Estimation of Ship Performance at Sea Using Neural Network

Abstract

In recent years, on-board monitoring has been performed widely to estimate ship performance at actual sea. To analyse monitoring data, the authors tried to make estimation models which have high predictive power and high explanatory power. At first, the authors cleaned data using reconstruction error by autoencoder. Then, the authors made estimation models using 24 neural networks and bagging to predict SHP and log speed of 2 ships. Prediction error of test data is as follows. MAPE is 1-3%, RMSPE is 2-8%, R2 score is 0.96-0.99. Also, the authors confirmed estimation models can estimate ship’s performance in calm sea and effects of hull fouling, aging and disturbance due to wind and wave. It can be said that the method proposed in this paper is effective to make the estimation models with high predictive power and high explanatory power.