Design And Application Of A Neural Network Based On Many-Valued Logic

Artificial neural network is a mathematical model that can partially simulate the functions of animals'and process information distributed and parallel. After decades of development, artificial neural network has been widely used in the area of production and life.First, this paper introduces the widely used artificial neural networks, BP neural network and SOM neural network, and their application in practice. Second, the concept of many-valued logic and DNA computing is introduced, in combination with the disadvantage of DNA computing. Third, with the idea of DNA computing, a neural network model based on many-valued logic is proposed. This kind of network model can be implemented by the circuit, which can smooth over the fault of DNA computing.There are three layers in the model of many-valued neural network, input layer, middle layer and output layer. Two sub-neurons, which make it possible to input data to the network in the form of many-valued logic, included in an input layer neuron. The neurons of middle layer have storage unit and can be used to calculate. The input layer and middle layer being connected in a special way. The process of network input is very quickly cooperating with the many-valued logic. After the calculating of the middle layer, the solutions are out. In the stage of getting solutions, the method of binary-out solution is being used. The binary-out method can improve the efficiency distinctly.Then, this paper introduces the process of solving problems by many-valued neural network. Two NP-Complete problems, SAT and SPLIT, are being solved by the network. The network is implemented by C language on PC. This paper mainly discusses follows:(1) A neural network model based on many-valued logic is proposed combined with the idea of DNA computing many-valued and neural network.(2)With the multi-valued neural network and specific problems, an effective parallel algorithm is designed.(3)Simulate the process of problem solving (20 variables) on PC to verify the validity of the method.