LNG船スポット運賃予測への人工知能に基づくハイブリッドモデルの適用
Abstract
In response to the increasing volatility of liquefied natural gas (LNG) spot freight rates, this study applies deep learning models to improve forecasting accuracy. The performance of standalone models—Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN)—is compared with that of hybrid models. The results show that the LSTM+CNN model achieved the highest prediction accuracy, reaching approximately 88% for short-term forecasts. While the standalone CNN model performed poorly in both short- and long-term predictions, its combination with LSTM significantly improved prediction accuracy. This improvement is attributed to CNN’s ability to extract local patterns and LSTM’s strength in capturing long-term dependencies, demonstrating a complementary effect. Furthermore, incorporating the attention mechanism (ATT), which has proven effective in previous studies, enhanced predictive performance, particularly in long-term forecasting. The LSTM+ATT model showed strong results due to the attention mechanism's capability to assign greater weight to important time steps and reduce information loss over longer sequences.