The Traditional LMS(least mean square) adaptive filtering algorithm is always in a contradiction state because it has a fixed step size and resolves the relationship between steady-state error and convergence. Based on the analysis of the filtering algorithm, according to the step-size adjustment principle of the variable-step LMS adaptive filter algorithm, a segmented variable step size LMS adaptive filtering algorithm based on the normal distribution curve was proposed by constructing the nonlinear function of the step-size factor and the error signal, and the influence of the parameter value on the performance of the algorithm was analyzed. Aiming at the problem of difficult selection of reference signal in actual signal processing, a method of reference signal selection based on the splitting array was proposed. The theoretical and sea trial data analysis results show that the convergence speed and steady-state error of the proposed algorithm are obviously better than the fixed-step LMS adaptive filtering algorithm and the variable step size LMS adaptive filtering algorithm based on the Sigmoid function.