关键词: 幂级数, 前向神经网络, 隐神经元数, 结构最优化, 权值直接确定法

Abstract: For a kind of feed-forward neural net with hidden-neurons’ activation-functions being a sequence of power functions，an optimal network-structure determination algorithm is proposed based on the weights-direct-determination method.Computer simulation and verification results indicate that the algorithm can determine the optimal number of hidden-layer neurons automatically，quickly and effectively，which achieves the best approximation ability of the neural net and thus realizes the network-structure optimization.

Key words: power series, feed-forward neural net, hidden-layer neurons, structure optimization, weights direct determination