Decision-Level Fusion of DNN and SVM for Transmission Line Fault Diagnosis

Abstract

Accurate fault diagnosis is important for transmission-line operation. This paper proposes a hybrid framework in which a deep neural network learns nonlinear feature representations and a support vector machine performs final classification. Wavelet transform, Z-score normalization, and sample augmentation are applied before a three-layer fully connected DNN extracts 16-dimensional representations; an SVM then constructs a maximum-margin boundary in the learned space. Controlled experiments are conducted on two synthetically generated structured-feature datasets. The model achieves 98.53% accuracy on a six-class dataset and 86.10% accuracy on a five-class dataset, outperforming the standalone DNN and CNN-LSTM baselines included in this study. These results support the feasibility of DNN-SVM fusion under the tested distributions. Because field-recorded data, preprocessing ablation, and broader hybrid baselines are not included, the findings should not be interpreted as evidence of field robustness or universal superiority.