A Multi-scale Feature Fusion Method for Demand Forecasting in Supply Chain Management

For data-driven supply chain product demand forecasting, traditional methods tend to oversimplify the complexities of supply chain data, often neglecting its inherent heterogeneity and failing to extract valuable features. In response, we introduce a novel approach called Multi-Scale Feature Fusion Approach with Attention Mechanisms (MO-MSA) for supply chain product demand forecasting. MO-MSA leverages the distinctive characteristics of historical supply chain data, conducts multiscale feature extraction on time series data, and employs a string-parallel hybrid multi-step forecasting strategy to significantly enhance forecasting accuracy. Our experimental validation on the Corporación Favorita Grocery Sales Forecasting dataset demonstrates that MO-MSA outperforms various existing methods, delivering superior average forecasting.