Abstract: The extreme sea state is generated by the combined action of various external factors. The traditional univariable wave prediction model cannot consider the influence of multiple variables. Therefore,the EMD-LSTM multivariable input model is constructed to predict the extreme sea state. By using the improved EMD algorithm to process the data set measured by wave buoy,the influence of variable end effect and non-stationarity can be eliminated and thus the prediction accuracy of the multivariable input model can be improved. The results show that the multivariable composite model can make an effective early warning on extreme sea conditions,and the model has the best prediction performance after introducing three key factors such as wave height,wind speed and gust speed into the input layer. Indicated by the root mean square error and Nash-Sutcliffe efficiency,the prediction performance of multivariable input model is significantly improved in comparison with the univariable input model.