鲁棒控制器设计，由于RBF网络可以实现任意逼近的非线性关系，它的目标是要做到误差平方和最小，与非线性PCA的目标一致，所以上述非线性PCA的模型可以通过采用两个RBF网络来实现非线性正变换 和反变换 。RBF网络是一个三层前馈网络，隐层采用径向基函数作为激励函数。第一个RBF网络把高维空间的数据映射到低维空间（如图4），第二个RBF网络将前面网络输出的低维空间数据再映射到高维空间，实现数据恢复(如图5)。这两个网络分别进行训练。-robust controller design, as RBF networks can achieve arbitrary nonlinear approximation, Its goal is to achieve the minimum squared error, and nonlinear PCA have the same goal So these nonlinear PCA model may be adopted by two RBF networks to achieve nonlinear transformation and inverse transform. RBF network is a feed-forward network, hidden layer RBF function as an incentive. RBF a network of high-dimensional data mapping space to the low-dimensional space (figure 4), second RBF network will be in front of the output of low-dimensional space mapping data again to a high-dimensional space. data Recovery (figure 5). The two networks separately for training.