Research And Design On Structure Artifical Neural Network

As one of the most important data modeling method, Artifical Neural Network is widely used in all kinds of areas. In the development of Artifical Neural Network, single mode has lots of problem such as poor generalization and unexplainable structure, all of them can not guarantee the user's requirements. Therefor, other kinds of neural network have been designed.Basd on the intelligent theory research, this subject contains three parts. One part is analysis of data space, the second part is about design of structure neural network, and the last part is about performance optimization. Meanwhile, in order to settle the problem of black box and reduce the dependence of expert system, a learning plantform of structure neural network is developed. The key part of this plantform is the application and trail about the new type of structure neural network.Through using the standard datasets from UCI and PTA, HDPE from real chemical factory to test, Tree typed Hierarcail Neural Network and Assocaite typed Hierarchical Neural Netowrk showed the effectiveness and priorty of explaincation. This steady performance makes this technology valuable and meaningful in the reality industry product processing.