大规模全局优化问题的高效协同演化算法研究

The mathematical models of many problems in real word are often large-scale global optimization problems, and existing algorithms for solving directly these problems are difficult and easy to fall into local optimal solution. Cooperative coevolution algorithms for solving large-scale problems have showed a good performance. Therefore, for large-scale global optimization problems, it has very important theoretical and practical significance to study the efficient cooperative coevolution algorithms. For the characteristics that these problems are large-scale, non-convex, and not differentiable, this project based on cooperative coevolution will study the following several technologies to improve the effective of large-scale global optimization algorithms. First, intend to design a new grouping technology to reduce the dimension of problems; second, intend to design some efficient evolutionary operators and local search techniques to solve the sub-problems, so that the algorithm search the feasible more overall and deeply; finally, due to the number of local optimal solutions is too much, so that the algorithm is easy to fall into local optimal solution and cannot find the global optimal solution, then a new method will be constructed that can escape from local optimum. Based on the above techniques, a large-scale global optimization algorithm with cooperative coevolution will be designed.

现实世界中的不少问题其数学模型往往是大规模全局优化问题，已有算法很难直接求解此类问题，并且容易陷入局部最优。协同演化算法在求解大规模问题时表现出了较好性能，因此对大规模全局优化问题研究高效协同演化算法具有十分重要的理论和现实意义。针对此类问题维数高、非凸、不可微等特点，在研究高效协同演化算法的基础上，本项目拟将提出多个优化技术以提高大规模全局优化算法的效率。首先，拟设计出一种新的分组技术，对大规模问题进行降维处理；其次，对降维后的子问题进行求解，拟设计出高效的进化算子和局部搜索技术，使得算法能更全面更深入地探索搜索区域；最后，由于局部最优点数目较多，算法容易陷入局部最优，拟设计出新的跳出局部最优解的方法. 综合以上技术从而设计出求解大规模全局优化问题的高效协同演化算法。

本项目研究使用协同演化算法求解大规模问题的相关算法及应用。针对此类问题维数高、非凸、不可微等特点，在研究高效协同演化算法的基础上，本项目提出多种优化技术以提高大规模全局优化算法的效率。首先，设计出一种新的分组技术，对大规模问题进行降维处理；其次，对降维后子问题进行求解，设计了高效进化算子和多个局部搜索技术，使得算法能更全面更深入地探索搜索区域；最后，由于局部最优点数目较多，算法容易陷入局部最优，设计了新的跳出局部最优解的方法。综合以上技术构造了求解大规模全局优化问题的高效协同演化算法，并研究了该算法的应用。