Intelligent Mining For Power System Security And Stability

With the development of power grid interconnection and power market, the running environment of power system becomes more complex. On the one hand, we should ensure the security and stability of power system; on the other, the available information that the decision maker want to use is limited. With the increasing development of power system simulation and measure techniques, large amount of data is created. It is very important to extract knowledge from the abundant data set, which can be used to meet the demand of power system operation and academic research. This thesis proposes a method to cope with the widening gap between large amount of data and information. The data mining technique is applied to analyze the power system security and stability. New contributions of this dissertation are as follows:1) An intelligent data mining model of power system security and stability is proposed firstly. A practical method is presented to analyze the power system dynamic data from both the digital simulation and the real measurements. Meanwhile, the dissertation presents a novel multiple knowledge models of mined information. The research result demonstrates that the proposed intelligent mining model can accomplish the analysis task of the power system security and stability with a great potential for further application in power grid.2) Using the current simulation software of power system, this dissertation discusses the simulation data acquisition method of intelligent mining for power system security and stability. Based on this, the mining procedure for various knowledge on power system security and stability can be processed automatically considering different simulation models. More specific influences of system model on power system security and stability can be available.3) This dissertation proposes the approach for the feature extraction, feature selection, sampling data of power system security and stability. The original features that are fit for intelligent mining are selected. The data quality is defined based on its validity as well as its usability. A new method to improve data quality is proposed.4) The data platform of intelligent mining for power system security and stability is built using the data warehouse technique. A multi-dimension data model of security and stability is proposed, which is much convenient and flexible for data exploration and data source visualization.5) This dissertation presents new predictive models for the transient stability and small signal stability based on support vector machine theory that can solve the problems such as finite samples. A new method of feature selection and sample condensation is proposed to build predictive model which improves the practicability of the model greatly.6) This dissertation presents a new rule model for power system security and stability. It includes the model of decision rule and the model of association rule. Based on the presented models as well as available knowledge of power system security and stability, this dissertation constructs rule models for transient stability decision, transient stability criterion decision based on phasor measurement units and small disturbance stability decision. During the modeling, feature selection, feature generalization, process of decision rules and new splitting method of decision tree are proposed.7) Several examples are employed to demonstrate the validity of dissertation approach. In order to enhance the intelligibility of the mining process and result, this dissertation uses the visualization technique to deal with the predictive model and the model of decision rule.It has been shown by this thesis that no single algorithm can meet all the requirement of security and stability analysis for all kinds of power grids. So we should construct an analysis platform with all kinds of data mining algorithms to obtain all the security and stability knowledge of target power grid automatically. In order to limit the analysis region and enhance the effi...