基于LSSVM的VVP推力系数预报

Hydrodynamic Coefficients of VVP Prediction Based on LS-SVM

针对神经网络存在结构较难确定、训练易陷入局部最优以及容易过学习等问题和标准SVM训练速度较慢等问题,提出最小二乘支持向量机算法,最小二乘支持向量机算法(LS-SVM)具有比其他非线性函数逼近方法具有更强的泛化能力;并且LS-SVM采用径向基核函数,得到LSSVM模型的待定参数比标准支持向量机少,仅为2个。将最小二乘支持向量机(LS-SVM)应用于全方位推进器周期螺距状态推力系数和转矩系数预报,与神经网络预测的结果比较表明:最小二乘支持向量机对全方位推进器周期螺距状态的推力系数和转矩系数的预报结构的精度明显高于采用BP神经网络能进行预报的结果,预测效果好,能够满足工程应用的要求。

Because NN has the problem of uncertain configuration,local optimization,over-learn,and the lower speed of standard SVM,we advanced the LSSVM which has more generalization ability compared with other method of nonlinear function approach,and the LSSVM model has less undetermined parameter which is two compared with standard SVM because LSSVM adopted radial basis kernel function.In this paper,LSSVM is applied in the thrust coefficient and model coefficient prediction of variable vector propeller in cyclic pitch,compared with NN prediction the result shows that,the value predicted by LSSVM has higher precision.So it is feasible to adopt LSSVM to predict the thrust coefficient and model coefficient of variable vector propeller in cyclic pitch,the result is better,and it can also satisfy the need of engineering application.