📄 中文摘要
时间序列预测在各个领域的实际应用中面临重大挑战。基于时间序列的分解，提出了一种改进机器学习模型架构的方法，以实现更好的多变量时间序列预测。研究重点分别分析趋势和季节性成分，并探索减少预测误差的解决方案。认识到可逆实例归一化仅对趋势成分有效，季节性成分则采用直接应用主干模型的方式，而不进行任何归一化或缩放处理。通过这些策略，成功降低了现有最先进模型的误差值。
重新审视季节性趋势分解以增强时间序列预测
📄 English Summary
Revisiting the Seasonal Trend Decomposition for Enhanced Time Series Forecasting
Time series forecasting poses significant challenges in real-world applications across various domains. This research enhances the architecture of machine learning models based on the decomposition of time series for improved multivariate time series forecasting. The focus is on analyzing the trend and seasonal components separately, investigating solutions to reduce prediction errors. Recognizing that reversible instance normalization is effective only for the trend component, a different approach is applied to the seasonal component by directly utilizing backbone models without any normalization or scaling procedures. These strategies successfully reduce the error values of existing state-of-the-art models.