人工知能を用いたLNG船スポット運賃予測に関する研究
Abstract
Natural gas is considered an environmentally friendly energy source, and its demand is continuously increasing. Consequently, the volatility of freight rates for Liquefied Natural Gas (LNG) carriers is also rising. Given this high volatility, there is a need for research to predict freight rates in advance, aiding the decision-making processes of shipping companies. While numerous studies have focused on freight rate prediction for various types of ships, research specifically targeting LNG carriers remains limited. This study exclusively utilizes freight data from 160K LNG carriers and employs the Long Short-Term Memory (LSTM) model, enhanced through hyperparameter tuning, to predict spot freight rates. Additionally, the study compares the predictive performance of the LSTM model with that of the Auto-Regressive Integrated Moving Average (ARIMA) and SARIMA models, which are well-established in time series analysis. The prediction experiments reveal that the LSTM-based model, in terms of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE), outperforms the others, offering the most accurate freight rate predictions. However, the hyperparameter-tuned model shows proficiency in predicting sudden increases in freight rates. This study suggests that using time series data alone can enhance the objectivity of freight rate predictions. Future research involving comparative analyses and experiments across various predictive models in high-performance computing environments is expected to improve prediction performance further. Such advancements would be valuable for forecasting shipping market conditions more effectively.