Research On Fault Location Of ±800kV UHVDC Transmission Line

Due to the distance between our energy resources and load centers are far apart,the UHVDC transmission project must be constructed in order to realize long-distance and large capacity transmission of power.Considering that the long transmission lines of UHVDC system and it passes through complex terrain,whether the fault ground fault in the transmission lines can be detected and eliminated timely and accurately,it will directly affect the reliability of system operation.Therefore,it is of great significance to study the fault location of UHVDC transmission lines.In this paper,the relationship between fault distance and energy ratio is analyzed,and the energy ratio can be gotten by using wavelet packet decomposition algorithm,the fault location algorithm based on wavelet packet decomposition and neural network is proposed.Compared with the wavelet analysis can only decompose the low-frequency part of the signal,the wavelet packet decomposition algorithm can not only decompose the low frequency part,but also decompose the high frequency part,so as to get more comprehensive fault information.Meanwhile,in order to obtain the mathematical relationship between the fault distance and the energy ratio,radial basis function(RBF)is used to train it.The RBF neural network has high precision approximation capability for any function,it can avoid the local optimum.However,whether its weight and training centers are chosen properly,it will directly affect the accuracy of the final training results.So the particle swarm optimization(PSO)algorithm is used to optimize the neural network,and then fault location is realized.The PSCAD/EMTDC software is used to build the model of ± 800kV UHVDC transmission system from Yunnan to Guangdong.Assuming that the ground fault occurs in the positive transmission line,the voltage signals of fault transmission lines are collected,the phase mode decoupling is performed to extract the outgoing mode components for different transition resistances and different fault locations.Then the wavelet packet decomposition is used to process the line mode component,obtaining the energy ratio of each frequency band,the energy ratio is used as the input of PSO-RBF neural network for training and learning.Finally,the fault location is realized.The simulation results show that the algorithm of wavelet packet decomposition combined with neural network has high precision,the error is within 1 km and the relative error is not more than 0.1%,which proves the validity of the method.