The Application Of Evolutionary Algorithm On Dynamic Optimization Problems

Evolutionary Algorithms have been widely applied in all kinds of static optimiza-tion problems, and a lot of research results which had important practical applications have been achieved in this aspect. In the real world, there is an important type of dynamic optimization problems, the objective functions, constraints and environment related parameters of these optimization problems may change over time, and it will also lead to the global optimum of the optimization problems changed accordingly. Dynamic optimization problems has broad application backgrounds, the research of its solution and its application has been developed as an important area.A common way to solve this kind of problem is restart the algorithm once the detection of environmental changes. But this kind of method has significant limitations, on the one hand, completely abandoning the historical information leads to decreased efficiency of the algorithm; on the other hand, it's hard to detect the environment' s change in practical application. At the same time, if environment's change cycle is short, it is difficult for the algorithm to find a satisfactory solution before start the algorithm from scratch, because the algorithm needs some iteration steps to converge to the global optimum. The conventional evolutionary algorithms can often finally converge to a satisfying solution after some iteration steps, however, the diversity of population is difficult to guarantee after the evolutionary operator's operation. In this paper, in view of the dynamic characteristics of the optimization problem, some effective evolutionary strategies are integrated into the conventional evolutionary algorithms to ensure that the algorithm to maintain the population diversity, and enhance the global exploring ability of the algorithms, which make the algorithm better adapt to dynamic environment, the specific research contents are as follows:1. From the formalized definition of dynamic optimization problems, to the sum-marize of the related researches of evolutionary optimization and evolutionary dynamic optimization, concentrating on a detailed review in the research of dynamic optimization problems. The formal definition of dynamic optimization problems, the characteristics of dynamic optimization problems, the category of dynamic optimization problems, the type of environmental change, as well as the basic framework for solving dynamic optimization problems and related performance evaluation are described in detail.2. The effect of multi-population strategy in dynamic optimization problems is an-alyzed, a master-slave dual population dynamic genetic algorithm based elite migration combined with the memory mechanism, which is suitable for0-1encoding optimization problems, is proposed. In this algorithm, the memorized elite individuals are involved in the evolving process, making the algorithm to keep exploring new global optima as well as make full use of the historical information. At the same time, different selection strategies in each population are adopted, thus keep the balance of exploration ability and exploitation ability of the algorithm. The effectiveness of the algorithm is verified by numerical experiments based on a set of commonly used benchmark functions.3. On account of the nature of easily lose population diversity of conventional evolutionary algorithms, a self-adaptive neighborhood search dynamic particle swarm optimization, combining with multiple roles in the process of the evolution, is proposed. The individuals in the population are assigned different roles according to its fitness value, and different update strategies are adopted in the process of updating process, so as to better maintain the diversity of population, providing a guarantee for the algorithm to track the optima changing trajectory. The ability of the algorithm to track optimum changing trajectory was verified by the numerical experiments on moving peak benchmark under different environment cycle and different environmental change intensity.4. In order to balance algorithm convergence speed and population diversity, a spatially neighborhood best search differential evolution is proposed. The proposed algorithm uses the spatially neighborhood best individual as the local best individual in the mutation strategy/DE/best/1. Two different individuals with the nearest between parent population and offspring population are selected to generate new population using tournament selection, thus keep the population diversity so that it can track the optima change trajectory. The effectiveness of the tracking of optima trajectory is verified by moving peak benchmark under different change cycle and different change intensity.