IJCAS

In the cross-border e-commerce industry chain for eco-friendly electronic products, the prediction of supply chain demand plays a pivotal role. It is essential to accurately forecast the future demand for each ecofriendly electronic product in various warehouses, enabling timely inventory distribution across the globe, reducing carbon emissions from logistics, and significantly enhancing customer experience. This paper explores multiple demand prediction algorithms for supply chain demand for eco-friendly electronic products. Addressing the shortcomings in existing methods, such as anomaly detection, vectorized representation of product information, and multi-step forecasting, we propose three innovative improvements to enhance prediction accuracy while considering environmental factors. Firstly, we introduce a linear regression method based on Huber Loss for data processing, effectively identifying anomalies in historical sales data. Secondly, we present a product Embedding vector representation method based on Pearson’s correlation coefficient. This method not only learns low-dimensional vector representations of product information but also reveals the sales correlations and competitiveness among products, aiding in optimizing the market placement of eco-friendly products. Lastly, we optimize the multi-time step demand prediction method by proposing a deep learning algorithm based on Seq2Seq+Attention, enabling end-to-end temporal multi-step forecasting to adapt to the dynamic and complex demands of the eco-friendly market. Experiments conducted on actual historical sales data of eco-friendly electronic products from an e-commerce platform demonstrate that the methods proposed in this paper surpass traditional time series prediction models and machine learning regression models in forecasting accuracy, contributing to more environmentally friendly and efficient supply chain management for electronic products.

Keywords: Attention, supply chain, demand prediction, Seq2Seq.