The invention discloses an in-situ power system voltage sag detection method. According to the technical scheme, the method comprises the steps that (1) a nonlinear system is expressed with a state equation, and a Cholesky decomposition method is adopted to calculate a covariance matrix and a volume fraction; (2) a volume point of nonlinear function propagation is calculated, a predicted measuredvalue in a k(th) step is calculated, and an innovation covariance matrix and a cross covariance matrix are calculated; (3) a state covariance matrix in the k(th) step is updated and calculated; and (4) STCKF is adopted to precisely track the state so as to estimate a signal amplitude, frequency and an initial phase. Based on volume Kalman filtering, the change process of voltage sag can be trackedin real time, and the amplitude, phase, frequency change and lasting time of the voltage sag can be accurately calculated.