Research And Discussion Of Artificial Neural Network Based On Non-parametric Regression Method

This paper proposes new methods to the estimation of partial variable co-efficient linear model and non-linear perceptron.It provides further theoretical basis for the application of variable coefficient model and Neural Network by combining the two method with each other.The first chapter is an introduction,which is a retrospect to the background knowledge of Artificial Neural Network and partial variable coefficient linear model, gives details of basic model, learning objectives and learning algorithm for the fuzzy perceptron and non-linear weighs perceptron, and points the advantage and disadvantage of the models applied to the reality.Chapter two presents the perceptron estimated method to the partial vari-able coefficient linear model. First,it proposes perceptron estimated method to multi-linear regression model, with which to solve the estimation problem of non-parametric regression using fuzzy perceptron model and its algorithm. And then,I compared the simulation result of the partial variable coefficient linear model estimation method based on perceptron with current local linear estima-tion method.Chapter three aims at non-linear weights perceptron based on local linear estimation method.For the research of previous author, there is no specific expressions to the non-linear weights perceptron. Therefore, it can't be applied to the reality. This chapter proposes the estimation method of non-linear weights perceptron using local linear method for providing further theoretical basis to the Neural Network perceptron, and discovers that there is a wider application to the non-linear perceptron compared to the traditional perceptron by stimulating.Chapter four emphasizes on generalization of non-linear weight perceptron.