论文部分内容阅读
针对最佳平方逼近3层前馈神经网络模型,采用子空间分析方法,讨论了隐单元的误差补偿性能,提出了隐层评测方法.研究结果表明隐单元选取策略应遵循其输出向量有效分量位于误差空间、回避耗损空间和尽可能靠近某一能量空间的原则,这一结果与隐单元采用什么激发函数无关,也允许各隐单元采用不同激发函数.网络的隐层性能评价可以通过隐层品质因子、隐层有效系数、隐单元剩余度来进行,而总体结果可采用隐层评价因子进行评测.评测实验表明,所提出的隐层评测方法是合理有效的.图1,表1,参11.
Aiming at the best square approximation 3-layer feedforward neural network model, the sub-space analysis method is used to discuss the error compensation performance of the hidden element, and a hidden layer evaluation method is proposed.The research results show that the hidden element selection strategy should follow its output vector effective component Error space, avoidance of loss space and the principle of as close as possible to a certain energy space, this result has nothing to do with the excitation function of the hidden unit, but also allows each hidden unit to adopt different excitation functions.The hidden layer performance of the network can be evaluated by hidden layer quality Factor, the effective coefficient of hidden layer, the residual unit degree of hidden unit, and the overall result can be evaluated by using hidden layer evaluation factors. The evaluation experiments show that the proposed hidden layer evaluation method is reasonable and effective. .