结合ResNet和特征工程的QAR数据预测方法
QAR Data Prediction Method Combining ResNet and Feature Engineering
潘卫军 1尹子锐 1冷元飞 1王安鼎1
作者信息
- 1. 中国民用航空飞行学院空中交通管理学院,四川 广汉 618307
- 折叠
摘要
为解决QAR数据存在的数据缺失、数据异常等问题,提出一种将神经网络模型与特征工程相结合的方法,用于实现对飞机的QAR数据中飞行状态参数的精准预测.首先通过引入ResNet(残差神经网络)模型大幅加深了预测网络模型的深度,同时缓解了深层网络带来的梯度消失/爆炸问题,完成了预测精度的一次提升.之后通过Pearson相关系数与随机森林相结合的特征提取方法提取训练特征,训练后得到更加精确的预测模型,完成了预测精度的二次提升.结果表明,特征工程与模型优化相结合的处理方法更加精确、高效,为QAR缺失数据补充和QAR数据异常检测提供了一种新方法,提升了数据的质量,可更好地进行数据分析与挖掘工作.
Abstract
In order to solve the problems of missing data and abnormal data in QAR data,this paper proposes a method that combines neural network model and feature engineering to achieve accurate prediction of flight state pa-rameters in aircraft QAR data.First,by introducing the ResNet(residual neural network)model,the depth of the esti-mation network model was greatly deepened,and the gradient disappearance/explosion problem caused by the deep network was alleviated,and the estimation accuracy was improved.Then,the training features were extracted by the feature extraction method combining Pearson correlation coefficient and random forest,and a more accurate estimation model was obtained after training,which completed the secondary improvement of estimation accuracy.The results show that the combination of feature engineering and model optimization is more accurate and efficient,providing a new method for QAR missing data supplementation and QAR data anomaly detection,improving the quality of data,and enabling better data analysis and mining.
关键词
特征工程/相关系数/随机森林Key words
Feature engineering/Correlation coefficient/Random forest引用本文复制引用
出版年
2024
NETL
NSTL
万方数据