Research On Interactive Genetic Algorithms And Its Application In Decision-Making Problems With Tacit Objectives

Decision-making problems with tacit objectives (DMPTO) are a class of complex optimization problems which widely exist in various aspects of the real life. Normally, the decision objectives of DMPTO are unable or difficult to be expressed in an explicitly structured way, such as fashion design problems, travel itinerary planning problems and automobile modeling design problems, et. al, therefore, the decision-making methods for DMPTO are worth studying both in theory and practice. In addition to having vaguely unstructured decision objectives, DMPTO also has two distinguishing features that decision-makers’preferences dynamically adjust with the interactive decision-making process, and the decision space is large and feasible solutions are numerous. In consequence, DMPTO is impossible to be directly resolved by means of traditional optimization decision methods, which need decision-making models with interactive mechanism to support the problem solving process. Interactive genetic algorithm (IGA) is a kind of intelligent decision methods with the human-computer interactive decision-making mechanism, in which the individual fitness values are given by decision-makers’subjective evaluations, and the traditional genetic algorithm (GA) is combined with interactive evolutionary idea, so that the person’s subjective initiative and intelligent ability in the field of information treatment are fully functional. As a consequence, IGA is suitable for solving DMPTO, and can be the foundation of the problem-solving methods. However, there exist several defects in the traditional IGA, such as the low convergence efficiency, user fatigue, population diversity reducing, et. al.. All of the above problems limit the further application and popularization of the algorithm to a great extent.On account of the analysis of the domestic and foreign research summary and elementary operation of IGA and GA, the problem solving model based on IGA is established for DMPTO. Aiming at settling the deficiencies of the existing study, this dissertation executes the pervasive research on IGA from the following three aspects: the improvement research on IGA combined with user preferences, the improvement research on IGA combined with niche techniques of sharing mechanism, and the improvement research on Multi-user IGA. In the dissertation, several improved IGA methods for DMPTO are proposed, and the intelligent decision support system facilitating the algorithm operation is built by introducing the automobile face modeling design problem as the application example of DMPTO, in order to verify the effectiveness of the proposed algorithms. The main achievements of the dissertation are as follows:(1) An interactive genetic algorithm based on user preferences model (UPM-IGA) is presented, which is used for solving the limitation of the user fatigue of IGA. Through a combination of the whole and part evaluation of evolutionary individuals, the user cognitive rule is explored and the user preference model is structured. According to the evaluations of users’partial individuals acquired from the algorithm interactive process, the fitness of each part of evolutionary individuals is extracted. In consideration of the impact on the user overall preferences brought from the weight of each part and the group between parts, the estimation formula of the individual fitness values is designed, which can directly execute the machine computation instead of the artificial evaluation when users feel fatigue. Simulation experimental results from AFMDS demonstrate that UPM-IGA can significantly reduce the number of iterations and evaluations, in consequence, the user fatigue can be effectively relieved.(2) An interactive genetic algorithm based on local user preferences (LUP-IGA) is proposed, which is used for solving the limitation of the low convergence efficiency of IGA. In the presented algorithm, the implementation of user local preferences’ fixing/removing operations are first discussed and then the improved crossover operator and mutation operator are described. Next, the process and performance of LUP-IGA are analyzed. At last, LUP-IGA is applied in the automobile face modeling design system (AFMDS), and the simulation experiment results show that LUP-IGA can effectively accelerate the convergence process and improve the success rate of finding out the optimal modeling individual.(3) Aiming at the population diversity maintaining problem, the niche techniques of Sharing Mechanism are overviewed, and the niche entropy is introduced to measure the population diversity. On the basis of the above contents and techniques, a sharing-based niche interactive genetic algorithm (Sharing_NIGA) is Constructed. In Sharing_NIGA, the adaptive selection strategy of the niche radius is designed, as well as the estimation method of the fitness values and the adaptive calculation scheme of the crossover probability and the mutation probability. User only needs to select the most satisfactory individual of the current generation, and does not need to give a specific fitness value of the most satisfactory individual, by means of the above strategy the user fatigue brought by the evaluation process can be effectively reduced. Simulation experimental results from AFMDS indicate Sharing_NIGA can effectively maintain efficient operation performance and sustain the population diversity.(4) On account of the multi-user group decision-making problem with tacit objectives (GDMPTO), the intelligent decision-making process is described, and a novel multi-user interactive genetic algorithm based on single population co-evolution (SPCE-NMUIGA) is proposed. In SPCE-NMUIGA, the adaptive setting of the multi-user weight and the crossover/mutation probability is mainly discussed, as well as the calculation way of the group fitness values, the selection of the convergence condition and the construction of the network platform based on the local area. In common, the automobile face modeling design problem is used as an example to execute simulation experiment, the results manifest that SPCE-NMUIGA is superior both in the optimizing efficiency and the group satisfaction of the optimizing result, and also can effectively sustain and command the population diversity.(5) For purpose of assisting the systematic implementation of IGA, the structure model of intelligent decision support system based on interactive genetic algorithm (IGA-IDSS) the intelligent decision support system structure model based on IGA is structured, which is exactly the general concept model of supporting for solving DMPTO. The six components of IGA-IDSS are elaborated in detail. In particular, the automobile face modeling design problem is used for a specific application example of DMPTO, and the automobile face modeling design system based on IGA (IGA-FMDSB). The coding mode of the automobile face modeling design is elaborately analyzed, and the implementation process and the system function of IGA-FMDSB are mainly addressed.