This paper presents an effective approach to identify power quality events based on IEEE Std 1159-2009 caused by intermittent power sources like those of renewable energy. An efficient characterization of these disturbances is granted by the use of two useful wavelet based indices. For this purpose, a wavelet-based Global Disturbance Ratio index (GDR), defined through its instantaneous precursor (Instantaneous Transient Disturbance index ITD(t)), is used in power distribution networks (PDN) under steady-state and/or transient conditions. An intelligent disturbance classification is done using a Support Vector Machine (SVM) with a minimum input vector based on the GDR index. The effectiveness of the proposed technique is validated using a real-time experimental system with single events and multi-events signals.

The economic impact associated with power quality (PQ) problems in electrical systems is increasing, so PQ improvement research becomes a key task. In this paper, a Stockwell transform (ST)-based hybrid machine learning approach was used for the recognition and classification of power quality disturbances (PQDs). The ST of the PQDs was used to extract significant waveform features which constitute the input vectors for different machine learning approaches, including the K-nearest neighbors’ algorithm (K-NN), decision tree (DT), and support vector machine (SVM) used for classifying the PQDs. The procedure was optimized by using the genetic algorithm (GA) and the competitive swarm optimization algorithm (CSO). To test the proposed methodology, synthetic PQD waveforms were generated. Typical single disturbances for the voltage signal, as well as complex disturbances resulting from possible combinations of them, were considered. Furthermore, different levels of white Gaussian noise were added to the PQD waveforms while maintaining the desired accuracy level of the proposed classification methods. Finally, all the hybrid classification proposals were evaluated and the best one was compared with some others present in the literature. The proposed ST-based CSO-SVM method provides good results in terms of classification accuracy and noise immunity.

No abstract