| High Voltage Direct Current(HVDC)transmission technology can improve transmission efficiency and resource utilization.It is the main way to afford the transmission need of long-distance,large-capacity,clean renewable energy.The flexible direct current(VSC-HVDC)transmission technology based on Voltage Scourced Converts(VSC)has a lot of advantages,such as flexible operation and strong controllability.It plays an increasingly important role in the field of HVDC transmission.The relay protection is the key technology to ensure the power system operation safety and stability.And the transient signal recognition is an important part to ensure the protection device operation correctly.Studying accurate and effective transient signal identification methods is of great significance for improving system power supply reliability and operation and maintenance quality.This paper focuses on the transient identification of flexible DC transmission lines,and analyzes the mechanism of common transient signals.From different perspectives of artifical intelligence technology,the transient signal recognition method based on machine learning is studied.The main work of this dissertation is as follows:The background and significance of transient identification of flexible DC transmission lines are introduced.The research status and characteristics of different methods are analyzed,such as traditional methods,the method of combination of artificial feature extraction and traditional intelligent algorithm classifier,and the method of combination of feature self-extraction based on unsupervised learning and classification models.The development mechanism and wave characteristics of three types of transient signals,such as unipolar ground fault,lightning fault and lightning interference,are analyzed.Based on ±500kV VSC-HVDC transmission system,the occurrence mechanism and the mathematic models of different transient signals are detailed.Two-terminal VSC-HVDC transmission system and transient signal simulation model are built in PSCAD.Combined with the simulation waveform,the transient characteristics of three kinds of signals are analyzed.Based on Frequency Spectrum Entropy(FSE)and Support Vector Machines(SVM),a transient signal recognition method is proposed.In this method,the newly defined FSE is used to extract single-ended transient current characteristics.And the multi-classifier of one-to-one SVM is used for classification.The calculation method of FSE essentially utilizes the relative magnitude relationship of each frequency content.It has the ability to withstand the influence of distance and transition resistance parameters.The use of SVM helps to make full use of features and soften classification boundaries.At the same time,the method can simplify the criterion setting and improve the recognition accuracy.Based on belief network,a transient signal recognition method is proposed.It realize the reliable identification of end-to-end transient signals.This method takes the raw transient data(current line mode component)as input,and uses the Restricted Boltzmann Machine(RBM)and Softmax layer to perform feature extraction and classification.It reduces the manual processing segments,and achieves joint adjustment of feature extraction and classification.In the simulation,a number of system data are used to increase the number of samples.Under the actual requirements of low sampling frequency(10kHz)and short time window(3ms),this method has high recognition accuracy and good generalization performance.In summary,this dissertation deeply investigates and analyses the present research state about the transient signal identification of flexible DC transmission lines.Based on the modeling and comparison of transient signal characteristics,this dissertation proposes new recognition methods from different perspectives,and achieved good simulation recognition effect. |
PDF Full Text Request