Power quality event classification using optimized Bayesian convolutional neural networks

Management of the electrical grid has an importance on the sustainability and reliability of the electrical energy supply. In the process, it is still crucial that power quality (PQ) is evaluated as part of any grid management master plan. This article provides a novel approach for classifying PQ disturbances such as voltage sag, swell, interruption and harmonics. In the proposed method, colorized continuous wavelet transform coefficients of the voltage signals are applied to convolutional neural networks as an image file. Thus, there is no need for extra feature selection and data size reduction steps as in conventional machine learning-based classifiers. Experiments were conducted on a dataset containing 1500 real-life disturbance signals measured from different locations in Turkey by Turkish Electricity Transmission Corporation. With the power of deep learning in image processing, the proposed method provides very high classification accuracy with a value of 99.8%. Comparisons with the other PQ disturbance classification methods, which are using traditional signal processing-based feature extraction and machine learning algorithm, prove that the proposed method has a simple methodology and overcomes the defects of these methods.