演化计算普遍增强策略及其在三峡库区滑坡研究中的应用

发布时间：2018-04-06 05:08

本文选题：演化计算　切入点：勘探与开采　出处：《中国地质大学》2017年博士论文

【摘要】：演化计算原理简单、易于操作、通用性强而且特别适合大规模并行处理,因而应用广泛。为了获得高质量的解,演化计算在整个运行过程中都需要保持勘探与开采的平衡。滑坡的发展是不确定的复杂非线性演化系统。三峡库区滑坡数据丰富,为基于数据的滑坡研究提供了可行性。然而,此类研究还有待进一步开展。基于上述两方面的背景,本文首先致力于研究能够在演化计算中广泛应用的维护勘探与开采平衡的机制,即演化计算普遍增强策略。然后在此基础之上进行三峡库区滑坡监测点位移时间序列聚类。文中首先提出了以精英为中心的演化群体多样性,其取值就是各个精英到各个其它个体的两两相异程度的平均值。如果演化计算中使用了精英策略,则可利用具有相同度的,基于该概念的群体多样性度量,精确表达以精英为中心的群体多样性,并显著降低计算的时间复杂度。另外,无论算法中是否采用了精英策略,都可以使用度为1的该方法计算群体多样性,从而以可接受的误差为代价,降低求解群体多样性的时间复杂度。随后,三种演化计算普遍增强策略被依次提出。它们的应用场景有所不同,可以相互组合使用。首先被提出的是分层分布式演化的移动式群体迁移策略。在这种策略中,全局迁移的对象是子种群,并采用了移动式迁出-替换方案。实验证明,在采用非随机迁移拓扑的前提下,该策略可以显著提高分层分布式演化算法的求解能力。基于子种群多样性选择分布式演化算法的迁移时机,是提出的第二个策略。该策略规定,迁入某子种群的个体进入这个子种群的概率由既定的公式决定。实验结果证明本策略可以显著提高分布式演化算法的求解性能。第三个策略是演化算法基于距离的勘探性辅助选择。根据该策略,在演化的任意一代中,都有一定的概率利用辅助选择替换主要选择。在辅助选择中,首先执行度为1的精英策略。某个体到精英的距离被当做该个体的临时适应值(以大为优)。然后,仍然按照主要选择中规定的选择模式选择个体。实验结果证明,添加该策略有助于解决停滞或早熟,从而提高演化算法解的质量。接下来是演化计算普遍增强策略在滑坡研究方面的应用。基于移动式群体迁移策略的分层分布式自动差分演化聚类算法被用于解决三峡库区滑坡监测点位移时间序列聚类问题。聚类结果为滑坡预测提供了一些启示。
[Abstract]:Evolutionary computing is simple in principle, easy to operate, versatile and especially suitable for large-scale parallel processing, so it is widely used.In order to obtain a high quality solution, the balance between exploration and exploitation should be maintained during the whole operation of evolutionary computation.The development of landslide is an uncertain complex nonlinear evolution system.The landslide data in the three Gorges Reservoir area is abundant, which provides the feasibility for the landslide research based on the data.However, such studies need to be further developed.Based on the above two aspects, this paper first studies the mechanism of maintaining the balance between exploration and exploitation, which can be widely used in evolutionary computing, that is, the general enhancement strategy of evolutionary computing.Then the time series of displacement of landslide monitoring points in the three Gorges Reservoir area are clustered.In this paper, it is proposed that the evolution population diversity centered on the elite is the average of the different degrees of each elite to other individuals.If the elitist strategy is used in evolutionary computing, the measure of population diversity with the same degree and based on this concept can be used to accurately express elitist diversity, and the time complexity of computation can be significantly reduced.In addition, whether or not elite strategy is used in the algorithm, the method with degree 1 can be used to calculate population diversity, thus reducing the time complexity of solving population diversity at the cost of acceptable error.Subsequently, three general enhancement strategies for evolutionary computing were proposed in turn.Their application scenarios are different and can be combined with each other.Firstly, a mobile group migration strategy with hierarchical and distributed evolution is proposed.In this strategy, the object of global migration is a subpopulation, and a mobile output-replacement scheme is adopted.Experimental results show that the proposed strategy can significantly improve the solving ability of hierarchical distributed evolutionary algorithm under the premise of adopting non-random migration topology.The second strategy is to select the migration time of distributed evolutionary algorithm based on the diversity of subpopulations.The strategy states that the probability of an individual moving into a subpopulation is determined by the established formula.Experimental results show that the proposed strategy can significantly improve the performance of the distributed evolutionary algorithm.The third strategy is distance-based exploration-assisted selection based on evolutionary algorithm.According to this strategy, in any generation of evolution, there is a certain probability to replace the main selection with auxiliary selection.In assistive selection, an elite strategy with a degree of 1 is first implemented.The distance from an individual to the elite is regarded as the temporary fitness of the individual.Then, the individual is still selected according to the selection mode specified in the main selection.Experimental results show that adding this strategy helps to solve stagnation or precocity, thus improving the quality of the solution of the evolutionary algorithm.Then there is the application of the general enhancement strategy of evolutionary computing in landslide research.The hierarchical and distributed automatic differential evolution clustering algorithm based on mobile group migration strategy is used to solve the displacement time series clustering problem of landslide monitoring points in the three Gorges Reservoir area.The clustering results provide some enlightenment for landslide prediction.
【学位授予单位】：中国地质大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：P642.22

【相似文献】