optimizing fault identification in power distribution systems by the combination of svm and deep learning models

Maintaining electrical grid stability and reliability requires the rapid diagnosis and classification of faults in power distribution systems. this study presents a hybrid model that integrates deep learning with support vector machine (svm) methodologies to classify distribution system faults. in the proposed approach, feature extraction is performed using a convolutional neural network (cnn), and an svm classifier is employed to identify fault patterns and establish generic fault classifications. the hybrid model is trained and evaluated using an extensive dataset comprising power distribution system fault currents under various fault types and conditions. the integration of deep learning feature extraction with svm classification enhances fault classification effectiveness. this study aims to contribute to the overall improvement of distribution system reliability, reduction of downtime, and more efficient grid management. to achieve this, pscad software is utilized to simulate faults and collect images of three-phase fault current data. initially, the fault classification problem is addressed using four pre-trained cnn models, with the collected images serving as input data. the hybrid model consists of two distinct components: an svm block, known for its efficient and precise data classification capabilities, and a cnn block, specifically designed for feature extraction. in the matlab environment, a combination of four pre-trained cnn models—alexnet, squeezenet, googlenet, and resnet-18—are utilized in conjunction with an svm to create hybrid models. the hybrid squeezenet-svm model has demonstrated exceptional performance, achieving an accuracy rate of 99.95%, a precision rate of 99.98%, a sensitivity rate of 99.6%, and a specificity rate of 99.7%.