Abstract: Traditional fault diagnosis and localization methods typically rely on linear models for power system fault detection, resulting in low positioning accuracy under complex fault scenarios. To address this, we propose a convolutional neural network-based fault localization algorithm for power systems. The method first employs the 3σ principle and median filtering to eliminate outliers and suppress noise from multi-source data, then combines normalization to achieve feature fusion and generate high-dimensional feature vectors. Next, we construct a convolutional neural network-based anomaly signal recognition model: Using 1D-CNN to extract local temporal features, 2D-CNN to capture spatiotemporal features, and implementing binary classification tasks for fault signal identification. Finally, Morlet wavelet transform is applied for time-frequency analysis of abnormal signals, combined with phase difference method for precise fault zone localization. Experimental results demonstrate that our proposed method achieves 100% diagnostic accuracy with a fault localization error of 5 mm, demonstrating high positioning precision.