提出最近邻Levenberg—Marquardt误差反向传播神经网络算法。针对BP神经网络收敛速度慢的不足,利用Levenberg—Marquardt优化算法进行改进。同时为了提高神经网络的泛化能力,进一步基于最近邻算法对样本进行修剪。试验表明,与一般神经网络算法相比,NN-LMBP在改善神经网络泛化能力的基础上,有效地提高了神经网络收敛的速度。
An algorithm of Nearest Neighbor Levenberg-Marquardt Back Propagation Neural Networks (NN-LMBP) was put forward. The optimization algorithm Levenberg-Marquardt was utilized to increase the convergence speed of BP Neural Networks. Besides, based on algorithm of Nearest Neighbor, a strategy of sample pruning was adopted to improve the generation performance of Neural Networks. Experiments show that compared with normal Neural Networks, NN-LMBP is better in speed and generalization ability.