Research On The Application Of Sparse Optimization In Power Quality Disturbance Analysis

In the complex power grid environment with multi physical field strength coupling,the power quality(PQ)of the system shows extremely complicated non-stationary propagation phenomenon,and the mutual coupling temporal-spatial characteristics between various disturbances lead to more complicated power quality disturbance(PQD)behavior.Presently,the traditional methods for PQD analysis are mostly from the spectrum transformation analysis theory under complete orthonormal bases framework,which attempts to represent arbitrary signals with a set of fixed basis functions that have the similar properties,but the own characteristics of the target signal itself are ignored,and the essential characteristics of the signal are difficult to describe succinctly and adaptively.To this end,taking the two critical problems of PQD signals denoising and classification as the objects of study,this dissertation plan to discuss a new method for PQD analysis based on the framework of sparse optimization modeling theory using overcomplete dictionary.And the related theoretical analysis and experimental research are finished,which provides helpful supplement for the theory and application of power quality analysis and control.Firstly,considering the shortcomings of traditional signal analysis methods,such as fixed base function and poor adaptability,resulting in the bottleneck in the application of PQD analysis in the orthogonal base framework.Therefore,a novel sight to address the two difficult problems of bad denoising results and low accuracy of classification accuracy of PQD signals in the spirit of sparse optimization modeling using overcomplete dictionary is proposed,which provides a simple and effective theoretical basis for the analysis of PQDs.Secondly,a time-frequency atoms dynamically searching method for PQD signals de-noising based on improved chaotic particle swarm optimization(ICPSO)is proposed,which effectively solves the difficult problem of the mutation feature point retention in noise suppression.The atoms and an overcomplete dictionary with similar characteristics to PQD signals itself are constructed to decompose the signal into a very concise sparse representation.In addition,in order to reduce the computational complexity of the sparse recovery process,the local optimization OMP algorithm is combined with the ICPSO global optimization(ICPSO-OMP)is developed,which significantly accelerates the recovery efficiency of the method.Then,taking six typical disturbing signals as the research object,the effectiveness and superiority of the proposed method is verified through the related experimental results.Finally,a novel sparse representation method for PQDs classification based on stack sparse autoencoder(SSAE)is proposed,which effectively solves the problem of low recognition accuracy results.By stacking 2 SAE networks and constructing a SSAE deep network model,which is used to automatically extract the depth abstract features of PQD signals,it solves the difficult problem of artificial extraction feature.The expression features extracted by SSAE are sent to the MSRC classifier for classification(SSAE-MSRC),which makes up for the existing classifiers that need to consume a large amount of computing resources for model training.Numerical experimental results demonstrate that the proposed SSAE-MSRC method can effectively improve the accuracy of the PQD classification results and achieve much beter recognition performances than other SRC extended algorithms as well as some traditional classifiers,such as SVM and ANN,in terms of robustness and classification accuracy.