以标准BP算法为基础,应用Levenberg Marquardt最优化方法,提出了一种快速收敛的BP算法———LMBP算法。经实验验证并与标准BP算法及其它改进形式比较,LMBP算法大大提高了收敛速度,而且性能稳定。这为BP神经网络应用于实时性要求高的场合(如在线检测)提供了算法基础。该算法的缺点是计算量大,所需计算机内存大,不适合大型网络的计算。 Based on the standard BP algorithm, a Levenberg-Marquardt optimization method is proposed, which is a fast convergent BP algorithm --- LMBP algorithm. Experimental verification and comparison with the standard BP algorithm and other improved forms, LMBP algorithm greatly improves the convergence rate, and the performance is stable. This provides an algorithmic basis for the application of BP neural network in demanding real-time applications such as online inspection. The disadvantage of this algorithm is the large amount of computation and the large memory required for the computer, which is not suitable for the calculation of large networks.