航段截尾油耗数据的区间估计方法  

Interval estimation method for flight segment truncated fuel consumption data

作　　者：陈静杰 梁国栋[1,2] 刘家学 CHEN Jing-jie;LIANG Guo-dong;LIU Jia-xue(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;Research Center for Environment and Sustainable Development of CAAC,Civil Aviation University of China,Tianjin 300300,China;National Engineering Laboratory for Integrated Traffic Data Application Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300 [2]中国民航大学中国民航环境与可持续发展研究中心(智库),天津300300 [3]中国民航大学综合交通大数据应用技术国家工程实验室,天津300300

出　　处：《计算机工程与设计》2023年第10期3103-3109,共7页Computer Engineering and Design

基　　金：中国民用航空局安全能力建设基金项目(SKZ49420210017);2021年天津市教委社会科学重大基金项目(2021JWZD39)。

摘　　要：对航段截尾油耗数据进行区间估计时,数据分布的稀疏性及非正态性会导致传统基于单因素的油耗估计区间难以建立。针对上述问题,提出基于分类和沙普利加性解释(classification and Shapley additive explanations,C-SHAP)的改进分位数回归森林区间估计(quantile regression forest,QRF)方法。通过C-SHAP方法,筛选全航程和各飞行阶段特征得到最优输入特征集;采用随机过采样算法增加训练集中截尾油耗样本的权值,提高QRF模型的估计性能;通过QRF估计给定上、下限油耗条件分位数,构建估计区间。实验结果表明,该方法的特征选择合理、估计区间质量较高。The sparse and non-normal characteristics of the data distribution when estimating intervals for flight segment truncated fuel consumption data can make it difficult to establish traditional single-factor based fuel consumption estimation inte-rvals.To address these issues,an improved quantile regression forest(QRF)method based on classification and Shapley additive explanations(C-SHAP)was proposed.The optimal input feature set was obtained by filtering the full range and each flight phase features using C-SHAP method.A random oversampling algorithm was used to increase the weights of the truncated fuel consumption samples in the training set to improve the estimation performance of the QRF model.The estimation intervals were constructed by QRF estimation given the upper and lower conditional quantiles of fuel consumption.Experimental results show that the method has reasonable feature selection and high quality of the estimated intervals.

关 键 词：截尾数据 数据分布 分类 沙普利加性解释 随机过采样 分位数回归森林 区间估计 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]