Frequency-Constrained Learning for Long-Term Forecasting
By: Menglin Kong, Vincent Zhihao Zheng, Lijun Sun
Potential Business Impact:
Predicts future events by understanding repeating patterns.
Many real-world time series exhibit strong periodic structures arising from physical laws, human routines, or seasonal cycles. However, modern deep forecasting models often fail to capture these recurring patterns due to spectral bias and a lack of frequency-aware inductive priors. Motivated by this gap, we propose a simple yet effective method that enhances long-term forecasting by explicitly modeling periodicity through spectral initialization and frequency-constrained optimization. Specifically, we extract dominant low-frequency components via Fast Fourier Transform (FFT)-guided coordinate descent, initialize sinusoidal embeddings with these components, and employ a two-speed learning schedule to preserve meaningful frequency structure during training. Our approach is model-agnostic and integrates seamlessly into existing Transformer-based architectures. Extensive experiments across diverse real-world benchmarks demonstrate consistent performance gains--particularly at long horizons--highlighting the benefits of injecting spectral priors into deep temporal models for robust and interpretable long-range forecasting. Moreover, on synthetic data, our method accurately recovers ground-truth frequencies, further validating its interpretability and effectiveness in capturing latent periodic patterns.
Similar Papers
Filter then Attend: Improving attention-based Time Series Forecasting with Spectral Filtering
Machine Learning (CS)
Makes computer predictions of future events more accurate.
FreqFlow: Long-term forecasting using lightweight flow matching
Machine Learning (CS)
Predicts future data faster and with less computer power.
Fre-CW: Targeted Attack on Time Series Forecasting using Frequency Domain Loss
Machine Learning (CS)
Makes computer predictions harder to trick.