支持向量机在导弹动力系统推力预测中的应用

应用改进型支持向量回归算法ν-SVR,研究了导弹动力系统推力预测问题,讨论了不同核函数和惩罚因子对推力预测的影响。发现选用多层感知器核函数和适当的惩罚因子时,得到的预测模型稳定性好,并且训练时间和预测误差相对较小;同时与BP神经网络模型进行了对比研究,仿真结果表明,支持向量机能够更好地预测发动机推力,是一种研究小样本情况下推力预测的有效方法。

Thrust Prediction of Missile Power System Based on SVM

Thrust prediction of missile power system was studied based on v-SVR, which is the improved regression algorithm of support vector machine (SVM), and the influence of different kernels and different penalty coefficients on the thrust prediction was also discussed. Experimental results show that the prediction model has good stability and relatively small training time and predictive error when multi-layer perceptron kernel and proper penalty coefficient are chosen. Comparison experiments between v-SVR and BP neural networks demonstrate that support vector machine can predict thrust better, and it is an effective approach on studying thrust prediction under small samples.