针对当前单一模型预测航空器滑出时间精度提升存在瓶颈的问题,提出一种结合最大互信息系数(Maximal Information Coefficient,MIC)的迭代注意力特征融合模块(iterative Atten-tional Feature Fusion,iAFF)和Stacking集成学习框架组合的航空器滑出时间预测模型.首先利用MIC提取出与滑出时间相关性较高的因素作为模型原始特征序列;然后以支持向量回归(SVR)、随机森林(RF)、多层感知机(MLP)和极限梯度提升机(XGBoost)为基学习器模型对原始特征进行特征构造,并利用iAFF模块对基学习器得到的构造特征和原始特征进行特征融合,通过MLP对融合后的特征进行学习,最终得到预测滑出时间.经实际算例对比验证表明,与单一模型相比,MIC-iAFF-Stacking集成学习模型在±2、±3、±5 min误差范围内的预测精度分别提升了6.14%、6.40%、2.31%,证明了该模型在滑出时间预测中的有效性.
Aircraft departure taxi time prediction based on MIC-iAFF-Stacking ensemble learning
An aircraft departure taxi time prediction model,which combines maximal information co-efficient(MIC),iterative attentional feature fusion(iAFF),and Stacking ensemble learning frame-work,was developed to improve the prediction accuracy of a single model for aircraft taxi-out time.First,MIC was used to extract the factors that highly correlate with taxi-out time to serve as the origi-nal feature sequence of the model.Subsequently,support vector regression(SVR),random forest(RF),multilayer perceptron(MLP),and extreme gradient boosting machine(XGBoost)were used as the components of the base learner model to construct the original features.The iAFF module was used to fuse the original and structural features obtained by the base learner model,and the fused fea-tures were learned using the MLP to finally predict the taxi-out time.The comparison and verifica-tion of actual examples show that the prediction accuracy of MIC-iAFF-Stacking ensemble learning model was higher than that of a single model by 6.14%,6.40%and 2.31%in the error range of±2,±3,and±5 min,respectively,which proves the effectiveness of the model in predicting the taxi-out time.
air transportationdeparture taxiing timemaximal information coefficientattentional feature fusionStacking ensemble learning
国家科技期刊平台
NSTL
万方数据
