Online Voltage Stability Prediction and Control Using Computational Intelligence Technique

Voltage instability has been reported as the main cause for many blackouts and become a major concern in power systems. This thesis deals with two specific areas of voltage stability in on-line power system security assessment: small-disturbance (long-term) and large-disturbance (short-term) voltage stability assessment. For each category of voltage stability, both voltage stability analysis and controls are studied. The overall objective is to use the learning capabilities of computational intelligence technology to build up the comprehensive on-line power system security assessment and control strategy as well as to enhance the speed and efficiency of the process with minimal human intervention.;The input variables of ANN are obtained in real-time by an on-line measurement system, i.e. Phasor Measurement Units (PMU). This thesis will propose a suboptimal approach for seeking the best locations for PMUs from a voltage stability viewpoint. The ANN-based method is not limited to compute voltage stability indices but can also be extended to determine suitable control actions. In this thesis, it is demonstrated that long-term voltage stability can be improved by re-scheduling real power generation based on the sensitivity of the ANN approach.;Load shedding is one of the most effective approaches against short-term voltage instability under large disturbances. The basic requirement of load shedding for recovering voltage stability is to seek an optimal solution for when, where, and how much load should be shed. Two simulation based approaches are proposed for load shedding to prevent voltage instability or collapse. In the first approach, a particle swarm optimization (PSO) algorithm is implemented which performs an efficient search for a global optimization. In the second approach, a sensitivity based algorithm is conducted through the sensitivity index of the load shedding buses. The proposed approaches are presented using the New England 39-bus test system. The second approach is found to be significantly faster than the first one and results in considerable savings in computer resources for the test system with which the methods were compared.;The voltage stability problems are quantified by voltage stability indices which measure the system for the closeness of current operating point to voltage instability. The indices are different for small-disturbance and large-disturbance voltage stability assessment. Conventional approaches, such as continuation power flow or time-domain simulation, can be used to obtain voltage stability indices. However, these conventional approaches are limited by computation time that is significant for on-line computation. The Artificial Neural Network (ANN) approach is proposed to compute voltage stability indices as an alternative to the conventional approaches. The proposed ANN algorithm is used to estimate voltage stability indices under both normal and contingency operating conditions.