Abstract: The ultrasonic detection accuracy of the transformer bushing is susceptible to noise, and it is difficult to accurately determine the state of the lead inside the bushing. This paper proposes an adaptive gray wolf threshold denoising method. Firstly, the ultrasonic echo signals are decomposed by multi-layer decomposition by wavelet transform, and the hierarchical threshold is estimated by the gradient descent adaptive threshold method based on multi-order continuous derivatives. Then, combined with the gray wolf algorithm, the minimum gradient value is used as the objective function to optimize, determine the threshold size and complete the denoising. Through simulation and case studies, it is found that the signal denoising ratio is higher and the mean signal error is smaller. At the same time, the smallest signal distortion and faster calculation speed can be obtained, and the useful information such as the vibration echo position of the ultrasonic echo signal can be effectively retained. The above-mentioned points can therefore improve the detection results and better reflect the state of the lead inside the casing, which shall have a certain application value.