【摘要】 为了提高神经网络的预测精度,设计一种动态扩展BP网络隐层的方法,在训练过的BP网络上动态增加一个具有线性激活函数的隐层,用改进的蚁群算法对新增权值进行训练,着重对算法的实现过程及算法分析进行论述。设计了算法改进前后用BP网络对催化剂活性进行预测的对比实验,结果表明,采用该模型及训练算法,可在不影响网络表达能力的基础上提高网络的训练精度及网络的泛化能力。更多还原

【Abstract】 In order to enhance the forecast precision of the neural network,a method was designed to extend an assistant hidden-layer of BP neural network.The hidden-layer had a linear activation function.The weights were trained by animprovedantcolonyalgorithm.The implement and analysis of the algorithm were elaborated emphatically.An experiment was designed to contrast the performance of the BP network to predict the catalyst activity using original algorithm and the improved training algorithm.The results show the hidden-layer extending model and the training algorithm can improve the training precision and the generalization ability of the network without affecting the expression of the neural network.更多还原