Research On Fault Diagnosis Method Of Multi Terminal HVDC Transmission Line Based On Parallel Convolution Neural Network

With the developing of economic and the shortage of global energy,the eastern coastal areas of China need a lot of power,but a large number of solar,wind power and other new energy power generation are distributed in the western areas of China,which are far away from large-scale urban agglomerations.So the large capacity long-distance transmission project came into being.Compared with AC transmission,HVDC transmission as a long-distance transmission mode has the advantages of less fault,low insulation requirements,no synchronization,etc.,which has been paid more and more attention.Multi terminal HVDC(MT-HVDC),which is composed of Modular Multilevel Converters HVDC (MMC-HVDC),has many characteristics different from AC system,such as fast rising speed of fault current,sensitive to over-current of power electronic devices in converter station,difficult to cut off of DC fault current and so on.In order to ensure the safety of the system,it is generally required to remove the fault in a very short time.When removing faults in MT-HVDC system,the fault location is required to be accurate,which affects other transmission lines as little as possible,so it is particularly important to determine the accuracy of fault location on the line where the monitoring point is located.To sum up,the research of fast and accurate MT-HVDC line fault diagnosis is very important.The research contents of the fault diagnosis method based on parallel convolution neural network are as follows:(1) The simulation model of MT-HVDC system based on MMC is constructed.Built the simulation model of multi terminal HVDC system by PSCAD simulation software.Studied and set the configuration parameters of MT-HVDC system.Study on the control method of MMC converter station.Study on system level control of MT-HVDC system.Parameter setting of line module,fault module and sampling module.The simulation model of multi terminal DC transmission system runs stably and can simulate the faults of DC line and AC system in the system.(2) The fault amplitude and frequency characteristics of multi terminal DC transmission lines are studied.When fault occurs in MT-HVDC system,the amplitude change of electrical signal waveform can distinguish each fault clearly.However,the change of fault distance in actual system can not be reflected in the amplitude obviously,which makes it impossible to distinguish the fault type and location reliably only depending on the amplitude change.The wavelet packet frequency analysis method is used to preprocess the data and input the amplitude characteristics and frequency characteristics to the neural network to fully capture the fault characteristics.(3) The fault diagnosis method of MT-HVDC transmission line based on P-CNN is studied.Through collecting the electrical signal waveforms on the DC line when each fault type occurs in the multi terminal DC transmission system,preprocessing the electrical signal waveform characteristics of each fault type,and then pre-training the two single branch convolution neural networks to identify the fault type and identify the fault area inside and outside;according to the principle of migration learning,the middle layer of each convolution neural network is The parameters are assigned to the two branches of the P-CNN respectively,so that the input and output layers of the P-CNN are trained under the condition that the parameters of the middle branches of the P-CNN remain unchanged,and the P-CNN pre-training for the fault diagnosis method of multi terminal DC transmission lines is completed;the parallel convolution neural network after the pre-training is completed is applied In the multi terminal DC transmission line fault diagnosis,when the multi terminal DC transmission line fault occurs,collect the electrical signal waveform on the DC line,preprocess the electrical signal waveform characteristics,input to the P-CNN to output the multi terminal DC transmission line fault diagnosis results.