Research On Artificial Intelligence-based Fault Location Of VSC-HVDC

With the rapid development of power electronics technology,high voltage direct current(HVDC)transmission has gradually replaced AC transmission in terms of long-distance and large-capacity transmission because of its advantages of lower loss,larger transmission capacity,and lower insulation requirements.The flexible direct current(VSC-HVDC)transmission technology based on Voltage Source Converter(VSC)has the advantages of flexible operation mode and strong controllability,which overcomes the shortcomings of traditional HVDC,such as commutation failure.Therefore,VSC-HVDC plays an increasingly important role in the field of HVDC transmission.However,high-voltage transmission lines have always been one of the components with the highest probability of failure in the power system,so fast and accurate fault location is essential to ensure the safe and stable operation of the power system.This thesis focuses on the fault location in VSC-HVDC,and analyzes the wave process of traveling waves and the mechanism of transient signal generation under single-pole-to-ground faults.With the help of artificial intelligence technology,the following research is carried out on machine learning-based fault location.First,the research background and significance of fault location for VSC-HVDC transmission lines are investigated,and the advantages and disadvantages of existing fault location methods are analyzed.The advantages and feasibility of applying artificial intelligence technology to fault location are discussed,and a machine learning-based fault location method is proposed.Secondly,the propagation of traveling waves on transmission lines is studied,including attenuation and reflection of traveling waves.The development mechanism and waveform characteristics of single-pole-to-ground faults on VSC-HVDC transmission lines are studied.Taking a ±200 k V VSC-HVDC transmission system as the research object,the development mechanism and mathematical model of the single-to-ground fault are analyzed.A point-to-point VSC-HVDC transmission system and transient signal simulation model are built in PSCAD software,combined with the simulation waveforms and transient characteristics of the fault signals.Then,a stacked denoising autoencoder(SDAE)-based fault location method is studied,and simple and fast fault location is realized.This method takes original transient data(decoupled current line mode component)as input,uses a structure of multiple denoising autoencoders to extract fault features.Then,the network is fine-tuned with the extracted features to establish the mapping relationship between transient signals and fault distances,so the fault location is realized.This method takes the actual fault distance(sample label)as the target and jointly adjusts the feature extraction and regression fitting to reduce the influence of human factors on fault location.At the same time,the transient data of multiple transmission lines in different transmission systems are used for training the network,and the number of samples is increased to improve the generalization ability of the model.In the case of a low sampling frequency(10 k Hz)and a short time window(5 ms),it has accurate fault location results.Finally,considering that the artificial intelligence-based fault location method requires a large number of samples,there are only a few fault samples in the actual power system.Because there may be a slight difference between the simulation model and the actual engineering system,it is difficult to ensure that the simulation data is exactly same as the field measured data.On the other hand,when the location model receives a new line with a large difference from the line parameters in the training set,the location error will increase.Aiming at the above problems,a transfer learning-based fault location method is proposed.This method uses transfer learning to complete the training of the network with a small number of samples,and enhances the actual application value of the network.In summary,this thesis focuses on fault location for VSC-HVDC transmission lines,investigates and analyses the existing fault location methods and their shortcomings.On the basis of simulation model,a stacked denoising autoencoder and transfer learning technology are carried out on fault location,and a good location effect is achieved.