Research On Interactive Determining Weights Based On BP Neural Network And Its Application

There are many multiple attribute decision-making (MADM) problems in the fields of society, economy and management, such as investment decision-making, project assessment, quality assessment, alternative selection, ranking of industrial departments and comprehensive evaluation of economic benefit, etc. Therefore, the theories and methods of multi-attribute decision-making show a prospect of wide range of applications. MADM mainly consists of the following two parts: (1) Collect decision information. The decision information generally includes the attribute weights and the attribute values. (2) Aggregate the decision information through some proper approaches. At present, two of the most common aggregation approaches are the additive weighted averaging (AWA) operator and the ordered weighted averaging (OWA) operator. This paper studies the above the two sides, how to decide the weights is mainly studied, and this paper is composed of the following six chapters:Chapter 1, some concepts and character on MADM problems are introduced. It makes a survey of the current study on the theory of MADM, and summarizes the main contents and structure.Chapter 2, attribute sets classification of MADM are introduced. In this chapter, the author gives the different methods of formalizing decision-making matrix, especially puts forward the method named transform formula based on exponential function to formalize decision-making matrix.Chapter 3, considering the defects of the present subjective and objective weighting methods, the author proposes a new kind of combination weighting method, which is objective and subjective synthetic approach to determine weights based on ideal-solution for multi-attribute decision making.Chapter 4, the author first presents an ordered weighted Euclid averaging (OWEA) operator, and then studies the relationship between the OWEA operator and the OWA operator in detail. A method based on the OWEA operator is proposed for the MADM problems, and an illustrative example is also given.Chapter 5, the principle and arithmetic of back propagation (BP) network are introduced firstly, and the interactive method of deciding the weights based on BP network is proposed. The method decides the weights using subjective weighting method firstly, and then adjust the weights by means of maximizing deviations method, then the results which got from OWEA operator are used as the learning samples to train the BP neural network. At last, the method is applied to evaluate the enterprise technology innovation capacity of Jiang Su province.Chapter 6 concludes the paper. The chapter points out the main innovative findings, and then discusses the future research.