动态多策略差分演化算法及其在无线传感器网络能量分配优化中的应用

The influence of differential mutation strategy is very imporant to differential evolution (DE). However, the choice of the best strategy is crucial and difficult for a specific problem. In addition, optimal power allocation is one of the core issues in designing a wireless sensor network (WSN). In order to improve the performance of DE and to provide an efficient technique for the optimal power allocation in WSNs, in this project, we will combine the learning automata with DE to implememt the dynamic multi-strategy adaptive selection in DE. Based on the learning algorithms in learning automata, the optimal strategy in the pool will be adaptively assigned the highest reward for the problem at hand quickly. And hence, it obtains the highest selection probability. According to the selection probabilities of strategies in the pool, better strategy will obtain more chance to be selected to proporgate offspring in the following generation..In this project, based on the theory of learning automata we will study the dynamic multi-strategy adaptation in DE and its application in optimal power allocation in WSNs. We will focus on the following key issues: (1) how to assign the reward that is able to balance the fitness improvement and diversity change for numerical optimization problems; (2) how to design the fast strategy selection technique based on learning automata; (3) how to implement the technique of dynamic strategy pool management; and (4) how to solve the expensive computational problems efficiently in optimal power allocation..The results of our research will further enhance the performance of DE and provide the examples to implememt multi-strategy adaptation in the DE algorithm. In addition, the proposed techniques will also be applicable to multiple opeators adaptation in other evolutionary algorithms. Furthermore, our proposed algorithms will provide alternatives for efficient optimal power allocation in WSNs. Thus, we can believe that our research in this project is very important in terms of both the development for the DE algorithm and the applications of improved DE variants in different fields.

差分变异策略对差分演化算法的性能具有十分重要的影响，但是针对所求解的问题选择最优策略是很困难的。能量分配优化是无线传感器网络（WSN）设计中核心问题之一。为了提高差分演化算法的性能，并把改进算法有效应用于WSN 能量分配优化中，本项目拟把学习自动机理论与差分演化算法有效结合，研究多策略自适应选择差分演化算法，并利用改进算法求解WSN能量分配优化问题。我们将重点对以下四个关键内容进行系统研究：（1）有效平衡适应值改进和多样性改变的信度分配方法；（2）基于学习自动机的快速策略选择技术；（3）动态策略库管理技术；（4）求解能量分配优化昂贵计算问题的高效优化技术。项目所得到的研究成果一方面能改进差分演化算法的性能，为演化算法多策略、多算子自适应选择研究提供示例，另一方面能为求解WSN能量分配优化问题提供有效的优化技术，因而在差分演化算法的改进与应用研究上具有十分重要的理论意义和应用价值。

差分演化算法的研究是当前演化算法领域的研究热点和研究前沿。差分演化算法提出了多个不同搜索策略，它们具有不同特点，适合求解不同的问题，然而如何根据所求解问题选择最合适的策略是很困难的。能量分配优化问题对无线传感器网络的设计具有十分重要的作用。为此，本项目结合学习自动机理论研究多策略自适应差分演化算法。项目主要研究内容：1）提出基于cheap surrogate模型的多算子自适应演化算法框架；2）提出基于排序机制的无约束和约束差分演化算法，有效平衡算法的勘探与开采能力；3）分析了杂交算子与杂交概率的本质关系，提出基于杂交概率修复机制的自适应差分演化算法；4）提出了基于迁移学习理论的迁移差分演化算法框架；5）改进差分演化算法在不同类型燃料电池模型参数提取中的应用；6）提出约束信度分配技术及其在工程优化问题和无线传感器网络能量分配优化问题中的应用；7）演化Kalman滤波器的设计。项目的研究发表论文14篇，其中SCI检索论文9篇（含IEEE Transactions on Evolutionary Computation论文1篇、IEEE Transactions on Cybernetics论文2篇），1篇论文入选ESI高被引论文。本项目的研究过程中培养硕士研究生9名（3人已毕业）、本科生13名（均已毕业）。项目的研究成果在多策略自适应演化算法和迁移学习演化算法等研究方面具有重要理论意义，对改进算法在不同领域应用起到推动作用。