Research On Identification Method Of Single Phase Series Arc Fault

With the development of electric power industry, electrical safety issues have received increasing amount of attention. The series arc faults of single-phase distribution lines is one of the main causes of safety hazards, however, it cannot be detected by existing low-voltage apparatus, which only monitoring overcurrent and leakage current. Therefore, a detection method for the series arc faults of distribution lines is imminent. In this thesis, we analyzed currents of electrical loads under various conditions and proposed a novel method for series arc faults detection. Proposed method contains algorithms for feature derivation, quantization, and classification.Test systems of series point contact arc faults and series carbonized arc faults are designed based on UL1699 standard, and it is taken as a platform to collect current waveform of general electrical load under normal operation, point contact arc faults and carbonized arc faults operation. Tested electrical loads include linear and nonlinear loads. Linear loads contain resistive and inductive load, and nonlinear loads contain most common types of electric equipments. Combining the collected current waveform of various types of electrical load, the individual characteristics of the current waveform of various types of electrical load under normal operation and series arc faults are studied. This thesis focuses on the extraction, quantification and classification methods of current waveform features of various types of electrical load under different working conditions. The specific methods are as following:Phase space reconstruction algorithm is applied in order to derive phase space feature of current waveform. The traditional coordinates delay phase space reconstruction algorithm is improved,sampling period is longer in the proposed method. With proposed improved phase space reconstruction method, current waveform can be represented as a shape on phase space coordinate. In order to maximize representable current phase space feature, the C-C method is introduced to perform a cross-examination for two important parameters in the process of reconstruction algorithm. Experiments show that chosen time delay and embedded dimension parameter are optimized.Fractal theory is introduced for quantization of derived phase space feature. The information dimension is adopted given its higher degree of distinguishability. However, based on experiments results, accurate determination of arcs faults would require features more than information dimension. A new parameter, zero current time, is introduced as another feature of current to enhance determination accuracy. Therefore, a two-dimensional feature consists of information dimension and zero current time is proposed.The least squares nonlinear support vector classification system based on support vector machine is proposed, taking the two-dimensional feature quantity as the classification object. Method of exhaustion is used for optimizing penalty factor and radius of the radial basis kernel function in the system. After selecting the optimal parameters, parts of sample data in the database of the two-dimensional feature quantity of current are used to train the classification system model. Following successful model training, a group of data samples is reselected for testing, and the test results show that the accuracy of system identification and location for the series arc faults can theoretically reach more than 90%. Also, several sets of data samples are not involved in the training classification system model, but directly involved in the testing system, and the results show the correct classification recognition rate can reach more than 90%, which indicates the nonlinear support vector classification system established in this thesis have good generalization ability.In the end of this thesis, based on the method of above analysis, the low-performance and low-cost PIC18F4520 microcontroller is used to design the hardware of single-phase series arc faults detection device, and the testing program is programmed by C18. The test results show that this method can effectively detect series arc faults of distribution lines. Therefore, the feasibility of single-phase series arc faults identification methods studied in this thesis is verified by both software simulation and prototype test.