A new fundamental component extraction algorithm is presented. The sampling signals are decomposed to smooth signals and detail signals by improved multi-scale mathematic morphology filter. The smooth signals are adopted to update the observations of the Kalman filter to reduce the interference of fault signals’ transient noise, which can improve the convergence speed of the filter algorithm. Detail signals are adopted to calculate the measurement noise variance in real-time to improve the convergence precision of the filter algorithm. The Matlab/Simulink simulation system is built and the results show theSalgorithm feasibilitySandSeffectiveness.