Quantum-enhanced time series forecasting: integrating amplitude-phase dropout attention model

The integration of quantum computing and machine learning has led to innovative algorithms in quantum machine learning that outperform traditional methods in time series forecasting. This paper introduces the Quantum Amplitude-Phase Dropout Attention Model (QAPDAM), which combines Amplitude-Phase Decomposition (APDM) measurement and quantum dropout attention to improve prediction accuracy and robustness while minimizing circuit parameters. The QAPDAM model consists of a Variational Quantum Circuit (VQC) for predictions and an APDM-based dropout self-attention module for feature extraction. Data is encoded into quantum states and processed with a parameter-efficient APDM self-attention mechanism to calculate query, key, and value. Quantum dropout is applied to adjust attention weights dynamically, reducing overfitting and adapting to data variability. Forecasts are obtained through circuit evolution and measurement. We tested QAPDAM's performance in Electric Transformer Temperature dataset, finding it surpasses other models in terms of predictive accuracy with a 25% reduction in circuit parameters.