The invention discloses a power grid fault classification method based on feature mapping migration learning. The method comprises the steps that 1 target domain data and auxiliary source domain data are selected; 2 fault feature extraction is carried out on the target domain data and the auxiliary source domain data based on micro-increment wavelet singular entropies; each micro-increment wavelet singular entropy is used as a fault feature; an eigenvector space corresponding to the target domain and an eigenvector space corresponding to the auxiliary source domain are formed respectively; 3 based on a feature mapping migrate learning method, basis vectors corresponding to an axis feature, the unique feature of the auxiliary source domain and the unique feature of the target domain respectively are found; and 4 the acquired basis vector corresponding to auxiliary source domain is used as a support vector; a similarity penalty term is set, and the constraints of a support vector training set is added to jointly train a classifier to acquire a corresponding classification result. According to the invention, three groups of basis vectors which can best embody the fault category can be accurately and quickly found.