本文根据统计学的稳健性原理,将柯西(Cauchy)函数作为新的神经网络目标函数。在网络参数相同的前提下,利用传统的均方目标函数和新的柯西目标函数对BP网络分别进行训练后,加入小噪声及异常值(Outliers)干扰对该网络进行测试。结果表明,具有稳健性目标函数的网络不但有更快的收敛速度,而且对异常值有更好的抵抗能力。 Based on the principle of statistical robustness, this paper uses the Cauchy function as a new objective function of neural network. On the premise of the same network parameters, the traditional mean square objective function and the new Cauchy objective function are used to train the BP network respectively, then the network is tested by adding small noise and outliers interference. The results show that the network with robust objective function not only has faster convergence speed, but also has better resistance to outliers.