首先,针对径向基函数(RBF)神经网络参数学习中最小二乘法(LS)难以获得较高鲁棒性的问题,假定训练数据扰动上界可知,并基于鲁棒最小二乘原理,提出一种RBF网的最优鲁棒参数学习算法;然后分析指出,扰动上界可依据训练数据集自适应学习估计;最后通过实验分析结果表明了所提算法具有较高的参数鲁棒学习能力.与LS相似,新算法无额外参数,易于实际应用. First of all, it is hard to obtain the least robust (LS) problem in parameter learning of Radial Basis Function (RBF) neural network. It is assumed that the upper bound of training data perturbation is known and based on the principle of robust least squares Then the analysis shows that the disturbance upper bound can be adaptively learned based on the training data set. Finally, the experimental results show that the proposed algorithm has higher robust learning ability. LS is similar to the new algorithm without additional parameters, easy to practical application.