样条神经网络的权值直接确定法

Weights direct determination of a spline neural network

根据样条逼近理论和神经网络原理构造了一种样条神经网络模型,以一组样条基函数作为隐神经元的激励函数。依据误差回传(BP)思想推导出该网络模型的权值修正迭代公式,利用该公式迭代训练可得到该网络的最优权值。而对于构造的具有特定网络结构的样条神经网络,依据伪逆思想提出了一种直接计算权值的方法,从而避免冗长的迭代训练过程。仿真结果表明该权值直接确定法不仅能一步确定权值从而获得更快的运算速度,而且能达到更高的计算精度。

Based on spline approximation theory and neural networks, a spline neural network is constructed, where the hidden layer neurons are activated by a group of spline functions. Based on the standard error back-propagation （BP） method, the neural-weights updating formula is derived. More importantly, a pseudo-inverse based method is then proposed, which could directly determine the network weights without iter- ative training. Computer simulation results show that the one-step weights-determination method could be more effective than the standard BP iterative-training method.