Power quality monitoring in emerging power systems using adaptive and intelligent techniques

Abstract

In the recent years, the Power Quality (PQ) has emerged as a major area of electric power engineering due to the rapid increase in the application of electronic equipments and their sensitivity to the electromagnetic phenomena. The high proliferation of nonlinear loads, power electronic-based devices in the power system and renewable energy sources are deteriorating the quality of power supply to quite severe levels. PQ monitoring is the process of gathering, analyzing and interpreting raw measurement data into useful information. The PQ monitoring helps in the understanding of PQ disturbances, their causes and their impact on the power system and the end-user equipment. Therefore, the continuous monitoring of load current and the supply voltage is a prerequisite for an appropriate remedial action to be taken for improving the PQ. The development of intelligent techniques to automatically analyze and process the large data not only helps in diagnosing the PQ phenomenon quickly but also eliminates the human interference and need for experts. In general, the currents are polluted mostly with harmonics and interharmonics while the voltages deviate from their nominal values. In the practical power system, the voltage and currents vary dynamically and are highly unpredictable. Therefore, the time-varying nature of PQ waveforms requires an adaptive tool that can analyze the disturbances accurately and visually identify the instants of transitions. Existing parametric and non-parametric methods are proved to be best in the analysis of either of the time-varying distortions, but only a few techniques are suitable for analysis of harmonics, interharmonics, and other PQ disturbances. Further, these techniques need improvements concerning the accuracy, computational complexity and robustness to the fundamental frequency deviation. The primary objective of this thesis is to analyze all sorts of time-varying disturbances including interharmonics in real-time while maintaining maximum possible accuracy. Another objective includes the development of a novel online automatic recognition approach for accurate classification of single and combined disturbances under noisy conditions. To fulfill these objectives, adaptive and intelligent techniques have been developed in this thesis based on the recent advances in signal processing and classification techniques. The Empirical Wavelet Transform (EWT) and Tunable-Q Wavelet Transform (TQWT) are two such techniques with adaptiveness in filter design and computationally efficient. This thesis ingeniously utilizes the potentials of these two adaptive signal processing techniques and customizes them for analyzing all sorts of PQ disturbances. The work carried out in the thesis is segregated into six parts. The first three works develop adaptive techniques based on the EWT for precise estimation of signal parameters and PQ indices of distorted time-varying signals. The main emphasis of these works is to extract the actual fundamental frequency component, which can reflect the voltage disturbances correctly. The results of synthetic and practical signals are presented to justify the feasibility and advantages of these techniques. Further, two new classification models are proposed based on the Support Vector Machines (SVM) for classification of single and combined disturbances utilizing only six basic features. In the first classification model, a multiclass SVM is built employing only nine binary SVMs, each dedicated for detection of a single disturbance. The second classification method categorizes the PQ disturbances into two based on the signal information and employs a dual multiclass SVM. It is a known fact that a linear classifier with basic features is adequate to identify the single PQ disturbances if the extracted fundamental frequency component is accurate. Therefore, the TQWT is utilized to tune the wavelet filters for extraction of the actual fundamental frequency component. Extensive testing on simulated and real disturbance signals reveals that both the proposed classifiers have an overall classification accuracy of around 97% with the mean classification time of at most 80 ms. Finally, the hardware implementation of the EWT-based PQ indices is performed on the TMS320F28377S DSP. Investigation on real signals confirms the practicality of the approach for online estimation of PQ indices with the maximum possible accuracy.