Scholarlink

<p>This paper presents an anomaly diagnosis framework for the Crude Oil Online Near-Infrared Spectroscopy Analysis System (CO-NIRS), addressing challenges such as high-dimensional features, susceptibility to local optima, and insufficient diagnostic accuracy. The framework ensembles an Improved Chaos Sparrow Search Optimization Algorithm-enhanced Support Vector Machine (ICSSOA-SVM), where ICSSOA is developed to overcome local optima limitations while retaining rapid convergence, thereby optimizing SVM parameters and enhancing classification performance and stability. Additionally, the AdaBoost algorithm is employed to ensemble multiple ICSSOA-SVM base classifiers, dynamically adjusting sample and model weights to improve the recognition of complex anomaly patterns and bolster model robustness. Experimental results on a 310-spectrum industrial CO-NIRS dataset covering six common anomaly types show that the ensemble ICSSOA-SVM model achieves a test accuracy of 95.70%, with a two-sided 95% Wilson score confidence interval of 89.46–98.31%. The ensemble model achieves the highest test accuracy among the evaluated classifiers while retaining second-level parameter optimization. A paired sensitivity analysis with Holm correction further indicates a statistically significant improvement over conventional SVM. Because the dataset is limited and commercially restricted, the conclusions are framed as an industrial case-study validation rather than a universal claim across all crude-oil streams.</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

Anomaly diagnosis of crude oil online near-infrared spectroscopy analysis system with ensemble ICSSOA-SVM

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Xiaojie Gu,
  • Yongjian Wang,
  • Xisong Chen

摘要

This paper presents an anomaly diagnosis framework for the Crude Oil Online Near-Infrared Spectroscopy Analysis System (CO-NIRS), addressing challenges such as high-dimensional features, susceptibility to local optima, and insufficient diagnostic accuracy. The framework ensembles an Improved Chaos Sparrow Search Optimization Algorithm-enhanced Support Vector Machine (ICSSOA-SVM), where ICSSOA is developed to overcome local optima limitations while retaining rapid convergence, thereby optimizing SVM parameters and enhancing classification performance and stability. Additionally, the AdaBoost algorithm is employed to ensemble multiple ICSSOA-SVM base classifiers, dynamically adjusting sample and model weights to improve the recognition of complex anomaly patterns and bolster model robustness. Experimental results on a 310-spectrum industrial CO-NIRS dataset covering six common anomaly types show that the ensemble ICSSOA-SVM model achieves a test accuracy of 95.70%, with a two-sided 95% Wilson score confidence interval of 89.46–98.31%. The ensemble model achieves the highest test accuracy among the evaluated classifiers while retaining second-level parameter optimization. A paired sensitivity analysis with Holm correction further indicates a statistically significant improvement over conventional SVM. Because the dataset is limited and commercially restricted, the conclusions are framed as an industrial case-study validation rather than a universal claim across all crude-oil streams.