Abstract: In order to predict the significant wave height accurately and quickly,our methodology utilizes an autoregressive modelsupport vector regression algorithm（AR-SVR） based on the average mutual information for feature selection.The AR-SVR model combines the statistical characteristics of the significant wave height series and considers the influence of driving wind field.This paper compares the prediction performance of AR-SVR model with AR and SVR model,and the performance study results demonstrate that AR-SVR performs better than the AR and SVR with higher prediction accuracy.