Power Quality Analytical Methodology Using Signal Processing and Machine Learning Techniques for Grid Fault Classification and Location Detection

The increase in global energy demand with a growing dynamic power system has made today's electrical network a complex infrastructure where increased faults and grid violations are being reported at higher rates. Also, with an increased penetration of intermittent Distributed Energy Resource (DER), the detection of transient events occurring in a large network is becoming challenging and therefore encouraging potential areas of research in this field. Furthermore, the deployment of high-resolution devices such as PMU gives access to vast amounts of data and carries vital information to ensure grid safety and protection. This paper analyzes a methodology to detect and predict power system disturbances and fault location in electrical power network (distribution & transmission) using data-driven & machine learning techniques. It uses signal processing techniques based on multi-level discrete wavelet transforms, with wavelet spectral energy/entropy-based feature extraction stage to train ANN (Artificial Neural Network) classifiers to predict and classify faults in power system applications. Machine learning classification techniques are used based on a dataset which is generated with single measurement point in simulation environment where grid model is used with varying fault conditions such as alternating phase & neutral impedance, inception angle & distance. A comparative analysis is performed with ANN and other potential classifiers to identify potential limitation & improvements in feature extraction stage to better classify faults and location. Furthermore, PQ events such as voltage sag, voltage swell, and transient event have been analyzed and the propagating effect of faults in the network, medium voltage to low voltage network are also discussed.