Real-Time Ship Motion Prediction Using Artificial Neural Network

Abstract Models based on Artificial Neural Networks (ANN) have been developed for predicting ship motions using the information about the wave field around the ship and historical time-series of motions. The ANN models developed in this study were able to predict all six degrees of freedom ship motions in irregular wave conditions with different significant waveheight, peak period and wave directions along with directional spreading. Preparation of training, validation and test datasets has been described along with the development and training of ANNs. The models were tested using the observed wave conditions recorded by a wave radar installed onboard the ship. A physics-based approach has been applied when selecting the length of input and output data. The effect of input and output window length on the accuracy of results was further studied by developing two sets of ANNs with different length of input and output window. Performance of both sets of ANNs on training, validation and test datasets has been presented along with detailed investigation on test dataset. Reducing the length of input window and increasing the length of output window was seen to reduce the accuracy of prediction.