Research On Complex Multi-attribute Group Decision Making Based On Neural Network

The multi-attribute group decision making mainly focuses on evaluating and ranking the alternatives against the attributes by a group of experts.For the complex decision-making problems,because of constraints about the experts' educational background,cultural knowledge,experience,and expression,it is often difficult to evaluate the alternatives accurately and quantitatively,and the experts often intend to employ interval numbers or fuzzy languages to express their evaluation information against some qualitative attributes.For the multi-attribute group decision making problems with different types of attribute value evaluation information,the experts often give overall evaluations on the alternatives,for example,the preference orderings.This thesis focuses on determining the attribute weights and expert weights so that the overall values of the alternatives and their rankings are obtained,based on the different types of attribute values and the experts' preference orderings on the alternatives.The preference orderings of experts on the alternatives are taken as the expected outputs,and the neural networks are used to obtain attribute weights to solve the complex multiple attribute group decision problem.It also extends the research to the case where there is no preference information on the alternatives.Firstly,for different types of attribute value evaluation information,this thesis normalizes them into the form of single point value.The fuzzy complementarity relation matrix is used to normalize the preference orderings.The ideal method is used to normalize the attribute values of interval numbers.The area mean method of triangular fuzzy numbers is used to normalize the language evaluation values.Grey relational degree method is used to normalize the attribute values of uncertain linguistic variables.Secondly,the thesis proposes mathematical models for the multi-attribute group decision makings problems with the overall preference ordering information on the alternatives given by experts.The linear neural networks are adopted to train the attribute weights and expert weights,where the experts' preference orderings are used as the expected output of the neural network,and the attribute values of the alternatives are taken as the inputs.The overall values of the alternatives are obtained based on the weighted summation method.An example is given for illustration and simulation is conducted based on Matlab to verify the proposed mathematical models and neural network models.Finally,the thesis investigates the multi-attribute group decision making problems where the experts do not give the overall preference information on the alternatives.The expected output of the neural network is constructed by using analytic hierarchy process and the maximizing deviation method,then the attribute weights and expert weights are trained by BP neural network.The overall values of the alternatives and their rankings are obtained by means of the weighted sum method.An example is also given to justify the proposed neural network by Matlab simulation.