基于随机森林算法的内河船舶油耗预测模型

摘要

准确的船舶油耗预测模型是船舶实现各项航行优化措施的基础.以长江干线某旅游船为研究对象,通过安装信息采集系统获得了大量的船舶实时营运数据.通过理论分析得出影响船舶油耗的主要因素为风速、风向、水深、水流速度和船舶航速;改进了随机森林建模时参数的设置方法,提出一种变量的重要性测度方法;对去噪处理后数据进行系统抽样并进行归一化处理,得到建模的样本数据;把样本数据按0.7∶0.3的比例随机分为训练样本和测试样本,对训练样本采用随机森林(RF)算法建立油耗预测模型;通过模型预测测试样本的油耗值,与实测数据对比,结果显示预测误差低于6.8%,优于BP神经网络与支持向量机(SVM)的预测结果;分析模型中各变量的重要性顺序为:航速>水流速度>水深>风速>风向,利用偏相关分析得到了单个因素与油耗间的定量关系.%An accurate model to predict fuel consumption of ships is the basis for optimizing ship navigation.Taking a cruise ship in the Yangtze River as a case study, a large volume of data on ship operations is collected by an information acquisition system.Based on theoretical analysis, the main factors that influence fuel consumption of the ship are identified, which are wind speed, wind direction, water depth, water velocity, and ship speed.A method of setting parameters of random forest model is improved and a way to measure the significance of variables is proposed.Sample data is obtained by systematic samples after de-noise process.The data is then randomly divided into training samples and testing samples by a ratio of 0.7 to 0.3.A prediction model of fuel consumption is developed by using random forest (RF) algorithm to address the training samples.Compared with the measured data, the errors are within 6.8%, which is better than the model established by utilizing BP neural network or support vector machine (SVM) with same samples.Order of the importance of each variable is: ship speed > water velocity > water depth > wind speed > wind direction.Finally, the quantitative relationship between a single factor and fuel consumption is analyzed by using partial correlation analysis.