Design of an Intelligent System for Early Fault Detection in Power ‎Transmission Lines Using Deep Learning Techniques

Abstract

Power transmission lines hold the fort of modern electrical infrastructure, however they come under continuous threat of different kinds of faults that can put grid stability at risk, endangering supply continuity. Conventional protection schemes adopt fixed threshold techniques which is one of the major obstacles in identifying the high impedance faults and also demand heavy investment on the special hardware. The objective of this work is to develop and implement a smart system for early faults detection in transmission line in an efficient and cost-effective manner, using lightweight deep learning methods. An experimental procedure was developed based on the simulation of the environment of the power transmission system in order to create a synthetic dataset which consists of multi-operational conditions and multiple fault types. Electrical signals were preprocessed by a sequence of well-defined operations, including normalization and Clarke transformation, to effectively reduce influences of unwanted noise and bring out the most essential temporal and spatial features. To allow automatic abnormal pattern recognition, a one-dimensional convolutional neural network model was trained on the generated data. As the results of the experiment, it was found that the proposed model is obviously superior to traditional protection methods in classification accuracy and response speed, and has better stability under complex operating conditions. It also demonstrated the ability to run on less expensive hardware, thereby reducing the computational burden and allowing the system to run in the field as well. The study concludes that the utilization of artificial intelligence algorithms in conjunction with the existing protection system is able to improve electrical grid reliability in addition to acting as an a cost-effective and scalable solution for future smart grid infrastructure.