EMD-SOFTS-based demand prediction for supply chain

Supply chain demand prediction can optimize inventory and improve efficiency. In the increasingly complex and volatile market environment, accurate demand prediction is crucial for enterprise operation and decision-making. But the dimensionality of raw time series data is often low. Current methods cannot fully capture the complex patterns and potential relationships in the data. To address this challenge, we proposed an Empirical Mode Decomposition (EMD) and Series-cOre Fused Time Series (SOFTS) based supply chain demand prediction method EMD-SOFTS in this paper. Firstly, we adopted EMD to decompose time sequence into multiple independent modalities and expand the data dimension. Demand features at different time scales and frequencies can be efficiently extracted. In addition, we take the SOFTS as the basic structure. SOFTS extracts correlations different channels with a STar Aggregate-Redistribute (STAR) module. It addresses the challenge of lacking inter channel correlation in traditional channel independent methods, while avoids the insufficient robustness of channel dependent methods. Finally, Agent Attention mechanism is introduced to fuse features from different channels to improve the predictive ability of the model. We conducted comprehensive comparative experiments and ablation analysis on the Alibaba Cloud Infrastructure Supply Chain Inventory Management Decision Dataset. Among the 8 compared methods, the proposed EMD-SOFTS has the lowest prediction error. The experimental results show that the proposed method can effectively capture the complex patterns of supply chain demand and significantly improve prediction accuracy.