Author: YUTONG LI (Harry)
Supervisor: Dr. Chiagoziem Chima Ukwuoma
GitHub: JSLEE-0703/Final-Project

This project explores the application of deep learning technology in anomaly detection of transmission lines, aiming to improve the reliability and efficiency of power systems. Transmission lines are an important part of modern power grids, which are responsible for transferring electrical energy from the source to the distribution network. However, due to environmental and other factors, they are prone to failure, which leads to significant economic losses and security risks. Traditional fault detection and classification methods, such as physical inspection and expert judgment, are time-consuming and error-prone. In this context, the emergence of machine learning and deep learning technologies offers a promising alternative to real-time monitoring and fault detection. The study used data sets named typical power systems. This dataset is used to train and evaluate custom ANN models and multiple other machine-learning models. The performance of the custom ANN model was compared to other existing machine learning algorithms such as random forest, deep support vector machine (SVM), K-nearest neighbour (KNN), decision tree (D-tree), and logistic regression. In addition, the effectiveness of the ensemble learning method is studied to further improve the accuracy and reliability of fault detection. The main research results show that the artificial neural network model proposed in this paper, especially the stacked model, has high accuracy, precision, recall rate and F1 score in fault detection and classification. The stacked model integrates the proposed artificial neural network model and the random forest model. Its accuracy rate reaches 0.9986. Meanwhile, the F1 score, recall rate and precision are all 0.9986, with the highest performance indicators among all models. Compared with the single model, this ensemble model shows better performance, highlighting the advantages of ensemble learning in this scenario.

Keywords: Deep Learning, Anomaly Detection, Transmission Lines, Artificial Neural Networks, Ensemble Learning

• Fault Detection: Binary classification (Normal vs. Fault)
• Fault Classification: Multi-class classification (5 fault types)
• SMOTE: Handles class imbalance
• LIME Explainer: Model interpretability
• Stacking: Combines ANN + Random Forest for best performance

Source: KingArthur000/Electrical-Fault-detection-and-classification (I optimized the data that was prone to cause confusion. Makes this dataset more suitable for deep learning models)

All visualizations: visualizations

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

git clone https://github.com/JSLEE-0703/Final-Project.git
cd Final-Project/Electrical-Fault-detection-and-classification-main-Refresh