关键词: 误差反传, 神经网络, 扩展隐层, 训练算法, 预测精度

Abstract: To enhance the forecast precision of the neural network, an improved BackPropagation (BP) training algorithm was proposed. The existing three-layer structure of BP neural network was extended. An assistant hidden-layer was dynamically increased in the trained-three-layer BP neural network to form a new neural network expansion model. The newly added weights and thresholds of BP neural network were trained by an improved ant colony algorithm. The implementation processes and analysis of the algorithm were elaborated in depth. An experiment was designed to compare the forecast precision and the training error of the BP network to predict the catalyst activity by its original algorithm and the improved training algorithm. Results showed that the hidden-Layer extension model and the improved training algorithm could improve the training precision and the forecast precision without affecting the expression of the neural network. On the other hand, it was also able to enhance the generalization ability of the network.

Key words: backpropagation, neural networks, extending hidden-layer, training algorithm, forecast precision