Abstract: A new traveling wave ranging method of the cross-bonded cable based on unsupervised learning is proposed to improve the fault location accuracy of cross-bonded cables. Firstly,the relationship among the current modes is analyzed in combination with the direct buried laying method of actual engineering environment. The algorithm gives the basis of taking the sum of the three-phase sheath current as the fault location signal. Secondly,the principal component analysis method of unsupervised learning is used to reduce the dimensionality of the high-dimensional matrix composed of the data collected from the direct grounding box and the cross-connecting box,and the density clustering algorithm is used to cluster the samples after dimensionality reduction.Finally,the segment by matching the cluster class with the smallest sample size is selected as fault segment,and double end ranging formula is used to measure the distance. The simulation results of PSCAD show that the proposed method can avoid the influence of cross interconnection points on the electrical wave propagation of the sheath,so that the fault location can still have high accuracy and reliability at different fault distances,and is not affected by the transition resistance.