基于决策树特征提取与RSM_LightGBM的涡扇发动机RUL预测

Prediction of remaining useful life of aircraft engines based on decision tree feature extraction and RSM_LightGBM

针对现有航空发动机剩余使用寿命预测(RUL)精度低及传感器监测参数提取困难等问题,提出了一种基于决策树特征提取与随机搜索算法优化LightGBM的航空发动机RUL预测模型。首先,对航空发动机历史监测参数进行分析,利用决策树算法计算监测参数对发动机寿命周期的重要性贡献程度,提取重要特征后对数据进行归一化处理,降低数据量纲对预测模型的影响;其次,根据航空发动机的历史衰退特征,为发动机设置阈值标签,表征发动机的性能退化特点。最后,利用随机搜索算法对LightGBM中的超参数进行寻优,获得RMSE最小值。在CMAPSS数据集上进行了实验验证。结果表明,与其他构建模型的最优值相比,所提方法在多个评价指标下具有更好的综合性能,有效提升了航空发动机RUL预测的精准度。

Aiming at the low prediction accuracy of current aircraft engine remaining useful life(RUL)prediction models and the difficulty of extracting sensor monitoring parameters,an aircraft engine RUL prediction model was proposed based on decision tree feature extraction and random search algorithm optimized LightGBM.Firstly,the historical monitoring parameters of aircraft engines were analyzed.The decision tree algorithm was used to calculate the importance contribution degree of monitoring parameters to the engine life cycle,and important features were extracted.Then,the data was normalized to reduce the impact of dimensional differences on the prediction model.Secondly,based on the historical degradation characteristics of aircraft engines,threshold labels were set for the engines to represent their performance degradation characteristics.Finally,the random search algorithm was used to optimize the hyper parameters in LightGBM and obtain the minimum RMSE.Experimental verification was carried out on the CMAPSS dataset.The experimental results showed that,compared with the optimal values obtained by other models,the proposed method had better comprehensive performance in multiple evaluation indicators,and effectively improved the accuracy of aircraft engine RUL prediction.