An Intelligent Technique for Fault Analysis in Smart Grids

Abstract

Power systems are moving towards smart grids through the incorporation of new digital technologies and equipment that increases the system complexity. The power systems become more prone to many types of failures such as cyber-attacks and sensors failure. Therefore, efficient fault analysis is essential to maintain normal grid operations. In this research, an intelligent technique is proposed to detect, classify, and understand the propagation behaviors of the short circuit faults induced in an IEEE 39-bus system. The proposed method is also capable of identifying faults causes and consequences, and suggesting remedial actions. The IEEE 39 bust system is modeled under normal and faulty conditions. A number of datasets are created from the smart grid model, which are then processed separately by Discrete Wavelet Transform (DWT) for fault detection. After that, statistical features are extracted from the coefficients generated by the DWT. The most significant features are identified by the Random Forest algorithm, producing reduced features matrices, which are used to train and test four supervised machine learning techniques, namely: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Bagged and Boosted Trees. The Frequency at the Synchronous generators dataset is found to be the best input to the classifiers based on the highest predictive accuracies attained by the classification techniques. The SVM achieved the highest average predictive accuracy of 98.4% and an F1 score of 0.975; therefore, it is selected as the best technique for fault classification. In addition, Bayesian Belief Networks (BBN) are built for fault propagation. The BBN can identify the fault location and the impacted buses through probability theory. Finally, a dynamic Fault Semantic Network (FSN) is constructed. The FSN utilizes the fault information and knowledge acquired by the Classifier and BBN for causes and consequences analysis and repair actions.

Date of Award	Jul 2020
Original language	American English