基于高斯混合模型的发动机稳态数据特征值提取方法

Eigenvalue Extraction Method for Engine Steady-state Data Based on Gaussian Mixture Model

因航空发动机的工作环境及工作特性所致,其稳态试验数据常伴有噪声干扰,对稳态值计算结果的准确性造成影响。在对稳态数据进行正态性检验后,利用稳态数据来源的正态特性,以及利用混合模型对数据的良好回归特性,基于高斯混合模型对稳态数据进行筛选分类。依托发动机稳态数据分布形式的相近性与数据本身的统计特性,来确定稳态数据特征值的提取方法。在对发动机稳态数据进行数据筛选以及对比不同稳态数据片段后,验证模型方法的数据降噪效果以及稳态数据特征值计算结果与同一稳定工作状态数据片段的选取不相关性。结果表明:从仿真数据的筛选结果以及不同稳态数据片段的验证结果可知,该方法具有较强的稳定性,可有效筛选出发动机稳态点数据,并准确计算出发动机稳态值,一般收敛结果相对误差在0.2%以内。

Due to the working environment and characteristics of aeroengine,its steady-state test data were often accompanied by noise interference,which affected the accuracy of steady-state value calculation results.After the normality test of the steady-state data,the Gaussian mixture model was used to screen and classify the steady-state data based on the normal characteristics of the steady-state data source and the good regression characteristics of the mixed model.Based on the similarity of the distribution form of engine steady-state data and the statistical characteristics of the data itself,the extraction method of steady-state data eigenvalues was determined.After data screening of engine steady-state data and comparison of different steady-state data segments,the data noise reduction effect of the model method and the calculation results of steady-state data eigenvalues were verified to be independent to the segment selection of the same steady-state data.The results show that from the screening results of simulation data and the verification results of different steady-state data segments,the method has strong stability,which can effectively screen the steady-state point data of the engine and accurately calculate the steady-state value of the engine.Generally,the relative error of the convergence result is within 0.2%.