适于高维空间搜索的自组织学习算法

Self-organizing learning algorithm for multidimensional optimization applications

提出了一种有别于当前优化算法框架的自组织学习算法(self-organizing learning algorithm,SLA),该算法融合遗传算法并行搜索与模拟退火串行搜索,结合粒子群学习机制和禁忌搜索机制,实现了系统与环境的交互学习,能够很好地处理传统优化方无法应对的高维非线性优化问题。SLA分自学习和互学习两个智能化学习阶段,先进行基于自学习机制的邻域禁忌搜索,保证局部极值的收敛,然后通过信息共享平台,进行基于互学习机制的广域禁忌搜索,保证全局极值的收敛。系统通过与环境交互学习而自适应地调整搜索策略和相关参数,使得搜索过程能够有效地避免盲目性,而具有相当的自组织性。最后,通过高维测试函数的对比仿真实验表明,SLA在由小型低维空间转入超大型高维空间时,仍能够与环境保持稳定、透明的交互学习,其全局搜索能力和整体稳健性明显优于其它搜索方法。

Traditional optimization methods are unable to deal with the multidimensional non-linear optimization problem which involves a great number of discrete variables and continuous variables. In order to cope with this situation, a self-organizing learning algorithm （SLA）is proposed, in which the parallel search strategy of genetic algorithm and the serial search strategy of simulated annealing algorithm are involved. Additionally, the learning principle of particle swarm optimization and the tabu search strategy are involved in the SLA, wherein the integrated frame work is different from the traditional optimization methods and the interactive learning strategy is involved in the process of random searching. The SLA is divided into two handling courses： self-learning and interdependent-learning. The local optimal solution will be achieved through the self-learning in the process of local searching and the globally optimal solution will be achieved via the interdependent learning based on the mechanism for information sharing. The search strategy and controlled parameters of the SLA are adaptively fixed according to the feedback information from interactive learning with the environment, thus the SLA is selforganizing and intelligent. Experiments for the multidimensional test functions show that SLA is superior to other optimization methods.