The invention provides a wave motion waveform analysis and prediction method and device based on an extreme learning machine under Dropout constraint. On the basis of deep learning, a method for analyzing wave motion by adding a dropout algorithm constraint is adopted, a traditional wave motion analysis method is improved, and the characteristic that an extreme learning machine improves the information processing capability under the condition of the dropout constraint and is resistant to fitting is utilized to analyze the wave motion. The method is faster and more accurate in the aspect of wave motion analysis, and the analysis capability on wave motion is improved. Firstly, a deep learning program is trained; the generalization ability and the operation speed of the training model are improved through constraint of a dropout algorithm, it is guaranteed that overfitting of wave analysis is prevented when data is limited through the method that part of nodes are hidden randomly, the calculation time is shortened, and the real-time performance and the accuracy of an analysis result are guaranteed.