Analysis Method Of Power Quality Disturbances Based On Wavelet Transform And Support Vector Machine

In recent years, power quality problem have drawn extensive attention of people. One reason is that power quality has been being disturbed heavily with the increasing number of polluting loads such as non-linear loads, time-variant loads, fluctuating loads, unbalanced loads, etc.; the other is that intelligent electrical devices have put forward more rigorous requirements for power quality. As an important index of the power industry product, power quality concerns the interests of generating power, supplying power and utilizing power. In the event that the power quality index departs the normal level too much, it will endanger the electric power generation, electric power transmission, electric power transform, electric power distribution and electric power consumption to a different extent. Therefore, power quality has become one of the primary problems in the electric power market. Power quality monitroing and analysis technology and power quality control technology have become new researching focuses of the electric power system field. Power Quality monitoring and analysis is the basis and premise of power quality control. Therefore, it is very necessary to build a power quality monitoring and analysis system to examine, evaluate and classify the power quality disturbances correctly. This dissertation makes a deep research on methods with regard to power quality analysis which include detection and location method of dynamic power quality disturbances based on complex wavelet transform, automatic recognition and classification method of dynamic power quality disturbances based on multi-class support vector machine classifier; data compression method of power quality disturbances based on lifting wavelet transform. Application of these methods may improve the power quality monitoring and analysis system in aspects of accuracy, real-time performance, automation, etc.In order to resolve the disadvantage of the common-used orthogonal real wavelet transform that it cannot draw the signal's phase information, based on analyzing the impact of wavelets'phase-frequency characteristic on wavelet analyzing results, it is brought forward that analyzing power quality disturbances by use of compound information of the complex wavelet transform may acquire better analysis results. Since the existing complex wavelets are all continuous wavelets, in order to overcome the disadvantages of complex calculation, this dissertation deeply researches a method for constituting orthogonal compact support complex wavelets by the existing orthogonal compact support real wavelets through changing the phase-frequency characteristic. This method is based on multi-resolution analysis and as long as the orthogonal compact support real wavelet has low-pass filter, it will meet the requirements of this method. Since the orthogonal compact support complex wavelet constituted by means of this method is orthogonal and compact supported, it can adopt Mallat fast wavelet algorithm. In order to acquire much better analysis results, Db4 orthogonal compact support real wavelet is chosen here to constitute a kind of db4 orthogonal compact support complex wavelet,which is used to detect and locate dynamic power quality disturbances. The Db4 orthogonal compact support complex wavelet filter group constituted by means of this method has the same length and the same support set as the original Db4 orthogonal compact support real wavelet. It also has the same characters as the Db4 orthogonal compact support real wavelet , respectively orthogonal, compact support, vanishing moment and regularity. Through simulation, the analysis results of Db4 orthogonal compact support complex wavelet are compared with that of Db4 orthogonal compact support real wavelet, and the analysis results of compound information of orthogonal compact support complex wavelet are compared with that of simple information.Whereas wavelet transform has the excellent characteristic of time-frequency localization, and support vector machine (SVM) has the excellent ability of statistic study, this dissertation brings forward a new method based on complex wavelet transform and multi-class SVM classifier for recognizing and classifying dynamic power quality disturbances. The complex wavelet transform is used to to extract the feature vector of dynamic power quality disturbances, and the multi-class SVM classifier is used to recognize and classify dynamic power quality disturbances according to the feature vector extracted. Based on the theory of structural risk minimization, SVM has stronger generalization ability, and can resolves several deficiencies of artificial neural network. In order to overcome the deficiencies of current multi-class SVM classifier algorithms, this dissertation brings forward a new multi-class SVM classifier algorithm based on fuzzy clustering analysis thought, i.e. hierarchical clustering SVM algorithm,and simultaneously constructs a classification tree for dynamic power quality disturbances based on hierarchical clustering SVM classifier algorithm. The hierarchical clustering SVM algorithm has the merits such as higher learning speed, lower error classification rate, less sub-classifier number, and requiring less memory, etc. In order to certify the correctness and validity of this method, through simulation this dissertation compares the recognition and classification method of dynamic power quality disturbances based on complex wavelet transform and multi-class SVM classifier with that based on artificial neural network, and also compares the hierarchical clustering SVM algorithm with several common-used multi-class SVM classifier algorithm.In order to reduce the storage space and the transmission time of power quality disturbance data, and also in order to enhance the real-time performance of the power quality monitoring and analysis system, this dissertation brings forward a new data compression method of power quality disturbance based on lifting wavelet transform which is as follows: firstly confirm the decomposition scale and make multi-scale decomposition to power quality disturbance signal by lifting wavelet transform; secondly, according to the relevant threshold strategy, make the threshold quantization and encoding operation to the higher frequency coefficients of each scale, and simultaneously store the coefficients with connection to the disturbance and throw away the coefficients without connection to the disturbance, and in this way compression of power quality disturbance data is accomplished; finally according to the compressed data reconstruct the original power quality disturbance signal by use of inverse lifting wavelet transform. In order to evaluate the correctness and validity of the compression method based on lifting wavelet transform, four evaluation indexes of data compression performance, respectively compression ratio, mean square deviation percentage, signal-to-noise ratio and energy ratio, are introduced,. In addition, the lifting algorithm of Db4 orthogonal compact support real wavelet is chosen to make simulation experiments to the compression results of power quality disturbances. This dissertation designs a power quality monitoring and analysis system. The main station is composed of industrial computer and PCI CAN card, and the slave station is just the power quality monitoring device. In addition, the software implementation procedure of power quality analysis method based on complex wavelet transform and SVM in the system is proposed. This dissertation brings forward a hardware scheme for the power quality monitoring device which is based on the DSP+DARAM+ARM structure. DSP is mainly used to gather power signals, calculate basic power parameters, calculate and analyze the power quality parameters, and implement the functions of event recording, malfunction alarming and remote signalling/remote control; ARM is mainly used to implement local display, CAN communication, setting up system parameters and keyboard operation, etc.; DSP and ARM change the data via DARAM. In order to resolve the problem of address bus and data bus multiplexing between ARM and CAN controller, a sequence circuit between ARM and CAN controller, which is implemented via CPLD, is designed. In order to increase the real-time performance, stability and reliability of device, the embedded real-time operation systemμC/OS-II is adopted in ARM software development.