摘要
针对汽油辛烷值损失数据中异常采样数据,本文提出一种基于机器学习的汽油辛烷值数据处理方法。该方法包括基于XGBoost的缺失值预测分析及基于Random Forest的异常值处理、再利用基于ARMA算法的特征降维模型,将模型的预测值与真实值进行对比,结果表明在测试集上的准确率为91.31%。经过异常值处理模型修复的数据满足辛烷值损失要求,可为后续降低辛烷值损失提供数据支撑和主要特征改善提供依据。
This paper proposes a machine learning-based method for processing petrol octane data with anomalous sampling data.The method includes XGBoost-based missing value prediction analysis and Random Forest-based outlier processing,and then uses an ARMA-based feature dimensionality reduction model to compare the predicted values of the model with the real values,and the results show that the accuracy of the model is 91.31%on the test set.The data repaired by the outlier processing model meets the octane loss requirements and can provide data support and a basis for subsequent octane loss reduction and major feature improvement.
作者
游长莉
唐成章
胡江宇
You Changli;Tang Chengzhang;Hu Jiangyu
出处
《时代汽车》
2022年第18期36-39,共4页
Auto Time
基金
贵州省教育厅青年科技人才成长项目(黔教合KY字[2022]111号)。
关键词
特征降维
随机森林
自回归滑动平均模型
BP神经网络
Feature dimensionality reduction
random forests
autoregressive sliding average models
BP neural networks
