基于数据驱动的导弹伺服液面预测

Prediction of Liquid Level of Missile Servo Based on Data Driven

针对导弹伺服机构液压源液面下降的问题,根据导弹伺服液面实测数据,采用支持向量机、神经网络集成和最小二乘多项式拟合3种数据驱动方法对伺服液面高度进行模型辨识。通过对实测数据仿真分析,发现不同输入维数对预测精度有所影响:当输入维数为4时,支持向量机预测误差最低;不同输入维数下,最小二乘多项式预测误差最稳定,且综合误差最小。

Aiming at solving the descending liquid level of missile servo, the support vector regression （SVR） , neural network ensemble （NNE） and least square polynomial fit （LSPF） are used for model iden- tification based on real data of liquid level of missile servo. The results of simulation demonstrate that differ-ent predicting results can be produced by input dimension, SVR has minimum prediction error, the predic-tion error of LSPF is most stable and LSPF has minimum synthetically error.