Study On Evolutionary Computation In Dynamic Environments

In this dissertation, studies are mainly focused on find effective solutions and innovative approaches to solving the dynamic optimization problems, propose some new optimization algorithms, which can adapt to environmental changes, from maintaining diversity and and evolvability, improving the eugenics mechanism, reusing historical information, detecting the change of environment, introducing the competitive relationship and other aspects. The main contributions of this thesis are as following:1) By maintaining the population diversity and evolvability, the environment adapting ability of evolutionary computation is improved. The effect of multi-objectivization in dynamic environment is investigated, and an instructional theory is proposed for constructing reasonable additional objective. Individual diversity and evolvability are introduced into evolutionary algorithms as additional information, and new multi-objectivization evolutionary algorithms are proposed. The computation results indicate that the individual diversity multi-objectivization evolutionary algorithm has the approving performance in maintaining population diversity, and dealing dynamic optimization. Evolvability is definted by two aspects, one based on firness improvement, the other based on genetypic change, which is introduced into new multi-objectivization optimization algorithms. The computation results indicate that the former can effectively solve the dynamic optimization problem with slight intensity, and the adapting ability of the latter is better in the random and intense dynamic environment.2) For reducing the process of individual adaptation in dynamic environment, by using family eugenics to improve the selection mechanism in dynamic environment, a dynamic evolutionary algorithm is proposed. In this algorithm, the global search ability is guaranteed by select parent individuals reasonably, middle individuals are confimed by orthogonal design methed, and diversity of population in maintained by interspecific competition. The computation results indicate that it can improve the global search ability and performance of maintaining diversity effectively without adding the computing burden excessively.3) In order to effectively and reasonable reuse the historical information to solve dynamic optimization problems, a new hybrid memory scheme, which is composed of short-term memory and long-term memory, is proposed, and based it, a dynamic evolutionary algorithm is proposed. This algorithm has a good adaptability and accuracy in dynamic environments with a variety of complex. Based on the hybrid memory scheme, a forecast scheme is introduced in dynamic optimization algorithms research, and a new multi-population evolutionary algorithm with forecast scheme is proposed. The useful environmental information is stored as memory sequences to build forecast models. When the environment changed, the dynamic environment can be predetermined by the forecast model. The computation results indicate that the introduction of forecast scheme can improve the optimum traceability effectively.4) With the purpose of predetermining the dynamic environment under the situation of none historical information, cloud model is introduced into dynamic evolutionary algorithms, which is a model of transforming a qualitative concept to a set of quantitative numerical values. Cloud predetermine scheme based on Y-conditional cloud model is proposed. When the changes are detected, according to forecast information, semi-drop cloud model with direction is utilized to tracking the dynamic optimum points. Inspired by the convergence property of cloud droplets, a cloud search scheme is produced to search local areas. A new hybrid genetic algorithm based on cloud model is proposed to deal with the dynamic multimodal problem. The computation results indicate that it has a good environmental adaptation capability and a high-precision dynamic searching capability.5) To improving the evlution quality of population in dynamic environment, the competitive relationship is introduced into cellular gentic algorithm. A predator-prey cellular genetic algorithm to solving dynamic optimization problems is proposed in which predator-prey model is used to replace the evolution rule in regular cellular genetic algorithm. The computation results indicate that it has the improving performance of maintaining population diversity, and can effectively solve the dynamic optimization problem with different complexity.