Methods for online voltage stability monitoring.

摘要

Online voltage stability monitoring is the process of obtaining voltage stability information in real time for a given operating condition. The prediction should be fast and accurate so that control signals can be sent to appropriate locations quickly and effectively. One approach is to get the stability information directly from the phasor measurements obtained for operating conditions. This approach is simple and requires few computations. The methods proposed are based on Thnin equivalent of a system. The Thnin equivalent, according to the maximum power transfer theorem, gives the upper limit of the power transfer to a load bus. To get the Thnin equivalent we need at least two sets of phasor measurements. It is found that Thnin equivalent gives a highly optimistic approximation of power margin. The work done in this thesis compensates the optimistic prediction by applying reactive power availability information of the system. The accuracy of this approach is very high compared to Thevenin equivalent. The thesis also presents improvement on decision trees method for online voltage stability monitoring by attribute selection. The role of data mining approach such as decision tree is vital in using the available accurate measurement data in the power system. Also, it is very important to extract important data or attributes so that the tree is robust, reliable and easy to compute. Data mining itself offers information based (gain ratio), statistical (k-nearest neighbor), probabilistic (na Bayes) and others for attribute selection. There are analytical approaches in power systems which can characterize attributes as well. Can these attributes be used for attribute selection for decision trees. The hypothesis has been tested using the tangent vector information of attributes. The accuracy of the selected attributes on decision trees is very high. Attributes with higher sensitivity were found to be better indicators of voltage instability. Attribute selection will be very helpful when it comes to large systems with a huge volume of data.