Parameter-performance scaling of hybrid quantum-classical policies for demand-charge-aware battery control
Abstract
Abstract Hybrid quantum-classical models are often proposed as compact alternatives to classical neural networks; however, the claim that they require fewer parameters has rarely been tested across full model size ranges against strong classical baselines. In this study, this claim was tested on a realistic energy-control task: 15-min, full-month control of behind-the-meter batteries that must minimize commercial electricity demand charges using 549 simulated office buildings under a standardized utility tariff. Three policy families—a classical neural network and two hybrid quantum-circuit variants sharing the same inputs and control head—were trained by behavioral cloning from a perfect-foresight optimizer and compared at four parameter budgets on a locked held-out test set. One compact hybrid design (~ 1,400 parameters) provided the best control performance among all tested model and modestly outperformed the size-matched classical network. However, the advantage is small, does not hold for other sizes, and depends on a task-specific control head rather than on the quantum circuit itself. The classical network remains a strong baseline and runs faster. Therefore, this study reports a narrow, bounded parameter-efficiency benefit rather than a general quantum advantage and outlines which hardware-level validation would still be required. All quantum circuits were evaluated using exact simulations.
Read the paper
Related studies
- Chaos and quantum thermalization
- The Dalton quantum chemistry program system
- Cord Blood Leptin and Adiponectin as Predictors of Adiposity in Children at 3 Years of Age: A Prospective Cohort Study
- Quantitative metabolomics by H-NMR and LC-MS/MS confirms altered metabolic pathways in diabetes
- Stable adiabatic quantum batteries
- A novel self-lipid antigen targets human T cells against CD1c(+) leukemias
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.