A modified pattern recognition algorithm based on the residual analysis and the recursive partition is proposed in this paper, which is more effective than the original algorithm in cases of high-dimension finite-sample practical engineering problems. The proposed approach is first verified with artificial testing data and then applied to power system transient stability assessment (TSA). Case studies show that the proposed pattern recognition scheme can successfully find out the relationships between the pre-contingency steady state quantities and the stability indices under each specified fault. According to the spatial description of the obtained patterns in the feature space, the transient stability levels can be predicted according to the system operation states. Thus the TSA scheme obtained can not only predict the stability index, such as the CCT value, of faults but also offer instability preventive control strategies. Simulation results on the IEEE New-England test system verifies the feasibility and good performance of the proposed algorithm. As a pattern recognition technique, it can be widely applied in different engineering domains to realize knowledge acquisition.
Tong-wen Wang, L. Guan, Yao Zhang
2008 IEEE Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century