Studies On Kernel Feature Recognition For Intelligent Stability Assessment

Stability assessment and security control is a key issue in power system operation and schedule. At present, security assessment is mainly carried out by digital-simulation-based fault scanning. For large power grid, one bottleneck is the huge calculation burden caused by fault enumeration. Besides, digital simulation cannot point out the related factors to cause instability as well as the possible operation means to improve the security, thus cannot meet the requirements of intelligent power grid operation and decision supporting. Stability and security assessment based on artificial intelligent provides a new way for online stability assessment. This method is treated as a very promising intelligent decision analysis tool for power system substitution of artificial experience considering its merits, strong learning ability, and fast evaluation and provides clear and easy-understood rules for stability assessment.However, how to select the input features for stability classifier is a key factor, which would affect the precision and speed of stability evaluation and the generalization of rules for stability discriminator, in the stability assessment based on artificial intelligence method. This thesis focuses on key features recognition for stability assessment based on artificial intelligent and the main research include:An embedded feature selection algorithm based on improved ant colony optimization algorithm and k-nearest neighbor (k-NN) classifier is proposed after comprehensive analysis of the advantages and disadvantages of various feature selection algorithm. A local search loop is designed to remove the redundant or strong-correlated features, and through artificial data to verify the validity of the algorithm.Dynamic features are chosen to form candidate feature set and the key dynamic features strong-correlated to stability are selected. The optimal feature subset suited for different grid size is extracted according to analysising the multi-set of feature subset supplied by the feature selection algorithm. Rule trees for stability assessment are obtained by using the optimal feature subset as the input of stability assessment classifier and the similarity and specificity in different size of grid are analyzed in the thesis.Steady-state power flow features are chosen to form candidate feature set and the key steady-state feature subsets that are strong-correlated to stability are selected. The impact of disretization on the results of feature selection is analyzed and scatter coefficient is defined to avoid the impact of the discrete boundary on feature analysis. In order to prevent the case of leakage alarm in the power system stability assessment, a fuzzy stability region is defined during the process of forming the rules for stability discriminator.This work was supported by National Natural Science Foundation of China under Grant 50407014.