Effective denoising is an important prerequisite for power quality detection, and wavelet features have a great impact on the denoising effect. To solve this problem, an adjustable threshold function is first proposed. By controlling the adjustable parameters, the proposed threshold function can be changed between hard and soft threshold functions. On the basis of deeply studying the wavelet features, such as orthogonality, support length, regularity, order of vanishing moment, symmetry, etc., the five features that should be satisfied in order to achieve the best denoising effect are proposed in this paper.Db5, haar, bior2.2 and rbio3.1 are used to decompose voltage sag, swell, interruption, oscillation and harmonics signals into four scales. On the basis of calculating the standard deviation of noise, the operator is introduced to modify the threshold, and the new threshold function is used to denoise the wavelet detail coefficients. After denoising, the details of the reconstructed signal are analyzed, and the SNR and MSE are calculated. The experimental results show that db5 wavelet has the best denoising effect under different noise intensity, which proves the correctness and reliability of the five wavelet characteristics in power quality disturbance signal denoising proposed in this paper.