一种基于混合对比学习网络的时间序列数据分类系统
By combining data augmentation and adaptive loss fusion mechanisms through a hybrid contrastive learning network, the problem of inaccurate time series data representation is solved, the robustness and classification performance of the model are improved, and more efficient time series classification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEILONGJIANG UNIV
- Filing Date
- 2025-02-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods struggle to learn accurate time series data representations, resulting in poor time series classification performance. In particular, under challenges of noise and complexity, the models lack robustness and applicability.
A time-series data classification system based on a hybrid contrastive learning network is adopted, including a data acquisition module, a dual data augmentation module, and a convolutional neural network module. Multiple data views are generated through a jitter-scaling or permutation-jitter strategy. The system combines perturbation contrastive and temporal contrastive losses and utilizes a KAN encoder and adaptive loss fusion mechanism to improve the robustness and generalization ability of the model.
By fusing dual data augmentation and adaptive loss, richer time series representations are captured, improving model stability and classification performance, reducing the impact of noise, and enhancing the accuracy and applicability of time series classification.
Smart Images
Figure CN120105190B_ABST