基于XGBoost的回归-分类-回归寿命预测模型

Regression-classification-regression life prediction model based on XGBoost

针对现有使用单个模型的剩余寿命预测方法存在精度较低的问题,提出一种基于XGBoost(极端梯度提升)的回归-分类-回归参数优化算法的预测模型。首先对涡扇发动机数据进行预处理,观察各特征与剩余寿命的相关性,筛选出初步的可用特征。然后根据故障预测与健康管理协会制定的有效剩余寿命预测值标准对连续寿命值进行离散标定,以信息增益法筛选出最终特征集。最后利用XGBoost算法分别建立离散标定后的多分类寿命预测模型和各个类别的寿命回归模型,并采用遗传算法优化各模型的参数,通过集成方式得出寿命的预测值。结果表明,基于XGBoost的回归-分类-回归算法预测模型明显优于XGBoost,其中均方根误差(RMSE)降低了27.5%,准确度(accuracy)提高了30%。

To address the problem of low accuracy of existing methods for remaining life prediction using individual models,a prediction model based on XGBoost(extreme random tree)regression-classificationregression parameter optimization algorithm is proposed.The turbofan engine data were first pre-processed to observe the correlation between each feature and the remaining life and to filter out the initial available features.The continuous life values are then discrete calibrated according to the effective remaining life prediction value criteria established by the Failure Prediction and Health Management Association,and the final feature set is filtered by the information gain method.Finally,the discretely calibrated multi-class life prediction model and the life regression model of each category are established by using the XGBoost algorithm,and the parameters of each model are optimized by genetic algorithm,and the predicted value of life is obtained by integration.The results show that the prediction model of XGBoost-based regressionclassification-regression algorithm is significantly better than XGBoost,in which the root mean square error is reduced by 27.5%and the accuracy is improved by 30%.