A Structural Formula Process Neural Networks and Its Applications

Aimed at the pattern classification and the system-modelling problem with complex time-varying signals that have singular values, a kind of structural formula process neural networks is proposed in this paper. This model is constructed based on the rational expression function's approximation character to complex process signals and the process neural networks' nonlinear transforming mechanism to time-varying information; its basic information processing unit is made up of two process neurons, logically constructing a structural process neuron. It is a type of extending for artificial neural networks at the architecture and the information processing mechanism. In this paper the continuity and functional approximation ability of structural formula process neural networks are analyzed, and the learning algorithm based on the function orthogonal expanded by bases is given. The experimental results demonstrate that the learning and the generalization character of structural formula process neural networks for the time-varying function samples with singular values are better than that of BP networks and the general process neural networks. The numbers of hidden layer and node numbers are greatly decreased, and the learning characters of the algorithm are the same with BP algorithm's.