Occam's Razor in Residential PV-Battery Systems: Theoretical Interpretation, Practical Implications, and Possible Improvements

TL;DR

This work analyzes a widely used rule-based EMS for residential PV-BSS, known as Occam's control, by embedding it in online convex learning to derive a theoretical performance bound. It then introduces MOS, a momentum-augmented online-learning dispatch that preserves the same computational footprint while delivering superior $L_2^2$ performance and stronger peak-reduction. Through real-world data and a 24-hour rolling-horizon baseline, MOS consistently outperforms Occam's control and, in most cases, rolling-horizon optimization, while maintaining a tiny data/storage footprint and low computation time. These findings suggest that online-learning approaches can deliver practical, economically favorable improvements for residential PV-BSS without increasing complexity, aiding distribution-grid stability in high PV scenarios.

Abstract

This paper presents a theoretical interpretation and explores possible improvements of a widely adopted rule-based control for residential solar photovoltaics (PV) paired with battery storage systems (BSS). The method is referred to as Occam's control in this paper, given its simplicity and as a tribute to the 14th-century William of Ockham. Using the self-consumption-maximization application, it is proven that Occam's control is a special case of a larger category of optimization methods called online convex learning. Thus, for the first time, a theoretical upper bound is derived for this control method. Furthermore, based on the theoretical insight, an alternative algorithm is devised on the same complexity level that outperforms Occam's. Practical data is used to evaluate the performance of these learning methods as compared to the classical rolling-horizon linear/quadratic programming. Findings support online learning methods for residential applications given their low complexity and small computation, communication, and data footprint. Consequences include improved economics for residential PV-BSS systems and mitigation of distribution systems' operational challenges associated with high PV penetration.

Figures (6)

• Figure 1: Residential PV-BSS structure.
• Figure 2: $e^{(-|P_{\mathrm{b}, t}|)}$ function.
• Figure 3: Annual 30-minute profiles for residential load and PV.
• Figure 4: Performance comparisons.
• Figure 5: Normalized avg. performance (wider better).