Research On Decision Neural Network Model With Applications

Artificial Neural Networks serve as a intelligently computational model for signal processing which can simulate the partial function and mechanism for biological Neural Networks. The researches about the Artificial Neural Networks mainly emphasis on two fields, including theory and hardware. Although recently electronic techniques are used to realize the hardware, it seems that human being don't find out true materials all the time which can succeed in the simulation of biological Neural Networks. In addition, the solver of optimization computation in theory is dependent on the all-connected Hopfield model. However, like the molecular Neural Networks, the partial connected-networks are to a large extent applied to the field such as image processing and so on. Consequently, it is a core subject about how to construct Artificial Neural Networks that are of local-connection, of certain practice, and also can be applied to solver the optimized problems.Therefore, this paper is aimed to establish a kind of Neural Networks model that are of local-connection, of simulation human's decision-making thinking, and also can be applied to solver the optimization for information. Thus we give detail discussions about the multi-stage decision problems by use of Graph Theory before establishment of models. Then we can continue to discuss the premature convergence of the network and even the optimized computation of the engineering technology. And then the Neural Networks model will be used to solutions of the TSP and Graph isomorphic problems and so on. Additionally, we also discuss how to apply the Artificial Neural Networks to vertex covering. There are mainly four important results in this paper.Firstly, we construct the graphic model for the multi-stage decision problems and discuss some basic theory and application problems, for instance, it describes how to directly or indirectly change the realistic problems into multi-stage decision problems. And it shows a standard directed graphic method, and the counting formula of multi-stage decision problems. Besides these two computation method of solving the set to decision-making are given. This paper establishes the model of the most shortest problem based on multi-stage decision-making problem of theory method and constructs the mathematic model of TSP problem based on multi-phase decision-making problem of theory method, that can be directly applied to the solving the Hamiltonian problem.Secondly, we establish the model of decision-making Neural Networks. This is a local area connection networks. Its merits is neither similar to the connected Hopfield Neural Networks, nor the "rigid" of cell neural networks, is close to the human thinking method in local area connection problem. The characteristics of the model are analogy to the modedecision-making of human, maybe we can't get the optimal solution, but are easy to obtain the satisfactory solution. This chapter gives the mechanism of the model, the framework of networks, and the circuit realization and so on.Thirdly, the chapter systematically discusses the phenomenon of premature convergence and applied to the theory of permutation group, graph theory and other mathematics tools to make efficient researches, which can be directly used to many optimization computation.Finally, we apply the model of decision-making Neural Networks to the research of traveling salesman problem, isomorphism of graph. The basic idea is to put the energy function to the decision-making, and then combine the equation of the networks. At last, establish the model of neural networks based on vertex coving. Based on the existed Hopfield neural networks, we combine the idea of decision-making neural networks to modify the model.