基于最小二乘法的BP算法

标准BP神经网络算法收敛速度慢是限制其广泛应用的主要原因．为此，以标准BP算法为基础，应用最小二乘法理论，提出了一种收敛速度快的BP算法——NLMsBP算法．仿真结果表明，和标准BP算法及其它改进形式比较，NLMSBP算法收敛速度大大提高，稳定性并未降低，这为BP神经网络应用于实时性要求高的场合提供了算法基础．该算法缺点是计算量大，所需计算机内存大，不适于大型网络的计算．

A BP algorithm for training neural networks based on solutions for a nonlinear least mean square problem

That standard backpropagation(BP) algorithm for training neural networks converges slowly is the main reason why it cannot be used widely in practical applications. Therefore, a new kind of BP algorithm, called the NLMSBP algorithm for short, is put forward in this paper by using solutions for a nonlinear least mean square problem. The experimental results have proved that the algorithm converges very fast and has good stability compared with the standard BP algorithm and the other modifications. It is suitable for training the network with a few thousands of weights and offsets and high training precision demand. If the computer memory is enough, the superiority of the algorithm over the others is very notable. Indeed, it is worth popularizing.