基于LASSO的船舶日油耗预测研究
Prediction of Daily Fuel Consumption of Ship Based on LASSO
-
摘要: 随着船舶燃油价格的不断上涨和国际海事组织对船舶碳排放的愈发限控,提高船舶油耗效率逐渐得到广泛关注,而船舶油耗预测是提高船舶油耗效率的重要前提。目前,在高维、稀疏的船舶油耗数据集中,众多算法存在可解释性低、泛化能力差的问题,为准确、有效地预测船舶日耗油量,提出一种基于套索(Least Absolute Shrinkage and Selection Operator,LASSO)算法的船舶日油耗预测模型。以某货船午时报告油耗数据为例,对船舶油耗数据进行了预测,并与岭回归(Ridge)、最小二乘法(Ordinary Least Squares, OLS)和人工神经网络(Artificial Neural Networks, ANN)进行对比,其均方误差均值分别降低了0.07、0.08和3.77。Abstract: Currently published or used fuel consumption prediction algorithms have drawbacks, such as low interpretability and poor generalization ability for the ship fuel consumption data set, which features high dimensionality and sparsity. A prediction algorithm based on least absolute shrinkage and selection operator (LASSO) is introduced to improve them. The application of the method is demonstrated with a set of fuel consumption data from a cargo ship's noon reports. The result is compared to those with Ridge regression, ordinary least squares method and artificial network, which shows that the average MSE with LASSO is 0.07, 0.08, 3.77 lower, respectively.