Predicting Inventory Fluctuation Patterns of Natural Fiber Products during Multi-Platform E-Commerce Promotion Cycles by Integrating T-GCN and Time Window Mechanism

This study proposes a spatiotemporal forecasting framework for inventory fluctuations of natural fiber products during multi-platform e-commerce promotion cycles. A cross-platform inventory correlation graph is constructed to encode inter-platform relationships, integrating inventory change trends and promotional rhythm into a weighted adjacency matrix. An adaptive time window mechanism dynamically segments inventory sequences based on local fluctuation intensity, enabling sensitive detection of sudden changes. Temporal Graph Convolutional Networks (T-GCN) are then applied to jointly model temporal dependencies and cross-platform graph features. Experiments demonstrate superior performance, achieving MAE=4.85, RMSE=7.12, and CPI=0.18, outperforming ARIMA, LSTM, and GRU models, while effectively tracking abrupt inventory variations with minimal response delay. The framework can be integrated with real-time data acquisition systems, edge-computing platforms, and networked information processing infrastructures, providing an engineering-oriented solution for inventory prediction, supply chain optimization, and decision support in dynamic e-commerce environments. The approach emphasizes spatiotemporal feature fusion and adaptive response to high-frequency inventory fluctuations, bridging predictive modeling with practical engineering systems.