Prediction of Ocean Velocity during Typhoon Using a GA-BP Neural Network Model

When a typhoon forms and approaches land, it generates strong winds and large waves, which significantly impact the ocean. Accurately predicting ocean current speeds during typhoons is crucial for maritime departments and related organizations to forecast and prevent potential maritime accidents, ensuring the safety of lives and property. Moreover, accurate predictions of ocean current speeds are vital for researchers studying marine environments, climate change, and other related fields during typhoons. In this study, we utilized high frequency surface wave radar (HFSWR) sea state inversion to obtain data on ocean current speeds during a typhoon. We employed a genetic algorithm-optimized BP neural network prediction model (GA-BP model) to forecast the ocean current speeds. The accuracy of the model’s predictions was evaluated using the mean average percentage error (MAPE). Furthermore, we compared the prediction results of the GA-BP model with those of the long short-term memory network prediction model (LSTM model), convolutional neural network model (CNN model), and BP neural network prediction model, using root mean square error (RMSE) as the evaluation metric for all four models. Our findings indicate that the CNN, LSTM, and BP models produced average absolute percentage errors greater than 20%. Conversely, the GA-BP model achieved an average absolute percentage error below 20%, meeting the requirements for medium to high-accuracy predictions. The RMSE values for the CNN, LSTM, BP, and GA-BP models were 0.2333, 0.15007, 0.1526, and 0.12614, respectively. Overall, the GA-BP prediction model demonstrated superior performance in forecasting ocean current speeds during typhoons, followed by the LSTM, BP, and CNN models in decreasing order of prediction effectiveness. This study successfully predicted ocean current speeds during a typhoon, considering the limited number of samples and the influence of a single time factor, using the GA-BP neural network prediction model.