Study On Fault Line Selection And Disturbances Classification Of Distribution Grid Using QNN-DS

Smart distribution grid (SDG) is a research hotspot in power system currently. Advanced distribution operation automation (ADOA) and advanced metering infrastructure (AMI) are major and prior technologies in the construction process of SDG. Fault line automatic selection of distribution grid in single phase fault is an important and difficult technology of ADOA, and it is still a problem which isn't resolved properly; the automatic classification technology of power quality (PQ) disturbances is a significant issue of PQ monitoring system in the structure of AMI. Simultaneously, the accuracy of fault line selection and disturbances classification has direct impacts on the reliability, continuity and quality of power supply in distribution grid. To achieve a higher accuracy of fault line selection and disturbances classification, increase the reliability of power supply and improve the PQ of distribution grid, the technologies of pattern recognition and information fusion are used for the research of fault line selection and power quality disturbances classification in this paper.The basic principles and mathematical description of quantum neural network (QNN) and Dempster-shafer evidence theory (DS) are introduced in the thesis, where QNN is a pattern recognition technology, and DS is a theory of information fusion. The architecture and implementation procedure of a novel pattern recognition method based on QNN and DS (QNN-DS) is presented, and the excellent pattern recognition property after the two combined is analyzed.To detect the fault line of distribution grid in single phase fault, the pattern recognition method based on QNN-DS is proposed. According to fast fourier transform (FFT) and wavelet packet transform (WPT) algorithms, three kinds of fault features including the fundamental component, the fifth harmonic component, and the transient component, extracted from zero sequence current are used to train and test three different QNNs, and then DS is used for global fusion at the decision level to gain a unified fault line selection result from the outputs of the networks. The simulation results indicate that QNN has a better convergence and pattern recognition performance than improved back propagation neural network (BPNN), the fault line selection method based on QNN-DS can effectively fuse the fault feature criterions and obtain higher accuracy and better properties of fault-tolerance and robustness, and the process is not sensitive to earth mode, fault distance, inception angles and transition resistance.The characteristic and key issues of PQ disturbances classification are analyzed, and recognition of PQ events using QNN-DS pattern recognition method is proposed. Ten kinds of typical disturbances models are created and 1000 disturbances data are randomly generated based on signal parameters which are amplitude, disturbance duration time, and signal to noise ratio. Three kinds of feature vectors extracted by discrete wavelet transform (DWT), WPT and S-transform are utilized to train and test three different QNNs, the simulation results indicate that each vector can correctly classify the PQ disturbances and has its own advantages; compared with improved BPNN, it is also verified that QNN has a better convergence and classification performance than BPNN. Via using the DS evidence theory for fusing outputs of the networks at the decision level can overcome the influence of the fault information to the certain extent and obtain higher accuracy. The recognition capability of the QNN-DS classifier is compared with BPNN-DS, probabilistic neural network with voting rules (PNN-VR) at the decision level, and only one QNN with information fusion at the feature level, it is found that the proposed classifier gives the best classification result. Therefore, the PQ disturbances classification method based on QNN-DS can effectively fuse the advantages of various feature vectors and obtain higher disturbances recognition accuracy and better properties of fault-tolerance and robustness, and it can be used as a PQ events classifier in practice.The simulation results of fault line selection and PQ disturbances classification based on QNN-DS indicate that the proposed method has inherited good classification property of QNN and excellent information fusion property of DS, and shown high pattern recognition accuracy, good properties of fault-tolerance and robustness, and strong adaptability applied in fault line selection of neutral non-effective grounding system in single phase fault and PQ disturbances classification.