MULTIPLE POWER QUALITY EVENTS ANALYSIS

Abstract

PQ events have become a serious concern for the power system operations. Literature in this thesis revolves around different signal processing and classification techniques for recognition of PQ events. Volterra series based techniques are the most suitable choice for non-linear and non-stationary signal leaving behind all other techniques based on basis functions like Fourier transform, STFT, S-transform and Wavelet transform etc. Therefore, this work revolves around Volterra analysis of PQ events in the power system. Then features are extracted which act as input to the interval type-2 fuzzy based classifier and finally the results are compared with other methods which show high accuracy of proposed novel technique for detection and classification of power quality events. A novel technique of adaptive finite element method based on sparse features based approach explored for analysis of power quality events. The role of adaptive finite element method is to generate a unique feature set and type-2 fuzzy system for the purpose of classification of power quality disturbances with minimum error. The concept of stiffness matrix is applied on power quality events. The type-2 fuzzy system utilizes the concept of membership functions to classify the single and multiple power quality events and then the proposed method is compared with other methods and finally with traditional type-I fuzzy logic approach for classification of power quality events. The results revealed that the novel methodology of adaptive FEM can reduce the computation time significantly with high accuracy. In today’s time, maintaining the power quality (PQ) in the electrical system is a major issue between the end user and the utilities due to increase in demand of sensitive microprocessor based controllers, heavy non-linear loads and solid state equipment attached to the grid. To deal with this issue an intelligent system is designed based on Long Short-Term Memory (LSTM)-Convolution Neural Network Based Hybrid Deep Learning approach for Power Quality events monitoring. LSTM is a part of deep learning, proficiency of training data and computational power makes deep learning efficient on complex pattern recognition and power-quality disturbances analysis. Therefore, one section in this thesis is devoted to detection and classification of PQ issues by using LSTM. ix This work has tailored an initial layer of CNN; the next stage is max pooling which performs the task of finding low level dependencies in layers. To develop a concise feature map, the features extracted in the first stage are applied as input to subsequent layers of CNN and max pooling layers. The event information is extracted through a two stage feature extraction process to bring out a high dimensional feature set to achieve correct classification of PQ events with less complexity and minimal time using convolution neural networks. Thus, this work addresses the detection and classification issues of single and multiple PQ events in power systems. The three main issues addressed in this thesis, first detection and classification of multiple power quality events under noisy and noiseless conditions and another is to develop adaptive intelligent systems which will change the mesh size of finite element methods so that memory size for monitoring can be optimized. Then finally the work explored the possibility to join two methods and to develop new techniques which have good features of both the techniques.