Real Time Estimation of Local Wave Characteristics from Ship Motions Using Artificial Neural Networks

Sea state identification using machine learning—A comparative study based on in-service data from a container vessel

“…for both wind and swell waves. Scholcz and Mak [22] extend the work of Düz et al [20] and present a deep learning methodology for the non-parametric estimation of the directional wave spectrum based on wave radar data using a convolutional encoder-decoder network applied to in-service time series data. Lastly, Han et al [23] provide an investigation for non-parametric SSE and establish an approach based on a generative adversarial network, in which the generator predicts the 2D spectrum relying on cross response spectra, and the discriminator classifies the validity of the prediction.…”

“…Their so-called SSENET features attention mechanisms and residual skip connections for enhanced performance. Moreover, Düz et al [20] present a real-time multivariate time series regression approach for integral sea state parameters applying multiple deep architectures on 2.5-min motion samples of a frigate-type ship. One distinct aspect of their work is the procedure of transfer learning: Initially, the model is trained on simulated data obtained from time domain potential theory calculations and then re-trained on in-situ measurement data.…”

“…For comparison of ResNet and Inception to state of the art literature, two models are adopted as benchmarks: The convolutional neural network (CNN) from Kawai et al [21] and the multichannel convolutional LSTM (MLSTM) network as proposed by Düz et al [20]. The former network was applied in the frequency domain on cross spectral sequences, whereas the latter one used raw time series as input for the prediction of sea state parameters.…”

Estimation of Sea State Parameters from Ship Motion Responses Using Attention-based Neural Networks

“…The before-mentioned papers are related to DP application without ship forward-speed. On the other hand, Duz, Mak, Hageman and Grasso (2019) and Mak and Düz (2019) successfully predicted wave height and wave direction under forward-speed conditions. Ship motions in 6-DOF are used as input, and the problem was treated as a multivariate timeseries regression.…”

“…The input data was 3-DOF ship motion time-series, and the output indicated the wave period, relative wave direction and significant wave height. As in Duz et al (2019), the problem was characterized as a multivariate time-series regression. The algorithms are trained with numerically simulated data generated according to the recommended guidelines.…”