Abstract: By using BP，RBF neural network and Matlab6.5 software，the model of prediction of specific impulse of solid propellant rocket engine under the condition of small sample data is established.Compared two kinds of methods，it is shown that the established model is reasonable.So the corresponding controlling method of prediction error and the mode selecting criterion of the specific impulse for new-type solid rocket engine are presented.Results indicate that the network mode provides in the paper can achieve high accurate prediction for the specific impulse property of solid rocket engine under the condition of small sample data.

Key words: solid rocket engine specific impulse, RBF neural network, BP neural network, Matlab software, prediction error control, mode selection criterion

摘要： 运用BP和RBF神经网络方法和Matlab6.5工具软件建立了小样本数据条件下的固体火箭发动机比冲的神经网络预测模型，并将两种神经网络预报的效果进行了比较。两种网络较好的预报效果，表明建立的预测模型是合理的。在此基础上，提出了与之相应的预报误差控制方法及其新型号发动机比冲预测的选模判据。数值实验结果表明，所提供的网络模型可在小样本数据条件下实现发动机比冲性能的高精度预测。

关键词: 固体发动机比冲, RBF神经网络, BP神经网络, Matlab软件, 预报误差控制, 选模判据