Online and Real-time Power System Stability Assessment using Data-Driven Analytical Techniques

Power system stability is concerned with the system’s dynamic behaviour following a physical disturbance. The loss of stability can lead to catastrophic consequences such as cascading failure and even wide-spread blackout. Therefore, maintaining power system stability has long been an essential requirement for secure and continuous electricity supply to the customers. With the smart grid development, the intermittent generation from renewable energy sources (RES) and the active demand-side participation brings more uncertainties to the system operation. In such highly volatile operating environment, online and real-time power system stability assessment is of great significance to avoid blackout events. The conventional stability assessment methods are based on time-domain simulations, which are not fast enough to satisfy the online and real-time application need. Wide area measurement systems are envisaged as the grid sensing and communication infrastructure in a smart grid for enhanced situational awareness. Based on the massive amount of data from measurement devices such as phasor measurement units (PMUs), data-driven methods have been identified as powerful tools for online and real-time stability assessment given the high complexity of the system and difficulties in modelling the physics behind the complex system dynamics. This PhD research develops a series of data-driven methodologies for accurate, fast, and robust stability assessment, covering both steady-state dynamic security assessment (DSA) and post-disturbance real-time stability assessment (RSA) areas. The main contributions through this PhD research are as follows. First, the data quality in data-driven analysis is improved. A new database based on uniform sampling is developed for online DSA. It requires less number of instances to achieve the same level of assessment accuracy. Besides, a generation rescheduling database is developed for rule-based preventive control. Second, in the area of DSA, an optimized intelligent early-warning system is developed for reliable online detection of risky operating conditions. It consists of an ensemble learning model based on extreme learning machine and a decision-making process under a multi-objective programming (MOP) framework, which optimally balances the tradeoff between warning accuracy and warning earliness. Third, in the area of RSA, a hierarchical and self-adaptive architecture is proposed for assessing short-term voltage stability (STVS). Under the proposed architecture, the assessment on voltage instability and fault-induced delayed voltage recovery (FIDVR) phenomena are coordinated hierarchically, and the assessment result is delivered in a self-adaptive way to achieve the fastest assessment without impairing its accuracy. Moreover, a probabilistic self-adaptive method is proposed to achieve more reliable and more robust real-time STVS assessment. At last, imperfection in PMU data is studied and the robustness of data-driven stability assessment against missing PMU data and PMU measurement noise is improved. A robust classification model and an ensemble data-analytics model are developed as robust DSA methods against missing-data issue, and a noise tolerant ensemble model is developed to improve the RSA assessment robustness against PMU measurement noise. All the methodologies have been tested on benchmark power systems to verify their effectiveness, and comparative studies with existing methods in the literature are conducted where applicable