Abstract The complexity of power quality disturbances (PQDs) is a significant risk factor in the electricity sector. An accurate and fast analysis of these disturbances provides crucial information to cover all the issues related to power quality. The main objective of this study is to explore a new analytic technique, including all kinds of disturbances that can appear in electrical networks, that differs from previous technologies such as the Fourier transform. Three methods based on the Stockwell transform, namely, the discrete orthonormal Stockwell transform (DOST), discrete cosine Stockwell transform (DCST), and discrete cosine transform (DCT), were used to analyze PQDs in time–frequency representation. These methods diagnose the disturbance's signal properties, which are dependent on resolution and absolute phase information. Nine PQDs, including normal sine waves, were mathematically modeled and used to evaluate the proposed methods. All the methods can effectively simulate and analyze PQDs. Among them, DOST is the most effective in providing clear and high-resolution time–frequency representations of signals. The classification of disturbances was fulfilled based on statistical features extracted from matrices derived from Stockwell transform-based methods, such as analytic approaches (mean, variation, standard deviation, entropy, skewness, and kurtosis). Neural networks, a method utilizing intelligence classifiers, were used for pattern recognition, and the patterns of the different methods were compared. Simulation results proved that DOST needs fewer samples than other methods; its capability to deal with signals in time–frequency resolution is also more viable. The neural network classifier has a higher accuracy rate than the K-nearest neighbor and decision tree methods and approximates the support vector machine method.
Read full abstract