Decoupling Temporal Resolution From Forecast Accuracy in MPC-Based Energy Storage Scheduling for Microgrids

Abstract

Model predictive control (MPC) of microgrid battery energy storage is shaped by two design choices that are almost always refined together: the temporal resolution of the schedule and the accuracy of the net-load forecasts it uses. This paper decouples these two factors. We show that the real-time regulating burden on the energy storage decomposes exactly into an intra-interval geometric mismatch between the step-function dispatch and the continuous one-minute signal, and a prediction-error component in the interval averages, which constitute two problems with different remedies. The calibrated sweep uses a multi-year, 113-home forecast panel and synchronized one-minute telemetry from ten homes over 72 h. At the nominal point, schedule refinement lowers regulating throughput by 312 kWh versus 126 kWh for forecast-error halving, giving a $2.5\times $ effect ratio. Across the tested PV-penetration and load-volatility perturbations, the ratio spans 1.0–6.8, so $2.5\times $ is not a general constant. The geometric mismatch accounts for about 84% of the regulating burden at hourly resolution and 70% at 15-minute resolution; refining the schedule therefore removes this dominant component, leaving only a prediction-error floor that no forecaster can eliminate. Under the baseline ERCOT prices, scheduled BESS cycling is zero, while finer resolution reduces real-time regulating throughput. This does not establish a general degradation reduction or a ranking of active-storage dispatch strategies. Controller costs are benchmarked across deterministic, robust, and scenario-based stochastic MPC in this baseline regime.