The disturbance signal recognition of power quality is the premise of distribution parameter analysis

localization of disturbance source and comprehensive control of power quality. As for such issue as insufficient performance ability of S transform in the analysis of power quality disturbance signal and low recognition precision due to random set input weight of extreme learning machine and threshold of hidden layer

a kind of new recognition method of power quality disturbance signal based on S transform（generalized S-transform

GST）and particle swarm optimization（PSO）extreme learning machine（ELM）.Firstly

such three adjustment factors as rough

fine and accurate adjustment are introduced to the Gauss window function of S transform

to adjust each factor value as per the frequency feature of the disturbance signal

so to obtain specific time-frequency resolution and increase feature. Secondly

the corresponding values of input weights and hidden layer threshold at maximum fitness are obtained by the optimizing capacity of PSO and improve the recognition accuracy of ELM. Finally

PSO-ELM is trained and its recognition ability is tested in accordance with the feature set of GST time-frequency module matrix. It is shown by the comparison experiment that the recognition method proposed in this paper

compared to the S transform and ELM method

has high recognition accuracy and strong noise-resistance performance and can meet the need for the recognition of power quality disturbance signal at the industrial environment.