Islanding Detection Using Computational Intelligence Techniques in a Smart Distribution Network

Abstract

Distributed generation (DG) offers solution to the ever increasing energy needs by generating the energy at the consumer end, in most cases by means of renewable energy sources. A microgrid with DGs will result in an enhanced performance in terms of continuity of the power supply for consumers. Microgrids may operate either in grid{connected or islanded mode. Islanding detection is one of the most important aspect of interconnecting a DG to the utility. Several islanding detection methods have been proposed over the years to improve the islanding detection in terms of detection time, accuracy. However, with the upcoming trends, such as smart grids, there is an imminent need for incorporating intelligence to the islanding detection methods. Also, it is important for the islanding detection methods to perform well at near zero power mismatch conditions and noisy conditions. This research work proposes islanding detection methods based on image classification techniques. Time{series data from point of common coupling is acquired and then converted to an image to enable this. A dataset for islanding detection based on several islanding and non{islanding events is created to be used in training and testing the machine learning and deep learning models. Three islanding detection methods are proposed in this research work. The first method is based on HOG feature extraction from the image and SVM classifier. The second method is based on transfer learning method. The third islanding detection method is based on custom designed CNN for islanding detection. In addition to islanding detection, a feature for early islanding detection is also proposed in this research work. Early islanding detection is proposed by monitoring the fault and normal conditions. Once a fault occurs, the time window between the operation of relay contacts and the opening of circuit breakers is utilized to detect the islanding event. All the methods are tested with the islanding dataset that is created which includes near zero power mismatch conditions and noisy data. The proposed methods demonstrate the potential of image classification techniques for islanding detection.