Research On Elder-keeping Immune Genetic Algorithm And Its Application To Optimization Of (N+M) Fault-tolerant Systems

Immune genetic algorithm, which combines immune algorithm with genetic algorithm, is a guided random searching method. Although the algorithm has better performance than original two algorithms, it still needs to be improved in convergence and searching capability. Elder-keeping immune genetic algorithm, which is an improvement of common immune genetic algorithm, is proposed in this thesis and satisfactory results are achieved when it is used for solving the optimization of typical test functions and model of (N+M) fault-tolerant systems.Firstly, the characteristics, the foundational theory, development and application of immune genetic algorithm are summarized, and the working principles, reliability modeling, optimal model and graphics-based problem-solving method of (N+M) fault-tolerant systems are discussed in this thesis. Secondly, elder-keeping immune genetic algorithm, which integrates the elder-keeping strategy and immune genetic algorithm, is put forward, and its entire design method is given in the thesis. Elder-keeping strategy operates individuals of population in selections, crossover and mutation,and reserves both elders and younger in the new population. The strategy is propitious to improve the excellence degree of population and quicken convergence speed of the algorithm. Vector distance based antibody density mechanism method can enhance population diversity and improve global searching capability of the algorithm. And with self-adaptive crossover operator and mutation operator, Elder-keeping immune genetic algorithm can obtain better global searching capability and convergence performance. Finally, the elder-keeping immune genetic algorithm, which is realized by Matlab language, is used to optimize two typical test functions and (N+M) fault-tolerant systems. Calculation result shows that the elder-keeping immune genetic algorithm obtains more satisfactory global convergence performance than other algorithms mentioned in the thesis.