<ul><li>Forecasting non-stationary time series is challenging due to changing statistical properties over time.</li><li>Existing methods overlook the multi-component nature of time series, where different components have distinct non-stationary behaviors.</li><li>Wavelet-based Disentangled Adaptive Normalization (WDAN) is a model-agnostic framework proposed to address non-stationarity in time series forecasting.</li><li>WDAN utilizes discrete wavelet transforms to handle low-frequency trends and high-frequency fluctuations with tailored normalization strategies for each part, showing improved forecasting accuracy.</li></ul>

Wavelet-based Disentangled Adaptive Normalization for Non-stationary Times Series Forecasting

Discover more