Study On Fault Diagnosis Of Power System Based On Wide Area Information

With the rapid development of smart grid in China, the operational safety and stability of power system are becoming more important. Meanwhile, the traditional backup protection based on local information has become less suitable to the operation requirements of large power system.The wide area backup protection is a beneficial supplement for traditional local main protection, in which multi-point measurement information is used for fault diagnosis of power system, so that the reliability of protection can be improved. The realization of wide area relay protection is supported technically by the rapid development of wide area measurement technologies, high speed network communication technologies and high precise time synchronization technologies, which promote vigorously the relay protection researches based on wide area information. On the basis of analyzing the system structure and related technologies of wide area relaying protection, fault diagnosis algorithms based on wide area information are mainly studied in this dissertation,concentrating on three aspects: wide area power frequency quantity, wide area traveling wave information and wide area state information. At the same time, in order to make full use of artificial intelligence technologies and wide area state quantities to realize fault diagnosis, contrastive researches are carried out on supervised and unsupervised pattern recognition methods respectively.Main works are as follows:(1) According to different fault types of power system, two fault diagnosis schemes based on wide area power frequency quantities are proposed, one is based on the comparison of wide area fault charges. The protective associated area of power system node is defined in the method, and relationships of positive sequence fault charges for associated area boundaries are sought. Then conceptions of the calculational charge and the reference charge for associated area are given. The state matrix of nodes and related branches is established based on relative size relationship of those two kinds of charges which are sent to the wide area decision center station by node Intelligent Electronic Devices (IEDs). The actual failure element can be determined by searching for matrix elements. The other scheme is based on current polarities and energy relative entropy. After fault components of fundamental positive sequence currents are obtained for different associated areas which are divided by the algorithm, concepts of boundary current, comprehensive calculation current and virtual current for associated areas are given. Energy relative entropy of the comprehensive current sampling values which is used to quantify the difference of currents is defined. At the same time, the angle between boundary currents of associated areas is calculated to reflect the polarity relationship of boundary currents. The fault diagnosis can be realized according to the significant difference between the energy relative entropy and the angle when internal fault and external fault occur in associated areas.(2) Two fault diagnosis schemes based on wide area traveling wave information are proposed.In the first method, wide area initial traveling wave voltage energies at nodes are calculated by using wavelet transform. The node with the biggest energy could be judged as the failure node.Then wavelet coefficients sequence angles of initial voltages and currents of all associated branches at the node are calculated. The size of angles determines whether a branch is the fault line. In the second method, a novel fault diagnosis scheme based on the distribution characteristics of wide area initial traveling wave reactive power is proposed. Initial traveling wave reactive power is defined and the reactive power amplitudes of associated branches at Substation in the protection area are compared, the line which has the maximum power amplitude is selected as a quasi-suspected fault line and its power amplitude is uploaded to the decision-making center station of the wide area traveling wave protection system to establish a power matrix for substations and associated branches. Further searching for suspected fault lines based on the matrix, and then the fault line can be determined by two-terminal power ratio of the suspected fault line.(3) With artificial intelligence technologies, the application of three supervised pattern recognition technology is studied for fault diagnosis of power systems based on the distribution characteristics of wide area status information. In the first, a new method of fault diagnosis making use of good classification and fault-tolerance characteristics of Probabilistic Neural Network(PNN) is proposed. The status information of fault direction component, zone II of distance protection and the main protection is taken as sample set for training and testing of the PNN network. In the second method, a new fault diagnosis algorithm of power system based on Least Squares Support Vector Machine (LSSVM) is proposed. Various components state information collected by node IEDs of power system in associated areas are fused after logic operations.Feature vectors who are formed by the fused information can be input into LSSVM to identify node IEDs of fault associated areas. After that, the fault components of power system would be identified. In the third method, a fault diagnosis network of lines and buses based on PSO-LIBSVM is established. The information collected by IEDs is fused by calculating the action coefficients of different associated areas. And then the sample matrixes can be established. Particle swarm optimization algorithm (PSO) is used to search optimal parameter combination to train the network that are tested by testing sample matrixes of random failure to identify the fault associated IEDs.After that the fault components can be determined. The simulation results of various fault diagnosis with uncertain information show that three methods proposed can identify faults of power system quickly, accurately and reliably with strong adaptability and good fault tolerance.(4) Combining fuzzy C-mean clustering analysis theory (FCM) with wide-area state information, a new diagnosis method of power system which is an unsupervised pattern recognition technique is proposed. Corresponding protection action information, direction component state information, the circuit breaker status information are acquired by line IEDs.And state information from line IEDs are made as objects of FCM clustering. The definition of power system associated IED is given, and then fault component IEDs can be classified into one category according to fault discrimination algorithm. IEDs with same direction of external faults can be classified into another category. With the fault area minimum principle, the category which has the minimum number IEDs in the clustering results is chosen as the fault element association IEDs. Components associated by the correlative IEDs is determined as fault elements. A large number of simulations show that the algorithm has better fault-tolerant performance, faster operation speed, and higher accuracy rate of discrimination compared to supervised pattern recognition methods. The method can identify the fault area with many inaccurate state information correctly.Finally, fault diagnosis methods based on wide area infonnation are summarized and follow-up works are prospected.