Currently, excellent performance has been demonstrated by machine learning-based fault diagnosis models in detecting latent faults within high-voltage switchgear. However, primary focus is placed on common latent faults by most diagnostic models, leaving less detectable faults such as excessive grease coating, abnormal contact ablation, and floating potential discharge of metal parts insufficiently addressed. To resolve this limitation, decomposition component analysis is integrated in this study to establish a fault diagnosis model for high-voltage switchgear, which is developed through weighted principal component analysis (WPCA) and a support vector machine (SVM) optimized by an adaptive sparrow search algorithm (ASSA). Training is conducted using latent fault sample data collected from in-service high-voltage switchgear, with model validation being performed through three evaluation dimensions: feature extraction methods, classifier performance, and optimization algorithms. Regarding feature extraction, superior performance is exhibited by the proposed model through effective reduction of feature overlap across different faults and enhancement of recognition accuracy. In classifier evaluation, an average accuracy of 93.70% is achieved, accompanied by excellent outcomes in mean Kappa coefficient, precision, and recall metrics, indicating consistently stable diagnostic performance across all fault categories. For optimization effectiveness, strong global optimization capability and accelerated convergence speed are demonstrated. The validation results confirm that robust diagnostic performance is delivered by the proposed model, enabling precise detection of excessive grease coating, abnormal contact ablation, and floating potential discharge. Significant enhancement is brought to the latent fault diagnosis capability based on decomposition component analysis for high-voltage switchgear through this research.