The Study And Realization On The Knowledge Acquisition Base On BP Algorithm

Knowledge Acquisition is the procedure of acquiring computer-applicable knowledge from certain knowledge sources, which is one of the key problems in the domain of AI (Artificial Intelligence). As a general rule, knowledge acquisition is the most difficult work of constructing Expert System. It is the key and bottleneck of constructing Expert System. As it's difficult and time-consuming to manually acquire knowledge and build knowledge-bases, many theories and methods have been tried to automate this knowledge acquisition procedure, among which are ANN technologies. ANN is suitable for solving those problems that are hard to build model accurately and have high non-linear. It also can resolve the problems in the system such as connection complex, boundary faintness and strong uncertainty which can not be strict described by the rule or mathematical model. ANN is the process of automatically discovering new knowledge from certain data sources. It can effectively increase the automation degree of knowledge acquisition.This paper is funded by National Science-Tech Tackle Key Project——Research on theOrange planting Expert System. Based on the study of Expert System and ANN, this paper presented Hybrids system integrating artificial neural network to knowledge acquisition, designed and realized one knowledge acquisition system..At first, a three-layer BP neural network model and one flexible network structure are designed—the number of input nerve cell and output nerve cell can be chose freely, the network can be in a position to normal best function.Secondly, a kind of improved BP algorithm-MSBP is brought forward. It's different from other algorithms that it's amplified at two aspects of normal BP algorithm. On one hand, it brings up self-adaptation of learning rate and adjusts them automatically in order to accelerate the speed of convergence. On the other hand, the momentum item is added so as to avoid falling into the local minimum. It accelerates the speed of convergence and avoids falling into the local minimum. The algorithm is high effective. The simulated conclusion indicates that the improved algorithm achieved the anticipative purpose.On the basis of the previous work, the SD algorithm is applied to the rule extraction module, the exact production rules are extracted. The SD algorithm is implemented with great efficiency and speed in feed-forward network.The information from knowledge source can be transformed into the rule which the computer can identify by the knowledge acquisition system. Knowledge base can be constructed in order to provide data support for the further reasoning machine.