Big multi-step motion and mooring load forecasting of fish cage using a novel hybrid model based on Bi-SLSTM neural network

Abstract

Accurate forecasting of motion and mooring loads in fish cages is vital for efficient field monitoring and numerical simulations. This research introduces an innovative hybrid model that leverages Bidirectional Stateful Long Short-Term Memory (Bi-SLSTM) neural networks to predict the dynamic behavior of fish cages. By incorporating bidirectional data flow and state-preserving mechanisms, the model enhances the accuracy of multi-step predictions. Performance is assessed through the Root Mean Square Error (RMSE) and the Trapezoidal-based Integral Similarity Index (R_trapz). Results show that Bi-SLSTM outperforms Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Long Short-Term Memory (BiLSTM) models, achieving R_trapz values above 0.85 in large multi-step predictions. Further analysis reveals that combining wave and motion data as inputs improves prediction accuracy, with surge and heave predictions reaching 91 % and 98 %, respectively. The hybrid strategy combining motion response and load data provides the best performance for mooring load prediction, with R_trapz values exceeding 0.85 across different output steps. The Bi-SLSTM model demonstrates strong prediction capability and robustness in forecasting cage dynamics.