Detection and Classification of Permanent Fault Using Multi-Layer Perceptron Model in a Distribution Network

The dynamic characteristics of modern power systems are becoming increasingly complex due to various factors, including the integration of renewable energy sources, the growing use of distributed generation, the rise of electric vehicles, and the adoption of advanced control technologies. These changes have led to new challenges in power system analysis and control. Hence fast and accurate estimation of the faulty section is necessary. To intensify the sensitivity and dependability of power system protection, fault identification, diagnosis, identification, and classification are essential. To address these issues, it is discovered that effective and cognitive protection methodologies, relying on sophisticated measuring approaches and intelligent problem diagnostics such as machine learning (ML) and neural network approaches are helpful. In this paper, an IEEE 14 bus system is modeled using MATLAB and simulation. Then a post fault single cycle current and voltage waveform is taken from each bus of this model using a three-phase fault block. Here 12 different cases are generated and maximum coefficients of each phase current as well as voltages are obtained using the wavelet packet transform algorithm. The data preprocessing is performed, and the best data are taken for training and testing purposes. A novel multi-layer perceptron method is implemented for fault detection as well as classification. A comparative analysis is performed with SVM and KNN methods, and the multi-layer perceptron performs better in fault detection and classification approaches.