Scholarlink

Purpose <p>Shanghai HIgh repetition rate XFEL aNd Extreme light facility (SHINE) is the first hard X-ray free-electron laser facility in China. As the volume of process variables (PVs) stored in the Archiver Appliance (AA) increases, managing this data presents challenges for web-based tools. To address these issues, a data management and analysis system was developed to streamline PV management operations through user-friendly interfaces and a backend dashboard.</p> Methods <p>The system was built using the Django framework for data processing. It incorporates several advanced features, including customized logging and integration with the DeepSeek model, which facilitates rapid fault localization, root cause analysis, and optimization recommendations. Additionally, abnormal user operations are detected through a combination of the Isolation Forest algorithm and the Interquartile Range (IQR) anomaly detection method. User interfaces and a monitoring dashboard were created using HTML and Grafana.</p> Results <p>During performance tests, the system handled 328.99 transactions per second with 100 concurrent requests. The F1-scores for detecting abnormal add and delete operations reached 0.9828 and 1.0000, respectively.</p> Conclusion <p>The system enhances PV management security, simplifies user operations, and reduces complexity. This demonstrates the meaningful integration of artificial intelligence for managing beam accelerator data, resulting in greater efficiency and reliability for the control system.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Operational data management and analysis system based on the IQR-isolation forest algorithm and DeepSeek model for SHINE

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Rui Wang,
  • Guanghua Chen,
  • Jianfeng Chen,
  • Huihui Lv,
  • Haifeng Miao,
  • Yingbing Yan

摘要

Purpose

Shanghai HIgh repetition rate XFEL aNd Extreme light facility (SHINE) is the first hard X-ray free-electron laser facility in China. As the volume of process variables (PVs) stored in the Archiver Appliance (AA) increases, managing this data presents challenges for web-based tools. To address these issues, a data management and analysis system was developed to streamline PV management operations through user-friendly interfaces and a backend dashboard.

Methods

The system was built using the Django framework for data processing. It incorporates several advanced features, including customized logging and integration with the DeepSeek model, which facilitates rapid fault localization, root cause analysis, and optimization recommendations. Additionally, abnormal user operations are detected through a combination of the Isolation Forest algorithm and the Interquartile Range (IQR) anomaly detection method. User interfaces and a monitoring dashboard were created using HTML and Grafana.

Results

During performance tests, the system handled 328.99 transactions per second with 100 concurrent requests. The F1-scores for detecting abnormal add and delete operations reached 0.9828 and 1.0000, respectively.

Conclusion

The system enhances PV management security, simplifies user operations, and reduces complexity. This demonstrates the meaningful integration of artificial intelligence for managing beam accelerator data, resulting in greater efficiency and reliability for the control system.