Open AccessPrediction of vessel propulsion power from machine learning models based on synchronized AIS-, ship performance measurements and ECMWF weather data
Author(s) -
Liang Qin,
Hans Anton Tvete,
Hendrik Brinks
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/929/1/012012
Subject(s) - computer science , spark (programming language) , propulsion , range (aeronautics) , container (type theory) , performance prediction , scope (computer science) , identification (biology) , machine learning , artificial intelligence , data mining , simulation , engineering , mechanical engineering , botany , biology , programming language , aerospace engineering
In this paper, AIS (Automatic Identification System), ship performance measurement and ECMWF (European Centre for Medium-Range Weather Forecasts) weather data were synchronized as a complete dataset. Detailed processing steps and methods are introduced which can be used as best practice for future related studies. All the data preparation was processed on a spark cluster. The optimization and turning of cluster performance will also be introduced. The synchronized dataset was adapted to train different machine learning models to predict ship propulsion power. The dataset includes 228 container vessels covering a time scale of 50 months. The performance of deep learning models with different architecture was compared and discussed. Compared to previous paper [1] in the same project, this paper is an extended scenario which combines the data adapted in scenario 1 and 2. The analysis of the models’ performance in different scenarios was discussed. More features were included in scenario 3 (this study). Hence, the best performing model from scenario 3 has more complex structure compared to scenario 1 and 2. The overall absolute R 2 score for test data is slightly lower than scenario 2. However, the performance for individual ships (relative R 2 score) is much better. This means, models that consider more features (operation, ship characters and environment) are beneficial for individual analysis. For general fleet or wide scope analysis, models that require less data and fewer features are better.