Ship motion multi-step real-time prediction mixing method and system based on reinforcement learning

The invention provides a reinforcement learning-based ship motion multi-step real-time prediction mixing method and system, and the method comprises the steps: obtaining original ship motion data through an attitude sensor disposed on a ship, dividing the original ship motion data into a training data set and a prediction data set, carrying out the RTWPD, and obtaining a prediction data set; decomposing a high-frequency component and a low-frequency component of the ship motion data into subsequences with a fixed number of layers; an ORELM basic prediction model is established for each sub-sequence obtained through decomposition, an AdaBoost.MRT reinforcement learning mode is introduced, iterative training is conducted continuously, and a plurality of trained ORELM models are combined together; and finally, reconstructing a prediction result of the sub-sequence to obtain a model multi-step initial prediction result, establishing an LSSVM error correction model for a multi-step prediction error, further extracting a change rule of a small part of ship motion data contained in a multi-step error sequence, and correcting to obtain a final ship motion multi-step prediction result and outputting the final ship motion multi-step prediction result. According to the invention, the stability and accuracy of ship motion attitude multi-step prediction are improved.