论文部分内容阅读
提出一种使用三层神经元网络的先验信息的新学习方法.通常,当神经元网络被用于函数逼近时,没有考虑权之间的关系而独立地学习,这样,学习的结果往往不理想.其原因是在学习中权是相互影响的.为了克服这一问题,首先,给出了一些有关权的先验信息,然后基于此提出部分权学习和其余权由精确数学方程计算的新学习方法.这方法在权的学习中几乎保持精确数学结构.另外,使用不等式先验信息的学习方法也被提出了.无论使用不等式还是等式先验信息的学习,因网络权的自由度被限制而加快了学习速度并保证误差较小.数值仿真的结果支持提出方法.
A new learning method using a priori information of a three-layer neural network is proposed.Generally, when the neural network is used for function approximation, learning independently without considering the relationship between weights, so the result of learning is often not The reason is that the weights of learning affect each other.In order to overcome this problem, firstly, some priori information about rights is given, and then a new partial-learning and the rest of the weights are calculated based on the exact mathematical equations Method of learning, which retains almost the exact mathematical structure in the study of rights.In addition, the learning method using inequality a priori information has also been proposed.With inequality or equality a priori information learning, the degree of freedom of network right Limit and speed up the learning speed and ensure the error is small.Numerical simulation results support the proposed method.