基于小波滤波和LSTM 神经网络的船舶运动极短期预报研究

Extreme short-term prediction of ship motions based on wavelet filter and LSTM neural network

为了提高船舶运动极短期预报精度及预报时间长度,本文采用小波多分辨率分析方法,将含有噪声的船舶运动信号进行了多尺度小波变换,通过采用阈值函数法对各尺度下细节信号的小波系数进行处理,对小波分解层数、小波基函数、阈值处理方法进行了深入研究,并通过模型试验数据对滤波效果进行了验证分析,实现了船舶运动信号的小波滤波。进一步针对船舶运动的非线性特性,基于深度神经网络的非线性映射能力,建立了基于LSTM网络的多步直接映射船舶运动极短期预报模型,并采用滤波后的船舶运动数据进行了不同工况下的预报分析。结果表明,不同时间长度的预报与试验结果幅值和相位吻合较好,验证了所建立的极短期预报模型的可行性。

In this paper,the wavelet analysis method is applied in the de-noising of test data in order to im⁃prove the prediction precision of ship motions.The test data of ship motions are decomposed by the multi-reso⁃lution theory.The filtering performance is validated by the test data of ship motions.On the basis of analytical results of the transforming characteristics of noise signal,the reasonable level of wavelet decomposition,wave⁃let basis function and threshold de-noising method are confirmed.Meanwhile,the extreme short time predic⁃tion model of ship motions is proposed based on Long Short Term Memory(LSTM)in order to improve predic⁃tion accuracy.The test results show that the prediction model has excellent performance.The most important outcome of the investigation is that the motion amplitude and phase are very well represented for different pre⁃diction interval lengths.