Evolving Multi-Scale Normalization for Time Series Forecasting under Distribution Shifts
This work is significant for researchers and practitioners in time series forecasting, offering a method to improve accuracy and robustness in the presence of distribution shifts, which is a common and critical problem.
This paper addresses the challenge of accurate long-term time series forecasting under complex distribution shifts. The authors propose EvoMSN, a model-agnostic framework that uses multi-scale statistics prediction and adaptive ensembling for flexible normalization and denormalization, coupled with an evolving optimization strategy. EvoMSN improves the performance of five mainstream forecasting methods on benchmark datasets and outperforms existing advanced normalization and online learning approaches.
Complex distribution shifts are the main obstacle to achieving accurate long-term time series forecasting. Several efforts have been conducted to capture the distribution characteristics and propose adaptive normalization techniques to alleviate the influence of distribution shifts. However, these methods neglect the intricate distribution dynamics observed from various scales and the evolving functions of distribution dynamics and normalized mapping relationships. To this end, we propose a novel model-agnostic Evolving Multi-Scale Normalization (EvoMSN) framework to tackle the distribution shift problem. Flexible normalization and denormalization are proposed based on the multi-scale statistics prediction module and adaptive ensembling. An evolving optimization strategy is designed to update the forecasting model and statistics prediction module collaboratively to track the shifting distributions. We evaluate the effectiveness of EvoMSN in improving the performance of five mainstream forecasting methods on benchmark datasets and also show its superiority compared to existing advanced normalization and online learning approaches. The code is publicly available at https://github.com/qindalin/EvoMSN.