In view of the existing adaptive variable step size least mean square (LMS) harmonic current detection algorithm in low signal-to-noise environment effected easily, an improved adaptive variable step size LMS algorithm was introduced. The difference of the error signal and the signal from the previous cycle is used as the feedback,which combined with Versiera function constructed servo dynamic factors. The dynamic factor as the momentum factor, which adjusting weight. The coherence estimation error was used to control step length. The algorithm compromised convergence rate and stability ,in addition,it reduced effectively the noise interference. Through detection and analysis of motor soft-starter producing periodically repeatedly harmonic in the process of working,which proved the new detection algorithm is feasible.