The invention discloses an online abnormal sound monitoring algorithm for mechanical defect diagnosis and positioning of a GIS body. The algorithm is formed by a front-end abnormal sound online diagnosis algorithm and a master station server platform fault diagnosis algorithm. According to the front-end abnormal sound online diagnosis algorithm, sound wave data of GIS equipment in operation are received; denoising and feature extraction are carried out on the data by using a fast kernel independent component analysis and a bispectrum analysis technology; whether a signal is fault signal is determined preliminarily based on a feature vector; and if so, lossless compression, data storage, and return operations are performed. According to the master station server platform fault diagnosis algorithm, a suspected fault signal is received; decompression is carried out and an intrinsic mode function energy entropy is calculated; the energy entropy is inputted into a trained fault diagnosis algorithm based on an online sequence extreme learning machine; and then secondary diagnosis is carried out on the suspected fault signal. According to the algorithm disclosed by the invention, the algorithm parameters are updated by using a GIS holographic sound database perfected continuously, so that the accuracy and practicability of the fault diagnosis algorithm are improved.