Hybrid Clone Competition And Stimulating Learning Strategies Roles Random Walk Particle Swarm Optimization

In the thesis, the intelligent optimization algorithms are analyzed and designed by the following viewpoint: the intelligent optimization algorithms can be modeled by two components. One is the stochastic search operation based on swarm; the other is the searching stategies which model the intelligent behaviors of life phenomena or biological systems. According to the above viewpoint, a new intelligent optimization algorithm can be got by designing the two components respectively.The stochastic search operation based on swarm is a basic characteristic of the intelligent optimization algorithm. In the thesis, the neighborhood intervals of a particle, the neighborhood set of a particle are defined in the object space Rn by a particle swarm located in the object space Rn. Using the above definition, a theorem which described that the neighborhood set of a particle can cover the objective space Rn is proposed and proved. According the theorem, a search operation called as the stochastic neighborhood search operation (SNSO) which can search the objective space Rn is proposed. When a particle swarm searches the object space using SNSO, the particle swarm becomes a randomly-walking particle swarm (RWPS). In order to make the randomly-walking particle swarm search the whole objective space Rn globally and efficiently, the randomly-walking particle swarm is decomposed into the different roles which carry out the search by the different strategies. So, the randomly-walking particle swarm becomes a multi-roles randomly-walking particle swarm (mRRWPS) which is capable to search the whole objective space Rn globally and efficiently.The other basic characteristic of the intelligent optimization algorithms is that their search is controlled by the stategies which model the intelligent behaviors of life phenomena or biological systems. In the thesis, the search stregies of the intelligent optimization algorithm are classified as the competitive strategies based on the selection mechanism and the collaborative strategies based on the heuristic information. In the multi-roles randomly-walking particle swarm, the clonal selection strategy and heuristic strategy based on the elite neighborhood are employed to control their stochastic neighborhood search operation, and a multi-roles randomly-walking particle swarm algorithm (mRWPSA) is built.In the thesis, the mRWPSA is analyzed by the theoretical demonstration, and tested by the experiments.In terms of the theoretical demonstration, firstly, according to SNSO and the the theorem which described that the neighborhood set of a particle can cover the objective space Rn, it is demonstrated that the mRWPSA is capable to search the objective space globally; secondly, according to the convergent theorem of Markov chain, it is demonstrated that the mRWPSA is convergent.Before the mRWPSA is tested by the experiments, the control parameters of mRWPSA are discussed. There are three control parameters in the mRWPSA. The first one is the number of particles of mRRWPS ( m ); the second one is the average number of cloned particles of each particle in one dimension during an evaluation circle ( Np ). The last one is local neighborhood factor ( lf ).In test experiment, firstly, the effects of three parameters (m, Np, lf) on the performance of mRWPSA are demonstrated by 30 benchmark functions; secondly, the test results of 22 benchmark functions are compared with ones of 7 classical intelligent algorithms. All test results demonstrate the following conclusions.1), Using the different values of the contral parameters (m, Np, lf), the mRWPSA can stably converge to the global optima or the close-to-optimal solutions of all 30 benchmark function, which demonstrate that the mRWPSA is capable to search the objective space globally, stably and rubustly.2), the control parameters——m, Np, lf respectively reflect the effects of the stochastic neighborhood search operation, the clonal selection strategy, and the heuristic strategy based on the elite neighborhood on the mRWPSA. The control parameters (m, Np, lf ) have the more effects on the convergent speed of the mRWPSA, but have the less effects on the stability and robustness of the mRWPSA. By selecting the proper control parameters according the characteristics of the optimization problem, the performance of convergent speed of the mRWPSA can be improved.