| In recent years,with the high development of the national economy,people will demand more and more electrical energy.The construction of first-class distribution networks is also progressing gradually.The distribution system network structure is becoming more and more complicated.The system running state is diverse.With more branches and a mix of overhead and cable lines,various faults inevitably occur.Quick and accurate identification of fault types and location of fault points will facilitate rapid fault removal,shorten outage times and reduce the labor intensity of line patrol persons.However,the fault mechanism of the power system is more complex.Fault signals show strong randomness and non-linearity,and the scale of data is enormous.There is an accumulation of errors in the fault analysis process.The existing fault diagnosis methods are challenging to reach the expected effect.This paper introduces a new generation of artificial intelligence technology,chaos theory,and modern signal processing techniques into distribution network fault diagnosis.The article can achieve fault type identification,fault line selection,fault section location,and fault distance measurement in distribution networks.This paper can improve the effect of distribution network fault diagnosis.(1)This paper investigates the identification of fault types in distribution networks under both adequate and inadequate samples.1)Combining the parameter optimized variational mode decomposition(VMD)and convolutional neural network(CNN),a fault identification method for the distribution network was proposed under adequate samples.The VMD was adopted to decompose the main transformer’s signals,including the three-phase current of the low voltage side,the three-phase voltage,and the zero-sequence voltage of the bus bar.According to the central frequency from low to high,the obtained intrinsic mode functions should be arranged in sequence.The VMD can obtain the time-frequency matrices.In order to decide the decomposition layers of the VMD adaptively,the partial mean of multi-scale permutation entropy is used to optimize it.The time-frequency matrixes can be transformed into pixel matrices by pseudo-color coding.The pixel matrices which were appropriately compressed can be used as input of CNN.The CNN can autonomously extract the eigenvector of fault.Support vector machine,naive Bayesian classifier,extreme learning machine,and random forest are used to train and test the feature vectors extracted by CNN,and so on,to verify the effectiveness of the method.2)Combining the modified complete ensemble empirical mode decomposition with adaptive noise(MCEEMDAN)and conditional generative adversarial network(CGAN),a fault identification method for the distribution network was proposed under the inadequate samples.The MCEEMDAN may decompose the electric signal into intrinsic mode functions.The raw time-domain signal can be transformed into the two-dimensional gray-level image by the pseudo-color coding of the intrinsic mode functions.The fault gray image can be labeled and put into CGAN to generate new samples to achieve data augmentation.In order to improve the quality of the generated samples,the least square loss function is introduced into the original CGAN network to make the generated samples close to the raw samples.The convolutional neural network(CNN)is used to mine the fault features autonomously.The Softmax classifier is used to achieve distribution network fault classification.The experiments show that the proposed method can effectively learn the distribution characteristics of the original sample.Furthermore,the fault recognition accuracy can be effectively improved.The proposed method has good stability,fast convergence speed,and high precision,and it can effectively complete the fault identification of the distribution network.(2)Aiming at complex characteristics of the un-effectively grounded system that the zero-sequence current contains intense noise and non-stationary feature,the approach based on modified complementary ensemble empirical mode decomposition(MCEEMD),optimal denoising smooth mathematical model,and chaos oscillator was given.The complementary ensemble empirical mode decomposition(CEEMD)was modified by combining the general regression neural network,and fuzzy C means clustering and generalized composite multiscale permutation entropy.The MCEEMD may decompose the original signal into a series of smooth intrinsic mode functions to complete the first filtering.The decomposition stage used the endpoint mirror method to suppress the end effect.The MCEEMD efficiency suppresses the mode confusing phenomenon of empirical mode decomposition(EMD).The MCEEMD has better completeness and orthogonality than ensemble empirical mode decomposition(EEMD)and complementary ensemble empirical mode decomposition(CEEMD).Establishing optimal denoising smooth mathematical model chose optimal intrinsic mode functions to complete the second filtering and to ensure that the reconstructed filtered signal has better smoothness and similarity.The MCEEMD optimal denoising can weaken noise and enhance faulty characteristics.The bifurcation characteristic of the chaotic oscillator was applied.The optimal denoising consequence of each line zero-sequence current was extracted as the external periodic dynamic.Trisection symmetry phase estimation was applied.Image texture parameters are used to autonomously discriminate the phase transition trajectory of chaotic oscillators to complete fault line selection.The research results verified the usability and effectiveness of the proposed method.(3)The fault-section location of the distribution network is researched.A method based on the wavelet packet transform AlexNet-GRU model is proposed to achieve fault segment location in distribution networks.The wavelet packet transform(WPT)is adopted to decompose electric signals.The wavelet packet coefficients of each node are rearranged from low frequency to high frequency to obtain the time-frequency matrix.The time-frequency matrix can be converted into the pixel matrix with the property of the image by color-coding.The pixel matrix can contain the working conditions of the current system.The AlexNet-gated recurrent unit(GRU)is built.The fine-tune AlexNet network can autonomously extract the pixel matrix features as predictive variables.The extracted predictor variables are used as input of GRU,and the fault area location for the distribution network is completed.Experimental analysis for the overhead/cable hybrid line with multi-branches is carried out.The testing results show that the proposed method is not affected by fault time,fault type,grounding resistance,and other factors.It can meet the distribution network’s fault location accuracy and reliability requirement.(4)A novel fault distance measurement method based on the spatial domain image fusion and convolutional neural network(CNN)is proposed.The three-phase traveling wave can be decoupled by the phase-mode transformation matrix for obtaining the line-mode component of the traveling wave.The 1D line-mode traveling wave can be converted into a 2D image by the Gramian angular field(GAF).The 1D line-mode component can be mapped into the color,point,line,and other characteristic parameters of the 2D image.In order to expand the invisible information of the line-mode traveling wave,the images obtained by the Gramian angular summation field(GASF)and Gramian angular difference field(GADF)are weighted and fused.The CNN can be used to autonomously mine the characteristic parameters of the weight-fusion image and realize fault distance measurement.The simulation results show that the proposed method does not need to be considered in the traveling wave head and the traveling wave speed.The fault distance measurement method is not affected by fault time,fault distance,or transition resistance factors.It has better robustness and higher fault distance measurement precision. |