传感器动态建模的最小二乘支持向量机方法

Sensor dynamic modeling using least square support vector machines

提出了应用最小二乘支持向量机(LS-SVMs)建立传感器动态模型的方法。LS-SVMs的训练过程遵循的是结构风险最小化原则,而不是通常神经网络的经验误差最小化原则,遵循该原则可获得更好的泛化性能,且不易发生局部最优及过拟合现象,因此可以克服应用人工神经网络建立传感器动态模型的缺陷。通过实例验证了该方法的实用性及可靠性。实验结果表明,即使传感器动态模型存在严重非线性,该方法也仍然有效。

The least squares support vector machines （LS-SVMs） are proposed for nonlinear sensor dynamic modeling. The LS-SVMs were established based on the structural risk minimization principle rather than minimized empirical error principle commonly implemented in the neural networks. The LS-SVMs can achieve higher generalization performance. Also, local minima and over fitting are unlikely to occur. Therefore, the LS-SVMs can overcome the shortcomings of neural networks in sensor dynamic modeling. The effectiveness and reliability of the method are demonstrated in two examples. The experimental results show that the method is still effective even if the sensor dynamic model is highly nonlinear.