Abstract: By transforming the functional neuron,we proposed a functional network learning algorithm with recursively base functions. The algorithm uses a recursive method for solving matrix's pseudo-inverse to achieve adaptive adjustment of base functions in functional neural network, finally realizes the functional network structure and parameters of the optimal solution together. The experimental results show that the learning algorithm has adaptive, robustness and high accuracy of convergence, and will have broad application in real-time online identification.

Key words: Functional neuron, Base functions, Matrix's pseudo inverse, Learning algorithm