IEEE International Conference on Advances in Computing Research on Science Engineering and Technology (ACROSET)
Quantum Kernel-Enhanced Federated Learning for Non-IID Solar Irradiance Forecasting Across Distributed Microgrids
Amrita Kundu, Vegi Fernando A, Souvik Ghosh, Mithaguru
Abstract
Accurate solar irradiance forecasting is critical for the stable operation of distributed microgrids, yet it remains challenged by heterogeneous (non-IID) local data distributions across geographically dispersed sites. Federated learning (FL) offers a privacy-preserving framework for collaborative model training without raw data sharing; however, classical local models struggle to capture complex non-stationary temporal patterns under severe distribution shift. In this paper, we propose QK-FedSolar, a novel federated learning framework that replaces classical kernel operations in local LSTM cells with quantum kernel methods—instantiated via parameterized quantum circuits (PQCs) and quantum feature maps—to form Quantum Kernel Long Short-Term Memory (QK-LSTM) units at each federated node. Leveraging the exponentially large Hilbert space of quantum kernels, our approach achieves superior inter-node generalization under the same communication budget as classical counterparts. We evaluate QK-FedSolar on a real-world dataset spanning 12 microgrid nodes across three climate zones, demonstrating a 23.7% reduction in root mean square error (RMSE) over FedAvg+LSTM and an 18.4% improvement over FedProx+Transformer baselines, while maintaining differential privacy guarantees (ε = 1.0, δ = 10‒5). To our knowledge, this is the first integration of quantum kernel methods into a federated, privacy-preserving energy forecasting context, bridging quantum machine learning and sustainable energy systems.