Research and Applications of Self-Organization Process Neural Networks

Aimed at the pattern classification problems relating to time process, a neural network model named self-organization process neural network is brought forward in this paper. The network consists of input layer and competition layer, and its input and link weights are functions relating to time. The nodes of input layer and competition layer link with each other totally. The network extracts the implicit pattern characters of input function and self-organizes them, and then represents the classification result at competition layer. In order to simplify computing, function orthogonal base is introduced into input space, and the input and weight functions are represented as the expansion form of orthogonal base. Using the orthogonality of base function, the adjustment of network weight coefficients can be made independent of time. The learning algorithms of competition learning and teacher demonstration are given. The effectiveness of the model and algorithms is proved by sedimentary faces identification in petroleum geologic.