节点文献

仿生智能优化算法及其应用研究

Research on Bio-inspired Intelligent Optimization Algorithm and Its Applications

【作者】 薛羽

【导师】 庄毅;

【作者基本信息】 南京航空航天大学 , 计算机应用技术, 2013, 博士

【摘要】 科学研究和工程应用领域经常遇到优化问题,如何设计有效的模型和算法求解这些优化问题一直是一个研究的热点。如,在电子对抗研究领域,国内外协同作战研究方面存在的武器-目标分配问题,这是一个典型的组合优化问题,本文针对电子对抗中的协同干扰问题建立了适合仿生智能优化算法求解的数学模型,并重点研究了如何改进现有仿生智能优化算法求解该模型。关于仿生智能优化算法,在理论和应用上需要进一步的研究,目前已有的IIGA等算法在背景项目的应用中仍然存在许多问题有待研究解决。其中,下面三个方面的不足最为研究人员关注:(1)算法的普适性、鲁棒性仍然有待提高。(2)算法的延展性不够,算法性能随着优化问题规模的增大而迅速降低。(3)已有算法如何应用于工程优化问题还需要进一步的研究。论文工作在分析前人已有成果的基础上,重点对(1)和(3)两个问题开展了仿生智能优化算法及其应用的研究工作。本文主要研究成果概括如下:(1)增强型自适应进化算法提出了增强型自适应进化算法(ESEA)。设计了贪婪繁殖算子、策略选择算子、X进化算子、种群多样性维持算子和进化策略学习算子来组成算法的进化结构;设计了多种有效的进化策略,即候选解产生策略;设计了概率选择策略并用于进化种群中的个体,采用改进的概率模型计算策略被选择的概率;引进了一种学习机制,可根据策略在进化过程中的表现自适应地学习每种策略的选择概率。通过对比实验发现提出的增强型自适应进化算法相比同类算法提高了普适性和鲁棒性,并且新的操作算子、多进化策略和自适应学习机制对算法性能的提升起重要的作用。(2)基于自适应学习群体搜索技术的集成进化算法针对第一算法难以适应战场态势多变、战场投入的武器和目标规模与日俱增的问题,提出了基于自适应学习群体搜索技术的集成进化算法(EEA-SLPS)。算法采用了多种群随机搜索技术和并行工作机制。与增强型自适应进化算法相比,本算法的主要特点在于集成了多种随机搜索技术,并使它们以有效的方式进化子种群。在该算法中,将整个进化群体分成三个子群体,并采用三个子算法分别对子群体进行进化。论文设计了多种不同性质的信息交流方式(IEMs)。做了大量的IEMs性能测试实验。通过对实验数据结果的分析,发现信息交流方向应由包含整个种群最优解的子群指向不包含整个种群最优解的子群,交流方向不应该是预定义的,应是动态自适应的。对比实验结果表明所设计的算法比同类算法在鲁棒性和普适性上均有所提高。(3)求解协同干扰决策问题的启发式自适应离散差分进化算法针对现代作战环境中多UCAV (Unmanned Combat Air Vehicle)协同对抗多部威胁雷达任务规划这一军事运筹决策问题,提出了基于多指标干扰效能综合评估方法的多UCAV协同干扰决策问题优化模型。为了有效求解多UCAV协同干扰决策问题模型,提出了启发式自适应离散差分进化算法。为了提高算法求解特定领域问题的效率,设计了基于威胁度的扩展型整数编码方案、基于威胁度的启发式个体调整操作和基于约束满足的个体修复等操作。实验结果表明提出的启发式自适应离散差分进化算法相比同类算法具有更高的鲁棒性和更好的求解效率。(4)自适应离散差分进化算法策略选择提出了求解协同干扰武器目标分配问题(CJWTA)的自适应离散差分进化(SaDDE)算法。好的策略池决定SaDDE算法的主要性能。论文中引入了基于相对排列顺序的标度法(RPOSM),通过RPOSM改进了层次分析法,提出了基于RPOSM的层次分析法(RPOSM-AHP)以解决策略选择问题,通过理论和实验数据结合的方法给出了解决策略选择问题的可行方案。

【Abstract】 Optimization problems widely exist in scientific and engineering fields. It has been a hot topicthat how to design effective models and algorithms to solve these optimization problems. Forexample, the WTA problems in the research field of ECM. It is a classical combination problem.This paper tries to construct optimization model for the WTA problem of cooperative jammingand solve it through bio-inspired intelligent optimization method.It needs research in theory and in application about these bio-inspired intelligent optimizationalgorithms. Many problems which exist in the algorithms such as IIGA for the WTA applicationneed to be overcome. Among which, the researchers mainly pay close attention to the followingthree aspects:(1) The universality and robustness of the algorithms remain need to be enhanced.(2) The ductility of the algorithms is insufficient, the performance decreases quickly with thedimension of the problem increases.(3) It needs further research that how to apply existingalgorithm to engineering optimization problems. Based on the analysis of previous work, wehave done some research work on the algorithms and their applications. The mainly researchresults are summarized as follows:(1) Enhanced self-adaptive evolution algorithmGreedy breeding operator, strategy selecting operator, X-evolution operator, populationdiversity maintaining operator, strategy learning operator are designed to consist the algorithmstructure. Besides, the algorithm uses many effective candidate solution generation strategies(CSGSes). An improved probability model is adopted to describe the probability of a strategybeing used to update an antibody. A self-adaptive learning mechanism is introduced into thealgorithm, so the selection probability of each strategy can learning from its experience ingenerating new individuals. Experimental results demonstrate that the proposed algorithm ismore effective than its competitors, and the new operators, multiple solution generatingstrategies and self-adaptive mechanism are effective for generating better solutions.(2) Ensemble evolution algorithm with self-adaptive learning techniquesAccording to the battlefield situation is easily change and the scale of the weapon and thetarget increasing day by day. Thus, an ensemble evolution algorithm with self-adaptive learningtechniques is proposed. The algorithm integrates many population based stochastic searchalgorithms in parallel manner. Different from the proposed algorithm in (1), this algorithmmainly focuses on ensemble of the sub-algorithms which with self-adaptive learning mechanismand make them evolve the sub-algorithms in effective manner. In the proposed algorithm, thepopulation is divided into three sub-populations and the sub-algorithms are employed to evolvethe sub-populations in parallel manner, respectively. We have designed many differentinformation exchange manners (IEMs). Then, many experiments have been done. We have found that the information exchange direction should be from the sub-population which with the globalbest individual at current generation to the sub-population(s) without, the direction should not bepredefined and should be self-adaptive. Experimental results demonstrate that the proposedalgorithm with better performance of robustness and universality.(3) Cooperative jamming decision making based on heuristic self-adaptive discrete differentialevolution algorithmAn optimization model of cooperative jamming decision problem is proposed based onmulti-index jamming effect comprehensive evaluation method for the military operations taskprogramming problem of multiple UCAVs (Unmanned Combat Aerial Vehicles) confrontmultiple threaten radars. In order to solve the model effectively, a Heuristic Self-adaptiveDiscrete Differential Evolution (H-SDDE) algorithm is proposed. In the algorithm, threatendegree based extensional integer coding scheme, heuristic individual adjust operator andindividual repair process are designed. The experimental results indicate that the proposedalgorithm is better than its competitors.(4) Research on strategy selecting of self-adaptive discrete differential evolution algorithmThe Self-adaptive Discrete Differential Evolution (SaDDE) is proposed to solve theCooperative Jamming Weapon-Target Assignment (CJWTA) problem. Obviously, the strategypool plays a significant role in the SaDDE algorithm. First, the Relative Permutation Order basedScale Method (RPOSM) is introduced. The analytic hierarchy process is improved by theRPOSM. Then, the RPOSM based analytic hierarchy process (RPOSM-AHP) is proposed for thestrategy selecting problem. Finally, a feasible scheme is given to solve the strategy selectionproblem by combining the theory and experimental results.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络