论文部分内容阅读
针对传统BP神经网络训练收敛速度慢、易陷入局部极小点的问题,将遗传算法与误差放大的BP学习算法相结合,提出基于切片模型的快速混合学习算法.该算法通过将传统神经网络的训练过程划分为许多小的训练切片,并利用遗传算法的并行寻优特性,对采用误差放大的BP训练过程进行监督.通过及时发现收敛速率较快的个体和过滤陷入局部极小点的个体,来保证网络训练的成功率和实现快速向全局最优区域逼近的目的.仿真实验表明,该算法在不增加网络隐层节点数的情况下,显著地提高了网络的收敛精度和泛化能力.
Aiming at the problem that the traditional BP neural network training converges slowly and easily falls into local minimum, a fast hybrid learning algorithm based on the slice model is proposed by combining genetic algorithm with BP learning algorithm with error amplification.The algorithm combines traditional neural network The training process is divided into many small training slices, and the parallel optimization of genetic algorithm is used to supervise the BP training process using error amplification.By finding the individuals with fast convergence rate and filtering the individuals who fall into local minima, So as to ensure the success rate of network training and achieve the goal of rapidly approaching the global optimal region.The simulation results show that the proposed algorithm can significantly improve the convergence accuracy and generalization ability of the network without increasing the number of hidden layer nodes in the network.