The invention discloses a method of optimizing a power quality disturbance classifier of an ELM. Power quality disturbance S transformation signal analysis is carried out, a power quality disturbance signal model is established, and feature extraction is carried out on the basis thereof according to extraction principles of features; a learning and training link is carried out; a GA is utilized to optimize design parameters of the ELM, and a large number of statistical tests are carried out to select superior experiment schemes; and finally parameter settings are determined for power quality disturbance identification classification verification, and a classifier design scheme applied in practice is provided according to obtained experiment results. Therefore, advantages that the ELM classifier has higher classification accuracy in all of a variety of noisy environments, and can meet classification requirements in practice are realized.