Research And Application Of Modular Neural Network Based On Cluster Analysis

A single artificial neural network has been widely used in various fields due to its strong learning ability and nonlinear approximation ability.However,a single artificial neural network often has the disadvantages of slow convergence and weak generalization ability when solving large-scale complex problems.To solve these problems,the researchers carried out a further imitation of the biological brain and introduced the concept of modularization into the design of the artificial neural network,from which the modular neural network was generated.Modular neural network adopts the idea of "divide and conquer" to decompose a complex task into two or more relatively simple subtasks and solve the whole task by processing these subtasks.The important idea of "divide and conquer" lies in "divide" and "conquer".The key to “divide and conquer” is how to decompose tasks into multiple subtasks and how to deal with these subtasks.This paper also builds a modular neural network based on cluster analysis according to these two problems:(1)This paper proposes a density peak clustering algorithm based on k-nearest neighbor optimization to decompose tasks.Clustering algorithm with good performance can better classify the subtasks and the peak density clustering has optimized the adjustment difficulty of traditional clustering algorithm based on density parameters,in which only one parameter can be achieved for the clustering of data.But when calculating the local density clustering algorithm,the peak parameters may cause unreasonable and sensitive problems of parameter selection.Aiming to solve this problem,this paper put forward the idea of using K nearest to redefine the local density calculation method and the experimental results prove that the presented clustering algorithm has higher precision.(2)When constructing modular neural network,k-nearest neighbor optimization density peak clustering algorithm is adopted for task decomposition.In terms of sub-network structure design,aiming at the characteristics of slow convergence speed and poor generalization ability of traditional BP neural network,genetic algorithm is introduced into the sub-network design,so as to improve theefficiency and accuracy of network learning.When designing sub network integrated strategy,the write puts forward the membership degree method according to the clustering center to calculate the new sample of dynamic membership degree of each network.This method is a kind of wide choices of sub network integration strategy.Due to the different new samples and different distance of every clustering center,the membership degree is a dynamic change,activating the largest network output.The model established in this paper is applied into the medical field to predict diseases.The iterative efficiency and accuracy of the model are verified by two medical data sets.The simulation experiment proves that the modular neural network designed in this paper is superior to the single artificial neural network in iterative efficiency and accuracy.