Abstract: The ambient data under the natural excitation of the system contains rich electromechanical characteristic information.Accurate identification of small signal stability characteristic parameters from ambient data is of great practical significance for guiding the safe and stable operation of the power system. An on-line identification algorithm of small signal stability characteristic parameters driven by ambient data based on subspace optimal mode decomposition（Sub-OpMD） is proposed. The algorithm obtains the singular sub-matrix by the matrix linear transformation of the input data based on the orthogonal projection, and the conjugate gradient algorithm is used to iteratively solve the optimal low-dimensional orthogonal space to achieve the optimal lowdimensional approximation of the high-dimensional mapping matrix between singular sub-matrices. According to the eigenvalue decomposition result of the optimal low-dimensional matrix, the small signal stability characteristic parameters of the system can be accurately obtained, i.e., the oscillation frequency, damping ratio and mode. The matrix linear transformation based on orthogonal projection and the introduction of conjugate gradient method make the dynamic mode decomposition（DMD） method better adapt to the ambient data. The calculation and analysis of IEEE 16-generator 68-bus system and actual system measured data verify the effectiveness and accuracy of the proposed algorithm.