Study Of Theory Of Hierarchical Genetic Algorithms With Applications

Theory of hierarchical genetic algorithms with applications to design of a system structure, parametic determination and job-shop scheduling problems are investigated in this dissertation.Firstly, theory of hierarchical genetic algorithms is presented. Based on the complexity of problems at hand and different from existing conventional genetic algorithms, a hierarchical representation of a chromosome is proposed with a chromosome arranged in a layered structure and, thus, a hierarchical genetic algorithm is proposed. Furthermore, an adaptive hierarchical genetic algorithm and combination of simulated annealing algorithm with the hierarchical genetic algorithm are proposed.Next, job-shop scheduling problems are dealt with using the algorithm proposed. A parallel machine scheduling problem with total weighted completion time is tackled based on the proposed algorithm through well-designed-coding of chromosomes which clearly reflects job scheduling policy. Results from case studies show that the algorithm proposed in this dissertation can be extended to applications to large-scale parallel identical and non-identical machine scheduling problems. Further, a flexible job-shop scheduling problem is solved, delivering optimal job scheduling policy with minimal completion time, which means the algorithm is encouraging in practical uses.Then, it is based on hierarchical genetic algorithm that structure of a fuzzy system and related parameters are determined at the same time and the system is then used for economic forecast. Compared with the BP neural networks, the model is simple and effective.Furthermore, structure and parameters of neural networks are determined based on the algorithm proposed. BP neural network, a financial crisis warning model, and a four-layer BP neural network for population forecast are studied, respectively. A well-designed representation of chromosomes is used to perform training task of the two BP neural networks with both connection weights and numbers of neurons in a hidden layer determined at the same time. It is shown that models trained based on hierarchical genetic algorithm are able to work well.Finally, a more complex problem in connection with determination of both structure and parameters of a system is again investigated. A wavelet neural network with continuous parameters is modelled based on the hierarchical genetic algorithm proposed and used for stocks market forecast. Following good performance of the new method applied to case study with practical data sets, it can then be concluded that the proposed algorithm can be widely applied to modeling and forecast of complex systems such as stock markets.