The invention relates to an acoustic velocity prediction method based on automatic coding and an extreme learning united network. The method is characterized by preparing well logging curve data of a reference standard well and a target well in a to-be-measured region, carrying out environmental correction and standardization processing on all curve data, constructing an automatic encoder network model and carrying out iterative calculation to obtain optimal associated curve data, building an extreme learning network mode, carrying out iterative calculation to obtain a weight matrix parameter, determining an optimal parameter of a prediction model and obtaining a prediction model, and applying the prediction model to a well with deficiency of an AC curve and predicting an AC curve. With the acoustic velocity prediction method based on the reasonable design, the AC prediction precision and stability are improved substantially; and reliable acoustic data can be provided for follow-up reservoir inversion and integrated geological interpretation and thus the method has the great application value.