针对航空发动机排气温度的变化过程受复杂非线性时变因素的影响而难以用精确数学模型描述的问题,提出了卷积和离散过程神经网络(CSDPNN)模型,并将其应用于航空发动机排气温度(EGT)预测。该模型以离散样本作为直接输入,采用卷积和算法实现对时间累积效应的处理。相较于以连续函数作为输入的过程神经网络(PNN),不需要拟合离散样本得到连续函数后进行正交基展开,减少了精度损失,具有更高的预测精度。给出了卷积和离散过程神经网络模型的学习算法,并通过对Mackey-Glass混沌时间序列的预测对提出的方法进行应用说明和验证。通过航空发动机EGT预测实例,并与卷积和离散过程神经网络模型的连续函数输入过程神经网格以及传统人工神经网络(ANN)的预测结果进行了对比。结果表明,相较于连续函数输入过程神经网络以及传统人工神经网络,卷积和离散过程神经网络具有更高的预测精度,且对于EGT的预测具有较好的适应性,因而为航空发动机EGT预测提供了一种有效的方法。
The changing process of aeroengine exhausted gas temperature (EGT) is affected by complicated nonlinear time varying factors, which make it difficult to construct its mathematic model. To cope with this issue, a convolution sum discrete process neural network (CSDPNN) model is proposed and used for EGT prediction. This model directly utilizes discrete sampling points as input, and uses the convolution sum to deal with the time accumulation process. Compared with the process neural network(PNN)with continuous function inputs, there is no need to fit the sampling points to get input functions and then to decompose them by orthogonal basis functions which can lead to precision loss. Therefore, this model can achieve higher prediction precision. A learning algorithm for this model is also developed, and the model is explained and validated via Mackey-Glass chaos time series prediction. Then, the model is adopted to predict a real EGT time series. The prediction results are compared with results obtained by the process neural network with function inputs and the traditional artificial neural network(ANN), which proves that CSDPNN model has higher precision than the other two networks, and it exhibits good adaptability to EGT prediction. This model offers an effective method for real EGT prediction.
[1] Mercer C R, Simon L D, Hunter W G, et al. Fundamental technology development for gas-turbine engine health management. NASA-20070022364, 2007.
[2] Liang J. On-condition maintenance decision research of aero-engine based on cost optimization. Nanjing: College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, 2004. (in Chinese) 梁剑.基于成本优化的民用航空发动机视情维修决策研究.南京:南京航空航天大学民航学院, 2004.
[3] Jaw L C. Recent advancements in aircraft engine health management technologies and recommendations for the next step. ASME GT2005-68625, 2005.
[4] Zuo H F, Zhang H J, Rong X. Condition based aero-engine maintenance decision method using proportional hazards model. Journal of Aerospace Power, 2006, 21(4): 716-721. (in Chinese) 左洪福, 张海军, 戎翔. 基于比例风险模型的航空发动机视情维修决策. 航空动力学报, 2006, 21(4): 716-721.
[5] Beijing Aircraft Maintenance and Engineering Corporation. The function and handling of ΔEGT and EGT margin in aircraft engine condition monitoring. The Reliability Research Program Summary of Civil Aviation Engine, 2000. (in Chinese) 北京飞机维修工程有限公司. ΔEGT与EGT 在发动机状态监控中的作用及运用. 民航发动机可靠性研究课题总结, 2000.
[6] Chen G. Forecasting engine performance trend by using structure self-adaptive neural network. Acta Aeronautica et Astronautica Sinica, 2007, 28(3): 535-539. (in Chinese) 陈果. 用结构自适应神经网络预测航空发动机性能势. 航空学报, 2007, 28(3): 535-539.
[7] He X G, Liang J Z. Some theoretical issues on procedure neural networks. Engineering Science, 2000, 2(12): 40-44. (in Chinese) 何新贵, 梁久祯. 过程神经网络的若干理论问题. 中国工程科学, 2000, 2(12): 40-44.
[8] He X G, Liang J Z, Xu S H. Learning and applications of procedure neural networks. Engineering Science, 2001, 3(4): 31-35. (in Chinese) 何新贵, 梁久祯, 许少华. 过程神经网络的训练及其应用. 中国工程科学, 2001, 3(4): 31-35.
[9] Ding G, Zhong S S. Approximation capability analysis of parallel process neural network with application to aircraft engine health condition monitoring. Lecture Notes in Computer Science, 2007, 4493(3): 66-72.
[10] Ding G, Xu M Q, Hou L G. Aeroengine exhaust gas temperature prediction using process neural network. Journal of Aerospace Power, 2009, 24(5): 1035-1039. (in Chinese) 丁刚, 徐敏强, 侯立国. 基于过程神经网络的航空发动机排气温度预测. 航空动力学报, 2009, 24(5): 1035-1039.
[11] Zhong S S, Li Y. Condition monitoring of aeroengine based on wavelet process neural networks. Acta Aeronautica et Astronautica Sinica, 2007, 22(1): 68-71. (in Chinese) 钟诗胜, 李洋. 基于小波过程神经网络的飞机民用航空发动机状态监视. 航空学报, 2007, 22(1): 68-71.
[12] Zhong S S, Luan S G, Ding G. Aeroengine condition monitoring based on process neural networks and aero thermo-parameters. Journal of Nanjing University of Science and Technology, 2006, 30(5): 533-536. (in Chinese) 钟诗胜, 栾圣罡, 丁刚. 基于过程神经网络与气动热力参数的民用航空发动机状态监视. 南京理工大学学报, 2006, 30(5): 533-536.
[13] Xu S H, He X G. Learning algorithms of process neural networks based on orthogonal function basis expansion. Chinese Journal of Computers, 2004, 7(5): 645-650. (in Chinese) 许少华, 何新贵. 基于函数正交基展开的过程神经网络学习算法. 计算机学报, 2004, 7(5): 645-650.
[14] Oppenheim A V, Willsky A S, Nawab S H. Signals and systems. New Jersey: Prentice Hall International Ltd., 1997: 70-79.
[15] Hagan M T, Menhaj M. Training feedforward networks with the Marquardt algorithm. IEEE Transactions on Neural Networks, 1994, 5(6): 989-993.
CJA