Intelligent fault location of optical fiber link in substation based on improved YOLOv8
TL;DR
Comprehensive experiments show that the improved YOLOv8 model outperforms the baseline detection network in terms of recall, precision, and F1-score, especially in the case of overlapping or low-amplitude faults.
Abstract
This paper presents a technically robust framework for intelligent fault location in substation optical fiber links using an improved YOLOv8 architecture. The proposed method incorporates a multi-scale attention mechanism and advanced feature fusion to enhance the model’s ability to identify and localize fault events under complex and noisy conditions. Wavelet-based denoising and enhancement techniques and systematic data preprocessing on a large real substation dataset containing multiple fault scenarios A composite loss function is used to optimize the detection pipeline. The function balances the focal loss, IoU regression, and attention regularization to ensure accurate training convergence and superior representation learning. Comprehensive experiments show that the improved YOLOv8 model outperforms the baseline detection network in terms of recall, precision, and F1-score, especially in the case of overlapping or low-amplitude faults. Quantitative analysis of ablation studies confirms that both the attention and fusion modules are critical for suppressing false positives and improving localization accuracy. In addition, confusion matrix visualization and case analysis reveal that the model achieves stable generalization across multiple substations and fault types. The enhanced deep learning pipeline has proven its advantages in monitoring substation faults, providing reliable and efficient solutions for utility infrastructure and supporting the progress of intelligent power system management.