Identification Of Compound Power Quality Disturbance Based On Optimal Feature Set And Improved Kernel Extreme Learning Machine

With the gradual improvement of the electrification of society,the rapid development of power electronic technology and the increasing maturity of microgrid construction,the issue of power quality has become a hot spot of global concern.The increase in the operation of electrified rails,the access of new power electronic equipment and the integration of distributed energy sources has led to an increase in the frequency and complexity of steady-state and transient disturbances,and the aliasing of power quality disturbance characteristics is serious.The aliasing is serious.Traditional recognition methods for single disturbances cannot meet the classification requirements of composite power quality disturbances.Therefore,high-accuracy and high-efficiency disturbance recognition based on the characteristics of composite power quality disturbances has become an important content for current domestic and foreign scholars to classify power quality.First,briefly describe the social background,academic background and far-reaching significance of the research on composite power quality disturbances,summarize the definition of power quality disturbances and the dis turbance models constructed by them,clarify the causes of the disturbances and their harm,and sort out the composite power The research focus of quality disturbance identification and the problems to be solved.Aiming at the problems of long extraction t ime and low time-frequency resolution of the extracted features in traditional composite power quality disturbance feature extraction,an improved incomplete S transform is proposed to achieve efficient and accurate extraction of composite disturbance feat ures.The window width factor function of adaptively adjusting the time-frequency resolution with the decomposition frequency is proposed to obtain the window width adjustment factor of the eigenfrequency,and the time-frequency decomposition matrix is obt ained by S transform at the eigenfrequency point of the disturbance signal,and the composite disturbance eigenvector is constructed by this matrix Finally,the time-frequency results obtained by the improved incomplete S-transformation of the composite power quality disturbance are analyzed,and the simulation comparison with the traditional feature extraction method shows that the method in this thesis not only reduces the time of feature extraction,but also has a higher time-frequency resolution.Aiming at the invalidity and redundancy of some eigenvectors constructed by composite power quality disturbances,a feature selection method of PCC-Gini is proposed.The importance of all the constructed feature vectors is sorted by Gini importance,and this sort is used as a search strategy to add features that meet the relevance threshold to the optimal feature set one by one,and finally obtain the optimal feature set with high importance and low redundancy.The feature set is visually analyzed and compared with common feature selection methods.It shows that the method proposed in this thesis has low cost of feature selection,can effectively remove a large number of redundant features,and due to the reduced feature dimension,the computational complexity of the classifier is lower.The training time is also reduced,especially for classifiers with higher computational complexity.Aiming at the problem that the recognition ability of composite power quality disturbances and the learning speed cannot be improve d simultaneously,the leave-one-out cross validation method is used to optimize the regularization parameters of the nuclear extreme learning machine,and the singular value decomposition method is used to reduce the computational complexity of the model.Propose a method for identifying comp osite power quality disturbances based on LOO-KELM Simulation and actual measurement results show that this method is effective for identifying 17 types of composite power quality disturbances under various noises.Compared with common classification methods,this method can achieve multiple types with higher accuracy through shorter model training.Recognition of complex disturbances,and the algorithm has good anti-noise performance and strong generalization ability.Finally,based on the feature extraction,feature selection and identification methods of composite power quality disturbances proposed in this thesis,a composite power quality disturbance recognition system based on MATLAB GUI is developed,the system framework and operating environment are briefly described,and the user login module and disturbance recognition module are designed.The feasibility and practicability of the method in this thesis are verified through simulation data,PXI acquisition data and actual power grid data and the functional testing of the above functional modules.