The invention relates to a method for extracting a feature vector of cable partial discharge signal based on wavelet transform modulus maxima, which comprises the steps of S1, acquiring a cable partial discharge signal detected in the field; S2, performing wavelet decomposition on the cable partial discharge signal to obtain a scale coefficient on the highest decomposition scale and a wavelet coefficient on each decomposition scale; S3, respectively performing single branch reconstruction on the obtained scale coefficient on the highest decomposition scale and the wavelet coefficients on the decomposition scales to obtain reconstructed signal sequences; S4, respectively acquiring a modulus maxima sequence for each of the obtained signal sequences, and building a modulus maxima matrix by using the modulus maxima sequences; S5, performing singular value decomposition on the obtained modulus maxima matrix to obtain a singular value sequence of the modulus maxima matrix; and S6, forming afeature vector of the cable partial discharge signal by using the obtained singular value sequence. According to the method, the feature vector extracted on the basis of advantages of the wavelet transform modulus maxima can better represent the partial discharge signal.