The Research And Application On Neural Network

The neural network as a kind of emerging information processing science, which can abstract and simulate some basic characteristic of the human brain. It is an information process method that a study auto-adapted, the non- procedural, takes person's cerebrum working pattern as a foundation. The characteristic of this kind work mechanism displays it own processing function in the massive neurons function through the network, from simulating the human brain structure and the single neuron function, achieved the goal that simulates the human brain process informationAt present, in the national economy and modernization of national defense science and technology, the neural network has the broad application domain and the application prospect, its mainly applied in information, automated, project, economical and so on, without a doubt, although it has widely applied domain, simultaneously also has many insufficiencies, thus becomes the hot topic which now the people continuously studies.Nowadays, the BP neural network applies universally, it has the basic character of the neural network, and therefore this article takes the BP neural network as the research object. The research content mainly contain, first, detailed elaborates the relation between the biological neuron and artificial neuron and the basic principle and inferential process of BP neural network, proposes existence questions of BP neural network and to analyze the reason. Next, conducts the research to structural design, parameter choice, sample choice and processing, initial power value and transfer function choice of the BP neural network, especially the momentum coefficient choice in the momentum item of BP neural network, in the network training speed influence of the different transfer function, to structure determination of the BP neural network. Once more, analyzes the standard BP nerve network problems which are slow convergence rate and easy to vibration, the main reason is because gradient drop method itself creates, therefore has made the improvement to the gradient drop method, then proposes the improvement BP neural network, proves the improvement BP neural network obviously surpassed the standard BP neural network whether in the training speed or training precision through computer simulation. Finally, applying the BP neural network to arm machinery gross power to forecast in Heilongjiang Province of our country, which has obtained the good effect, has provided the certain theory instruction for development agricultural mechanization of the Heilongjiang Province.Certainly, because author's knowledge is limited, the research to the BP neural network is insufficiently comprehensive, has the unavoidable deficiency, need to have the further research to BP neural network theory and the application of he BP neural network.