基于CRBMs-RVR的涡轴发动机输出功率衰退预测

Output Power Degradation Prediction of Turboshaft Engine Based on Continuous Restricted Boltzmann Machines and Relevance Vector Regression

针对涡轴发动机全寿命期内输出功率衰退预测问题,提出一种含多层连续受限玻尔兹曼机(CRBMs)深度特征提取的相关向量回归(RVR)功率预测方法。对发动机气路部件测量数据进行重构,利用CRBMs深度网络提取数据深层特征,将特征数据作为RVR模型的输入,实现对输出功率的预测,并对预测结果提供概率分布。以某型双转子涡轴发动机部件级模型为试验对象,模拟全寿命期内发动机气路部件性能退化,对输出功率进行衰退预测。试验结果表明:基于CRBMs-RVR的预测模型与传统的RVR预测模型相比,训练时间缩短30.2%,预测结果的均方根误差减小64.6%;与基于主成分分析(PCA)进行特征提取的PCA-RVR预测模型相比,预测结果均方根误差减小42.4%,验证了所提出的预测方法具有模型结构简单、预测精度高、可提供概率式输出的优点。

Aiming at the problem of output power degradation prediction of turboshaft during the whole lifetime,a Relevance Vector Regression(RVR)power prediction method with multi-layer Continuous Restricted Boltzmann Machines(CRBMs)depth feature extraction was proposed.The measurement data of engine gas components were reconstructed,and CRBMs deep network was used to extract the deep features of data.The feature data was used as the input of RVR model to realize the output power prediction of,and the probability distribu⁃tion of the predicted results was provided.Taking the component-level model of a twin-rotor turboshaft engine as the test object,the perfor⁃mance degradation of engine gas path components in the whole lifetime was simulated,and the output power degradation prediction was car⁃ried out.The test results show that compared with the traditional RVR prediction model,the training time of the prediction model based on CRBMs-RVR is reduced by about 30.2%,and the root mean square error of the prediction results is reduced by about 64.6%.Compared with the PCA-RVR prediction model based on Principal Component Analysis(PCA)for feature extraction,the root mean square error of the prediction results is decreased by about 42.4%,which indicates that the proposed prediction method has the advantages of simple model structure,high prediction accuracy and providing the output probability.