PQ-Net: Periodic Quantum Networks for Multivariate Time Series Forecasting

Yijun Wang, Qiyuan Zhuang, Xiaogang Qiang, Xiu-Shen Wei

Keywords: Quantum Machine Learning, Multivariate Time Series Forecasting, Data Re-uploading Quantum Circuits, Periodic Quantum Networks

Abstract: Multivariate time series forecasting (MTSF) requires capturing both periodic structures and cross-channel dependencies from complex temporal signals. To address this challenge, we propose Periodic Quantum Networks (PQ-Net), a quantum--classical hybrid forecasting architecture that integrates a learnable temporal query mechanism for cycle alignment and a channel aggregation module for modeling inter-channel correlations. PQ-Net preserves permutation equivariance across variables while jointly representing frequency-domain and cross-channel information in a principled manner. At the core of PQ-Net lies the Data Re-uploading Quantum Circuit (DRQC), whose representational capacity we theoretically analyze. We show that DRQC are mathematically equivalent to truncated Fourier series, enabling natural encoding of periodic patterns, while quantum entanglement provides a means to capture inter-variable dependencies. This interpretation establishes DRQC as a rigorous and interpretable foundation for periodic modeling within PQ-Net. Extensive experiments on twelve real-world datasets demonstrate that PQ-Net consistently achieves state-of-the-art forecasting accuracy over strong classical and quantum baselines, and preliminary hardware results further validate its practicality on real quantum devices.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 2038