基于正交校正共轭梯度法的快速神经网络学习算法研究

前馈神经网络由于具有理论上逼近任意非线性连续映射的能力,因而非常适合于非线性系统建模及构成自适应控制。为了提高前馈神经网络的权的学习效率及稳定性,该文提出一种基于正交校正共轭梯度优化方法的快速神经网络学习算法,通过与其它学习算法(如:BP算法、变尺度法、用差商近似代替导数的Powell法等)的比较,经仿真试验表明,本算法是一种高效、快速的学习算法。

Study of fast learning algorithm for neural networks base on CGM-OC approach

Because the feedforward neural network has an ability of approach to arbitrary nonlinear mapping, it can be used effectively in the modeling and controlling of nonlinear system. In order to improve the learning efficiency and stability of feedforward neural network, a fast learning algorithm for neural networks base on CGM-OC approach is presented. Compared with other learning methods such as BP method, Broyden Flecher Goldfarl Shanno method, Powell method etc., simulation results show that the proposed method is an efficient and fast methed.