Predicting Ship Fuel Consumption based on LSTM Neural Network

With the rapid development of marine trade and transportation, a large number of ships sail at sea. The ship energy conservation and environment protection are becoming public interested problems. Precise prediction of ship fuel consumption is vital for controlling emissions from ships. Some scholars used regression analysis to interpret the relations between fuel consumption and its influence factors. However, considering that there are a large amount of influence factors affecting ship fuel consumption, such as pitch, ship speed, wind and wave, some machine learning models are employed. While, ship fuel consumption data is time-related. A time series problem arises. The regression analysis and machine learning models cannot capture the time series characteristics. Therefore, LSTM neural network is applied in this paper. Experimentally, we compared LSTM with three traditional machine learning models, namely linear regression (LR), support vector regression (SVR), and artificial neural network (ANN). It found that the prediction accuracy can be improved with 11.8% compared with artificial neural network. The LSTM model can catch the time series characteristics of fuel consumption data and get a higher accuracy.