时变最优化信息处理技术及其应用（Ⅱ）：前馈神经网络学习算法

Time Dependent Optimization for Information Processing and Its Applications(Ⅱ) On Line Neural Network Learning

多层前馈神经网络应用成功的关键之一在于寻找一种有效的学习方法。尤其是在线应用，其学习和效率显著极为重要。而且前应用最广泛的ＢＰ算法却存在收敛慢和振荡等缺点。

The key to successful applications of multi layer feedforward neural networks is to find out an efficient learning algorithm. This is extremely important especially in on line applications. The disadvantages of the most widely used BP algorithm are slow convergence and oscillation. With time dependent optimization, on line feedforward neural network learning algorithm with exponential convergence rate is derived. Both theoretical analysis and simulations show that this algorithm has a fast convergence rate and enable the network to learn and adapt to new input output patterns quickly.