# Ship Motion Attitude Prediction Model Based on FMD-IBKA-BTGN

## TL;DR

This paper introduces a new hybrid model for predicting ship motion in rough seas, improving accuracy over existing methods.

## Contribution

The novel FMD-IBKA-BTGN model combines signal decomposition, optimization, and deep learning for enhanced ship motion prediction.

## Key findings

- The model reduced MAPE by 20.38% compared to LSTM in Sea State 4 conditions.
- It achieved lower RMSE, MAE, and MSE than six other models, showing better generalization.
- FMD decomposition and IBKA optimization significantly improved prediction accuracy.

## Abstract

Accurate prediction of ship motion attitude remains a significant challenge due to the inherent non-stationarity and strong stochasticity of marine environmental conditions. To address this issue, this study proposes FMD-IBKA-BTGN, a hybrid model combining Feature Mode Decomposition (FMD), Improved Black-winged Kite Algorithm (IBKA), and a Bidirectional Temporal Convolutional Network with Gated Recurrent Unit (BTGN). First, FMD decomposes motion signals into intrinsic modes. Subsequently, IBKA—enhanced with chaotic mapping and Lévy flights—optimizes BTGN hyperparameters for global search efficiency. Finally, predictions from all components are ensembled for final output. Experiments on a 240 m vessel in Sea State 4 show our model outperforms six models, reducing MAPE by 20.38%, RMSE by 7.4%, MAE by 4.2%, and MSE by 0.97% versus LSTM. The model enhances both prediction accuracy and generalization.