Energy Demand Forecasting using Quantum Machine Learning

In collaboration with E.ON, this project investigated the application of Quantum Machine Learning (QML) methods to improve demand forecasting for modern electrical grids. Accurate energy forecasting is essential for grid stability and operational planning, particularly as renewable generation and distributed energy resources introduce increasingly complex and nonlinear dynamics into power systems.

The WISER research team developed two quantum learning frameworks designed to model multi-stream energy time-series data, capturing temporal correlations, seasonal effects, and cross-variable dependencies that often challenge conventional forecasting approaches. The first framework introduces a kernelized Quantum Reservoir Computing architecture, where coupled quantum reservoirs encode multiple time-series inputs simultaneously. The second approach applies Gaussian Process Regression within a quantum-enhanced kernel framework. Instead of using fidelity-based quantum kernels, the model employs projected quantum kernels derived from local expectation values of reduced density matrices, improving noise resilience and computational efficiency on near-term quantum devices.

Experimental evaluations were conducted on both simulators and quantum hardware across multiple system sizes. The quantum reservoir computing framework achieved a best average mean absolute error (MAE) of 0.0811 on simulator experiments and 0.1524 on hardware implementations. The quantum Gaussian process model achieved MAE values as low as 0.078 on simulator and 0.082 on hardware, while larger experiments demonstrated continued predictive capability. In a 100-qubit utility-scale experiment, the model achieved high-precision forecasting (MAE < 0.15) in 36% of cases and viable predictive accuracy in 76% of evaluations.

These results demonstrate the feasibility of scalable quantum multi-output forecasting across systems ranging from small experimental devices to larger quantum simulations. The project highlights the potential of hybrid quantum–classical learning frameworks to enhance predictive modeling for complex energy systems, offering new approaches for improving demand forecasting and grid management in next-generation energy infrastructure.

WISER Research Fellows: Mackenson Polche, Varun Puram, Aditi Lal, Weronika Golletz, Joan Arrow