Research On Intelligent Decision-Making Methods Of Tacit Objective Decision-Making Problem Based On IEC

In the management and decision-making field, there exists a class of special decision-making problems which are called Tacit Objective Decision-Making Problems (TODMP), such as the fashion design problem, the car styling design problem and the travel itinerary planning problem. It is very complex to solve the TODMP because it has three special characters that the decision objectives are unable or difficult to be defined explicitly in a structured or quantitative way, the preference of the decision-maker can be changed during the decision-making process, and the solution space is very large and it has a huge number of feasible solutions. It's impossible to sovle this class of problems in the direct way of exhaustive comparison of all decision solutions, and also impossible to search the the optimal solution or satisfactory solution directly in some optimization searching ways. It needs the intelligent decision-making methods with an interactive mechanism to sovle the TODMP. TODMP is a class of complex decision-making problems, and these problems are pervasive in our real life, so it's important to study decision-making methods of solving the TODMP in both theroy and practice.In recent years, with the development of intelligent optimization algorithms and artificial intelligence technologies, the interactive evolutionary computation (EEC) with human-computer interaction mechanisms has become a strong advantageous method of sovling the optimization problems with uncertain objectives, especially the interactive genetic algorithm (IGA) in the field of interactive evolutionary computation. The IEC/IGA-based methods which are suitable for sovling the TODMP become the foundation of the decision-making methods of the TODMP. From the point of view of the TODMP, this dissertation analyses the difficulties of solving this class of problems, and then studies some EEC-based intelligent decision-making methods of the TODMP on the basis of IEC/IGA, in combination with other intelligent computation and artificial intelligence techniques. These studied methods aim to solve several problems existing in the TODMP, which are low solution efficiency, decision-maker's fatigue, solution individual diversity reducing, and so on. In the dissertation, the car styling draft design problem and the travel intinerary planning problem are introduced as two applications of the TODMP, in order to verify the accuracy and effectiveness of the proposed intelligent decision-making methods. The main achievements of this dissertation are as follows:(1) Intelligent methods combined with multi-agent system technology are studied and an interactive multi-agent genetic algorithm is proposed, which is used to sovle the problems of the low solution efficiency of the TOMDP and the decision-maker's fatigue in the interactive decision-making process. In the proposed algorithm, the idea of agent-based computing is applied to the domain of IEC, and the concepts of Agent, Agents' environment and Agent's behaviors in the environment are defined. The behaviors of Agent are evolution, competition and self-learning. In addition, an effective fitness strategy is designed in the proposed interactive multi-agent genetic algorithm, so it's not necessary for the decision-maker to evaluate each solution individual in the interactive decision-making process. The decision-maker only needs to select the most satisfactory solution individual in each generation, and does not need to give a specific value of its fitness, which reduces the fatigue of decision-maker. The result of the travel itinerary planning experiments verifies the effectiveness of the proposed method.(2) The problem of solution ndividual diversity reducing in the decision-making process of the TOMDP is studied, and a method based on improved sharing scheme to maintain population diversity is proposed. Because there is a drawback that it's difficult to set a correct value to the niche radius in traditional sharing method, in the proposed method, the niche radius is first designed to be calculated automatically according to the average distance of all individual in population. And then the concept of niche entropy is introduced to measure the population's diversity. The evolutionary parameters of the crossover probability and the mutation probability are calculated automatically on the basis of the evolutionary generation number and the niche entropy in that generation, which can control the genetic operation in the evolutionary process effectively. The results of the benchmark function experiments and the ca(?) styling draft design experiments verify the effectiveness of the proposed method of maintaining diversity and the algorithm with it. The proposed method of maintaining diversity can effectively avoid falling into the local optima early, and it's more useful for the decision-maker to determine his/her own preference gradually in the interactive decision-making process, with no increase of evaluation fatigue.(3) A special adaptive diversity strategy based on an improved niche identification method is proposed to settle the conflict between maintaining diversity and quick convergence. Because of the inaccurate niche identification in traditional niche methods, an improved approach to identify niches is designed firstly, and then on the basis of the identified niches, the calculation of the niche entropy is improved, which is more suitable to measure the population's diversity. In the proposed strategy, the evolutionary parameters are adjusted adaptively by the calculated value of the niche entropy, which establishes an adaptive mechanism of "population diversity→evolutionary parameters→population diversity". The results of the benchmark function experiments and the car styling draft design experiments verify the effectiveness of the proposed method and strategy which can effectively settle the conflict between maintaining diversity and quick convergence.(4) The car styling draft des(?)gn problem - a specific example of the TODMP is studied. In this part of the dissertation, the car styling draft design based on modeling parts and the car style coding are introduced, and an IGA-based aided system prototype for the car styling draft design is constructed and realized.