基于循环神经网络的电商报单时序趋势预测方法
By improving STL decomposition and Hax process, combined with recurrent neural networks, the modeling problems of trends, cycles and event intensity in e-commerce order reporting time series are solved, achieving more stable predictions and higher reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI PANTI TECHNOLOGY CO LTD
- Filing Date
- 2025-12-04
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to simultaneously handle trend components, cyclical components, and short-term jump components in e-commerce order reporting timelines. Furthermore, traditional methods exhibit unstable prediction results when faced with high noise and activity interference, lacking effective modeling of event intensity and resulting in insufficient prediction reliability.
An improved STL decomposition method is used to identify daily and weekly cycle structures. A composite trigger structure is constructed by combining an improved Hax process. A recurrent neural network is used for dynamic prediction, and a confidence vector containing probability, stability, data quality, and temporal continuity is output.
It improves the accuracy of trend stripping, enhances the adaptability to activity and behavioral fluctuations, strengthens event-triggered modeling capabilities, provides interpretable prediction credibility, and improves the practical value of prediction results.
Smart Images
Figure CN121563649B_ABST