结合先验知识的涡喷发动机神经网络点火识别

在分析某弹用涡喷发动机燃烧室点火特性的基础上，提出利用结合先验知识的神经网络来识别发动机点火时刻的方法。一方面，该方法通过先验知识对神经网络原始输人样本进行更合理的重组。另一方面，用神经网络的信息分布来描述燃烧室的点火条件，可以全面融合影响点火性能的各种因素。根据与试车数据的比较，结果显示：在少量训练样本情况下，神经网络识别器能保证较好的收敛速度和泛化能力。该方法可用于发动机点火点在线识别。

Identification of turbojet ignition based on neural network with a prior

The method of neural network incorporating a prior was given to recognize ignition time after analyzing ignition property of some missile turbojet burner. With this method, the initial input parameters were better reconstructed and all factors of ignition performance were completely synthesized, using the neural network distribution information to describe the ignition condition of burner. The results compared with test data indicate that neural network recognition has good convergence speed and generalization capability in case of small training samples. The proposed method can be used in online recognition of engine ignition.