Research on Ocean Current Prediction Based on POD-LSTM Hybrid Model

Although ocean current prediction is crucial to climate, ecology, and shipping, its complex nonlinear characteristics make it difficult to predict accurately by traditional methods. Advances in ocean observation technology have opened up new ways for data-driven models. In general, in this study, the data set was decomposed into average snapshots by Reynolds decomposition Matrix and pulsation snapshot matrix, and POD decomposition of pulsation, and then introduced Long Short-Term Memory (LSTM) to predict the ocean current, and then obtains the ocean current prediction model, and then uses the resulting ocean current model to find a missing submarine that lost contact and power and predict the location of the submarine. Then, a submarine model is established to predict the position of the submarine by applying the ocean current model and the principle of fluid mechanics. The accuracy and practicability of the ocean current prediction model are demonstrated by comparing the prediction results with the actual results The uncertainty of the predicted results of the model was only . From the data, it can be seen that the model predicts better with smaller trajectory offsets.This experiment shows the feasibility of applying LSTM neural network combined with POD decomposition to the field of ocean current prediction, and also provides experience for applying more methods of machine learning to the field of navigation and sea in the future.