Improved monitoring and security assessment of power systems using machine learning techniques and phasor measurements along-with their optimal placement

Abstract

Phasor Measurement Units (PMUs) are able to provide fast and accurate synchronized measurements than the conventional Supervisory Control And Data Acquisition (SCADA) systems. These synchronized measurements have been recently employed for various power system applications due to their ability of providing updated system information at a particular time. Thus, the synchrophasor technology can play a vital role in enhancing the overall system monitoring, protection and control of power systems. This thesis aims to utilize these synchronized measurements to build some intelligent classifiers for the fast and accurate prediction of power system security. The first part of this thesis is concerned about the optimal placement of PMUs in order to achieve complete observability of the system during normal operation as well as during contingencies. The placement methodology proposed in this thesis is based on a new intelligent search technique, which works in two stages. Stage I uses Best First Search (BFS) algorithm to determine the sub-optimal placement locations of PMUs and stage II uses pruning to remove the redundant PMU locations from the results obtained in stage I. Thus, the proposed method offers complete topological observability with a placement set containing minimum number of PMUs. Further, the proposed method is able to incorporate the presence of single and multiple flow measurements in the system. Cascaded failures, which often result in islanding, are considered as a severe disturbance and therefore, maintaining system observability during such disturbances is of utmost importance. Considering this perspective, the BFS based two stage method has been further extended to find the optimal locations of PMUs for keeping the system observable during cascaded failures. A topology based algorithm has also been proposed to identify whether a particular cascaded event leads to islanding condition or non-islanding condition. In addition to observability, measurement redundancy is also incorporated in the placement scheme to enhance the state estimation process. The synchronized measurements are then utilized for developing an improved monitoring and assessment scheme for power system security using intelligent classifiers. To achieve this, a new framework for Static Security Assessment (SSA) and Transient Security Assessment (TSA) has been presented. The proposed framework for SSA consists of four classifiers, where classifier I is used to predict the static security status of the system as secure or insecure for a particular loading condition and classifier II determines the type of violations (either line overload/bus voltage violation or both) that causes insecurity in the power system. Classifier III is used to predict the security status of that particular loading condition with respect to all probable N-1 contingencies. Similarly, classifier IV predicts the type of violations causing insecurity in those insecure patterns identified by classifier III. The inputsto these classifiers are the synchronized measurements viz. bus voltage phasors, branch real and reactive power flows measured by PMUs. Some of the intelligent classifiers which have been utilized for this assessment include Wavelet Support Vector Machine (WSVM), Case Based Reasoning (CBR) and AdaBoost algorithm. These classifiers are found to have better generalization capability than the traditional classifiers used for security assessment. Simulation results obtained using WSVM, CBR and AdaBoost classifiers are compared with the results obtained using existing techniques. For TSA, the measured rotor angles of the generators are used as inputs to the proposed classifier models. This TSA framework consists of three classifier models in which classifier I determines the transient security status and classifier II is used to determine the generator coherency. Classifier III is a hybrid classifier, which determines the individual generator synchronism state for a given operating condition. This hybrid classifier consists of an array of parallel classifiers, where one classifier is assigned to each generating unit of the power system. Finally, the proposed approach is implemented and tested on standard benchmark systems such as IEEE 14-bus, IEEE 30-bus and on a practical Indian 246-bus networks. Simulation results reveal that the proposed method can enhance the overall monitoring and assessment of power system with the use of synchronized measurements and with classifiers having high generalization capability.