The embodiments of the invention disclose a two-dimensional
Hankel matrix multi-scale SVD transformation method. The method includes the following steps that: S1, three-phase traveling wave data are collected, and modulus transformation is performed on the three-phase traveling wave data, and linear traveling
wave mode signals are obtained, and the linear traveling mode signals are analyzed; S2, Xtransformation is performed on the traveling wave linear mode signals, and an approximate component Pj and a detail component Qj under the
decomposition scale of a j-th layer are obtained; S3, a
composite matrix Cj is constructed based on the S2, SVD transformation is performed on the Cj, a singular entropy corresponding to the
decomposition layer is calculated, and a singular entropy increment deltaEj relative to an upper layer is calculated; if it is judged that the deltaEj is larger than a certain value epsilon, the approximate component Pj is assigned to a one-dimensional matrix X, and the step S2 is repeated, an optimal
decomposition layer number r and detailed components Q1, Q2, ..., and Qr which are obtained after the decomposition of the first r
layers are outputted; S4, dynamic threshold
noise reduction
processing is performed on a series of matrices Qi (i=1, 2, ..., and r) which are obtained after the r
layers are decomposed, so that matrices which are represented by a symbol in the descriptions of the invention are obtained; and S5, transformation is performed on traveling wave signals X', and
head wave and reflected wave time points are determined according to a singular point position in a traveling wave
signal approximate component Qr which is obtained according to the decomposition of an r-th layer; and S6, a fault location interval is determined according to the
head wave time of each monitor, and therefore, a fault point can be accurately located.