Abstract: To address the challenges of local optima, high-dimensional feature, and insufficient diagnostic accuracy in fault diagnosis of crude oil near-infrared (NIR) spectroscopy analysis system, an ensemble CSSOA-SVM-based fault diagnosis method is proposed. The chaos sparrow search optimization algorithm (CSSOA) is introduced to optimize the support vector machine (SVM) parameters, overcoming the local optima limitations of the traditional sparrow search algorithm (SSA) while preserving its rapid convergence, thus enhancing classification performance. By integrating the AdaBoost algorithm, multiple CSSOA-SVM base classifiers are combined, with dynamic adjustments to sample and classifier weights improving the recognition accuracy and robustness for complex fault patterns. Experimental results demonstrate that the proposed ensemble CSSOA-SVM model achieves a diagnostic accuracy of 95.48% across six common fault types, outperforming traditional methods in accuracy, convergence speed, and robustness, offering an effective solution for fault diagnosis in crude oil NIR spectroscopy analysis system.

Key words: crude oil near-infrared spectroscopy analysis system, fault diagnosis, chaos sparrow search optimization algorithm, support vector machine optimization, ensemble learning