Research on performance time series and the discovery of data patterns, enabling intelligent applications in industry, AI, and smart healthcare.
We focus on advancing time series intelligence through methodological innovation and real-world deployment. Our work bridges theoretical advances in pattern discovery with robust implementations in industrial and clinical settings.
See collaborators' information: Click Me.
Core Research Areas
Our research centers on three interconnected pillars:
🔍 Performance Time Series Analysis
We study nonstationary, high-noise signals from industrial and physiological systems using adaptive modeling techniques to enable reliable monitoring under dynamic conditions.
🧩 Data Pattern Recognition
We identify latent temporal structures — such as trends, rhythms, and anomalies — by combining signal decomposition with interpretable machine learning for robust diagnostic inference.
⚙️ Hyperparameter Optimization
We automate model configuration through hybrid search strategies (e.g., coordinate ascent-enhanced grid search) to improve deep learning performance in resource-constrained environments.
Together, these areas form a coherent methodology for building intelligent systems grounded in rigorous data science.
Scholarly & Technical Contributions
Our research produces two complementary types of outcomes:
Publications
We have published peer-reviewed works on time series analysis, deep learning optimization, and AI-driven diagnostics, including:
- TRLLD: A threshold recognition-based load level detection algorithm for nonstationary load time series (CMC, 2025).
- Trend-aligned Feature Correlation: A holistic metric for feature relevance in time series (SPIE BDCIA, 2025).
- Pretrained CNNs for Lorenz Plot Classification: Enabling early cardiac risk assessment from RR intervals (JIM, 2025).
- Hyperparameter Optimization: Grid search enhanced with coordinate ascent for deep convolutional networks (ICMIDA, 2024).
Deliverables
Our technical impact is realized through patented inventions and copyrighted software systems:
- Patents: 12 granted IP rights, including a database load analysis method (CN112307042B), a reinforcement learning-enhanced image recognition training system (CN213361678U), and multiple hardware protection and monitoring devices.
- Software Copyrights: 17 registered systems, such as the Multi-terminal Health Monitoring Platform, ACO-based Path Optimization Software for Smart Warehousing, and Real-time Visualization & Management System for Massive Data.
- Deployed Tools: Systems supporting load balancing, disaster recovery, HRV/Lorenz-RR analysis, and secure identity authentication in enterprise environments.
For collaboration inquiries, please contact: song.qq@scifs.ac.cn