LSTM-PSO: Long Short-Term Memory Ship Motion Prediction Based on Particle Swarm Optimization

Yuxin Yao,Liang Han,Jiangyun Wang
DOI: https://doi.org/10.1109/gncc42960.2018.9018688
2018-01-01
Abstract:According to the nonlinear characteristic of ship motion, the ship motion pose will be disturbed by coupling, indefinite period, noise signals, chaotic and some other factors, which leads that it is hard to predict ship motion in the future precisely. Based on the above, and considering the sequence of ship movement, many neural networks have been applied in ship motion prediction, such as LSTM (Long Short-Term Memory, LSTM) and ESN (Echo State Network, ESN). However, there are problems in the parameter setting of ANN (Artificial Neural Network) algorithm, that how to update network parameters during training iterations of the network to avoid iterates getting into local optimum. LSTM with PSO optimization is proposed in this paper. Testing simulation results show that the combination LSTM and PSO improves the accuracy of ship motion prediction.
What problem does this paper attempt to address?