关键词: 神经网络, 权值直接确定, 反插值, 时序

Abstract:

By using numerical algorithms to solve inverse-interpolation problems, the accuracy of positive solution is influenced by the choice of initial values, and the computational speed is slow. In order to solve the problems, a power-excitation feedforward neural network is employed to solve inverse-interpolation problems. As the adopted neural network is only suitable for the inverse-interpolation problem with one-to-one mapping but not for that with multiple-to-one mapping, the timing control condition is introduced into the neural network to construct a timing power-excitation feedforward neural network model. Theoretically derived and simulated results indicate that the constructed neural network effectively solve the inverse-interpolation problems with both one-to-one and multiple-to- one mappings.

Key words: neural network, direct weight determination, inverse interpolation, time sequence