Research On Granular Computing Based Multi-attribute Group Decision Making Methods

As one of the most important research fields in decision analysis, multi-attribute group decision making widely exists in human's daily life. Decision data in multi-attribute group decision making always include uncertainty, fuzziness, multiple views and multiple levels information. Facing with the complicated decision data, decision makers are not satis-fied with the functional and relation models that need strict assumptions and more subjective experiences, but instead the methods which analyze and evaluate the practical decision data, and acquire the effective deci-sion information for supporting the complicated decision. These kinds of data-driven decision methods are the new trend of the modern deci-sion analysis. Granular computing is a natural pattern that simulates human's thinking and problem solving procession, and it is also a new paradigm for analysing data and solving problems based on granules and the relations among them. Granular computing can provide the satisfying solution by multi-granularity, multi-view, and multi-level description and reasoning. This thesis introduces the mechanism of granular computing into the solution of multi-attribute group decision making, and investi-gates the solving strategies for multi-attribute group decision making from granularity analysis of decision elements, multi-granularity decision infor-mation expression, multi-level solving models. The main contributions are summarized as follows:(1) Analyzing the granularities of decision element and constructing the preference granular structure of the decision alternative set provide the necessary expression basis for solving multi-attribute group decision making from multiple granularities and multiple levels. Firstly, the gran-ularity levels of the three decision-making tasks (ranking, sorting and choice) are interpreted through the decision results. Secondly, the gran-ularity change mechanisation and solving transformation of decision in- formation are revealed from the delivered uncertain degree. Thirdly, the preference granular structures for the alternative set are constructed un-der the different decision expression formats. The last, the granularity order for the classical aggregation operators is given from the perspective of the data relevance.(2) To overcome the different sematic scaling and the complex trans-formation procedure in the computation of multi-granularity linguistic, a normalized scaling method is proposed for multi-granularity linguistic based on the priori scaling information. Compared with the other scaling methods, the proposed method can obtain the normalized scaling at the generation stage, which avoids the multifarious transformation procession. At the same time, it constructs a platform for semantic commutations of linguistic terms with different granularities.(3) By introducing the intuitionistic rough granular judgments into the decision process, a multi-attribute group decision making method is proposed based on preference granular structures. The consistence weight-s of experts are determined by depicting similarity degrees between the corresponding preference granular structures under each attribute. And the weights of attributes are set according to the decision data, rough granular sorting judgements and the subjective preference of decision makers. The proposed method can obtain the effective decision result-s by integrating the thought of multi-granularity problem-solving into the decision model, fully considering the intuitionistic rough granular judg-ments, and objectively acquiring the more interpretable decision index from the decision data.(4) By introducing the experts'preference of alternative pairs, a hy-brid multi-attribute group decision making method is proposed based on fuzzy information granules. We first generalize the similarity degree of ac-curate preference granular structures under the fuzzy circumstance, and then use it to determine the weights of experts, which are irrelative the types of decision data. Similarly, the weights of attributes are deter- mined by combining the heterogeneous decision data, expert preference of alternative pairs and the decision makers'preference. The model can conveniently calculate the decision results that carries abundant decision information and easily apply in real decision problems.(5) On the basis of human's reference decision principle, a hybrid multi-attribute group sorting method is proposed based on preference granules evidence fusion. Not need the complex and intricate parameters and thresholds, the method provides the sorting of objects completely driven by data. It compares an object with the inner and outer objects with respect to a decision granules and the generates the sorting evidences for it. These sorting evidences are fused from different granularities and multiple levels to make the final objective and credible sorting for the objects.This thesis proposes a series of methods for solving the multi-attribute group decision making problems from multiple granualrities, multiple views and levels, and it also provides a new perspective for solving the multi-attribute group decision making problems. This study enriches the theories and methods of multi-attribute group decision making.