SIA OpenIR  > 数字工厂研究室
生物启发式协同优化算法及其在交通流预测中的应用研究;
Alternative TitleBiological heuristic collaborative optimization algorithm and its application in traffic flow prediction
王丹萍1,2
Department数字工厂研究室
Thesis Advisor胡琨元
Keyword生物启发式计算 协同进化 短时交通流预测
Pages122页
Degree Discipline机械制造及其自动化
Degree Name博士
2018-11-28
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract近十年来,随着人工智能理论的发展和生物群体社会行为研究的深入,国内外学者在简单地模拟个体间“竞争与协作”模式的基础上,进一步借鉴了生物系统中不同物种或同一物种不同群落间的依存关系,引入“协同进化”机制,提出了诸多改进策略,从而弥补了数学机理模型匮乏的不足。这种具有群体协作机制的生物启发式算法已成为计算智能领域研究的热点和重要分支。本文以此为背景,广泛地查阅了相关领域文献,分析国内外已有研究工作及存在的不足。在此基础上,借鉴层次演化模型与协同进化策略,针对粒子群优化算法、差分进化算法两类典型生物启发式计算方法进行改进,设计了多算法协同优化框架,以智能交通领域的短时交通流预测问题为背景,开展了应用研究。具体工作和创新性成果概括如下:1. 基于搜索历史的多种群协同粒子群优化算法。针对标准粒子群算法收敛速度过快且容易陷入局部最优的缺陷,引入了多种群协同优化策略,采用动态K均值聚类,实现解空间分布的种群划分,延缓种群快速收敛;为了提升高维问题求解过程中算法局部搜索的效率,采用适应度值树技术,防止粒子重复性访问,进而提出了基于搜索历史的多种群协同粒子群优化算法(Multispecies Coevolution Particle Swarm Optimization Based on Previous Search History, MCPSO-PSH)。通过仿真测试,结果验证了改进策略能够明显提升协同粒子群算法的全局搜索能力和局部开发能力。2. 基于动态种群策略的协同差分进化算法。针对差分进化算法求解多峰问题鲁棒性不强的缺陷,采用多种群协同进化策略提高算法的全局搜索能力,限制其因早熟而引发的搜索停滞现象。为进一步提升局部探索效果,采用K均值聚类的方式实现基于空间的种群划分;同时,针对DE/current-to-best/ 1/bin算法收敛过快的情况,设计了基于邻居关系的突变策略以限定个体的信息交流范围;为提升早期全局搜索能力和后期局部开发能力,引入了先进的年龄机制与生命周期策略,实现动态调整子种群数量,在淘汰与补偿之间寻到一种动态平衡,进而提出了基于动态种群策略的协同差分进化算法(Cooperative Differential Evolution with Dynamical population,DynCDE)。仿真测试结果证明了该算法的收敛速度适中并具备较强的局部开发价值。3. 基于动态种群的多算法协同优化框架。针对“没有免费午餐原理”所提出的单一算法普适应问题,采用经典的生物启发式算法作为子种群的优化因子,通过生命周期策略对子种群中个体性能进行评价,以营养值为核心,使占据优良解空间的个体分裂(复制),而占据非优空间的个体消亡(移除),进而突出优势算法的核心地位,实现资源的自适应集中;通过采用相互学习策略,有效保持处于劣势算法依然具有搜索效用;在此基础上,设计了基于动态种群的多算法协同优化框架(Self-adaptive Framework based on Lifecycle of Multiple Evolution Algorithms, SaFLMEAs),通过标准函数测试和对比分析,证明了该方法的有效性。4. 基于改进协同差分进化的短时交通流量预测方法。分析了短时交通流量预测问题的背景及国内外研究现状。针对传统BP神经网络在短时交通预测中存在的缺陷,采用基于动态种群的协同差分进化算法优化神经网络权值以获得更高的交通流历史数据模拟精度,进而实现指定交口、特定时段内交通流信息的精确预测,并通过仿真实例证实了该方法的有效性,对实际交通诱导系统的开发具有指导意义。
Other AbstractIn the past decade, with the development of artificial intelligence theory and the in-depth research on the social behavior of biological groups, after simply simulating "competition and collaboration" between individuals, domestic and foreign scholars introduced the mechanism of "co-evolution" and proposed many improvement strategies, which can made up for the lack of mathematical mechanism model. Biological heuristic algorithms with collective collaboration mechanism have become a hot spot and important branch in the field of intelligent computing. Based on this background, this paper has extensively reviewed relevant literature, and has analyzed existing research work and existing deficiencies at home and abroad. On this basis, referring to hierarchical evolution model and cooperative evolution strategy, two typical biological heuristic computing methods, particle swarm optimization algorithm and differential evolution algorithm, are improved, and a multi-algorithm collaborative optimization framework is designed. In this paper, application research is carried out against the background of short-term traffic flow prediction in the field of intelligent transportation. The specific work and innovative results are summarized as follows: 1. Multispecies Coevolution Particle Swarm Optimization Based on Previous Search History, MCPSO-PSH. Aiming at the PSO’s defects of fast convergence speed and easily falling into local optimum, a multi-population collaborative optimization strategy is introduced and dynamic k-means clustering is adopted to enhance the population division of spatial distribution solution and delay the rapid population convergence. In order to improve the efficiency of local search in solving high-dimensional problems, the technology of fitness tree is adopted to prevent the search repetition of the same solution space, And then Multispecies Coevolution Particle Swarm Optimization Based on Previous Search History (MCPSO-PSH) was proposed. Through the simulation test, the results verify that the improved strategy can significantly improve the global search ability and local development ability. 2. Cooperative Differential Evolution with Dynamical population, DynCDE. Aiming at the DE’s defect of weak robustness to solve the multimodal problem, the multi-population co-evolution strategy is adopted to improve the global search ability of the algorithm and limit the search stagnation caused by premature. In order to further improve the effect of local exploration, the spatial population division is achieved by the way of k-means clustering. Meanwhile, for overcoming the fast convergence of DE/current-to-best/ 1/bin algorithm, a mutation strategy based on neighborhood relationship was designed to limit the information exchange scope of individuals. In order to improve the ability of global search in early stage and local development in later stage, advanced age mechanism and lifecycle strategy are introduced to implement dynamically adjusting the number of sub-populations and keeping a dynamic balance between elimination and compensation. Finally, Cooperative Differential Evolution with Dynamical population (DynCDE) is designed. Simulation results show that the algorithm has moderate convergence speed and strong local development value. 3. Self-adaptive Framework based on Lifecycle of Multiple Evolution Algorithms, SaFLMEAs. For general adaptation of single algorithm proposed by No free lunch, in this paper, the classical biological heuristic algorithm is regarded as the optimization factors for subpopulations. And the individual performance of the subpopulation is evaluated by the lifecycle strategy. With the nutrition value as the core, lifecycle strategy divides (duplicates) the individuals occupying the good solution space and eliminates (removes) the individuals occupying the non-optimal space. Furthermore, the core position of the superior algorithm is highlighted and the adaptive resource concentration is achieved. By adopting the mutual learning strategy, the disadvantage algorithms still own search ability. And then Self-adaptive Framework based on Lifecycle of Multiple Evolution Algorithms (SaFLMEAs) is designed. The validity of the method is verified by the test against standard benchmark functions and their comparative analysis. 4. Short-term traffic flow prediction method based on DynCDE. This paper analyzes the background of short-term traffic flow prediction and the research status at home and abroad. Aiming at the BP neural network’s defects in short-term traffic prediction, DynCDE introduced in Section II was employed to optimize the weight of neural network to obtain higher simulation precision of traffic flow history data. Furthermore, the accurate prediction of traffic flow information at specific intersections and time periods is achieved. The effectiveness of the method is verified by a kind of simulation examples, which is of guiding significance to the development of practical traffic induction system.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/23637
Collection数字工厂研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
王丹萍. 生物启发式协同优化算法及其在交通流预测中的应用研究;[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
Files in This Item:
File Name/Size DocType Version Access License
生物启发式协同优化算法及其在交通流预测中(2990KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王丹萍]'s Articles
Baidu academic
Similar articles in Baidu academic
[王丹萍]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王丹萍]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.