequencies perform best under switching costs, and that stochastic optimization extends viable commitment lengths by mitigating forecast sensitivity, with practical implications for robust EMS design.

Abstract

This study investigates the integration of forecasting and optimization in energy management systems, with a focus on the role of switching costs -- penalties incurred from frequent operational adjustments. We develop a theoretical and empirical framework to examine how forecast accuracy and stability interact with switching costs in online decision-making settings. Our analysis spans both deterministic and stochastic optimization approaches, using point and probabilistic forecasts. A novel metric for measuring temporal consistency in probabilistic forecasts is introduced, and the framework is validated in a real-world battery scheduling case based on the CityLearn 2022 challenge. Results show that switching costs significantly alter the trade-off between forecast accuracy and stability, and that more stable forecasts can reduce the performance loss due to switching. Contrary to common practice, the findings suggest that, under non-negligible switching costs, longer commitment periods may lead to better overall outcomes. These insights have practical implications for the design of intelligent, forecast-aware energy management systems.

Figures (8)

• Figure 1: Diagrams of the Online Optimization algorithms. RHC: Receding Horizon Control, FHC: Fixed Horizon Control
• Figure 2: Rolling Origin Forecasting with Fixed Input Window and Forecast Horizon. The diagram illustrates how forecasts with horizon $H = 2$ are updated every $v_f = 1$ time steps, creating overlaps between prediction windows.
• Figure 3: Diagrams demonstrating the concepts of vertical and horizontal stability between predictions.
• Figure 4: Illustration of the FHC expected competitive difference as a function of the commitment level for (a) Low and High switching costs with predictions of exponentially decaying correlations, and (b) Stochastic and Deterministic optimizations. Stochastic scenarios exhibit a lower level of forecast error sensitivity ($B_s < B$), shifting the optimal commitment level to higher values.
• Figure 5: System setup (a) and data flow (b) for the battery scheduling problem. The data flow diagram illustrates how input features are processed through forecasting models with varying update frequencies to generate load predictions, which then inform the optimization stage to produce battery control actions.