均方误差
人工神经网络
计算机科学
回归分析
回归
遗传算法
机器学习
统计
人工智能
计量经济学
数学
作者
Payman Eslami,Kihyo Jung,Daewon Lee,Amir Tjolleng
摘要
Short-term prediction of tanker freight rates (TFRs) is strategically important to stakeholders in the oil shipping industry. This study develops a hybrid TFR prediction model based on an artificial neural network (ANN) and an adaptive genetic algorithm (AGA). The AGA adaptively searches satisficing network parameters such as input delay size. The ANN iteratively optimizes a prediction network considering parsimonious variables and time-lag effects as predictors. Three parsimonious variables (crude oil price, fleet productivity and bunker price) are selected by a stepwise regression of TFR variables. The article compares the performance of its hybrid model with two traditional approaches (regression and moving average), as well as with the findings of existing ANN studies. The results of our model (root mean squared error (RMSE)=11.2 WS) are not only significantly superior to the regression approach (RMSE=21.6 WS) and the moving average approach (RMSE=17.5 WS), but are even slightly superior to the results of existing ANN studies (RMSE=14.6 WS–15.8 WS).
科研通智能强力驱动
Strongly Powered by AbleSci AI