论文部分内容阅读
介绍一种特殊的前向神经网络——自联想神经网络（Ａｕｔｏａｓｓｏｃｉａｔｉｖｅａｒｔｉｆｉｃｉａｌｎｅｕｒａｌｎｅｔ－ｗｏｒｋｓ，ＡＡＮＮ），然后将发动机参数在全包线、大范围工况下的变化规律与神经网络的非线性映射能力结合起来，开展了将ＡＡＮＮ应用于发动机全包线、大范围工况下参数估计的仿真研究。本文提出的选取测量矢量加入样本集的ＥＭＰ方法，有效地减少了样本集中样本矢量的数目，简化了网络的训练。用ＥＭＰ方法在全包线内仅用７４６组测量矢量作为样本集，在网络训练好后，任选包线内的一工况点作为算例运行发动机模型，所得各参数的稳态估计及动态估计的平均百分比误差＜０．５％。仿真结果表明，上述的参数估值方法是可行的，为进一步实现对发动机控制系统传感器的状态监视和故障诊断打下了基础。
A special kind of forward neural network, called Autoassociative artificial neural network (AANN), is introduced. Then, the variation law of engine parameters under all-envelope and large-scale working conditions is combined with the nonlinear mapping ability of neural network , Carried out the AANN applied to the engine all-enveloping, a large number of operating conditions under the parameter estimation simulation study. The EMP method proposed in this paper to select the sample vector to join the sample vector effectively reduces the number of sample vectors in the sample set and simplifies the training of the network. Using EMP method, only 746 sets of measurement vectors are used in the whole envelope as a sample set. After the network is trained, an operating point in the optional envelope is used as an example to run the engine model. The steady state estimation and dynamic The estimated average percentage error is <0.5%. The simulation results show that the above method of parameter estimation is feasible, which lays the foundation for further monitoring and fault diagnosis of engine control system sensors.