面向PM2.5空气污染的多重分形与协同群体智能算法研究

PM2.5, as the primary air pollutant, has attracted wide attention. How to effectively analyze and forecast PM2.5 has become a hot topic. In view of the high-dimensional and nonlinear characteristics of air pollution system, the multifractal theory and swarm intelligence optimization algorithm will be used in the research of the meteorological environment. First of all, the theory of cooperative coevolution is used to glowworm swarm optimization algorithm, combined with the MapReducde programming model for parallel processing. The fractal feature selection method based on cooperative coevolution glowworm swarm optimization algorithm is proposed, which improve the precision and efficiency of the traditional fractal feature selection method. The problem of the selection of PM2.5 the influence factors is solved. Secondly, the spatial and temporal evolution mechanism of PM2.5 based on multifractal theory is reseached, and the correlation between PM2.5 and influencing factors based on multifractal theory is also reseached; Finally, the artificial fish algorithm and glowworm swarm optimization algorithm under the framework of culture algorithm is structured, then belief space and population space are designed. The selective ensemble learning method based on the culture under the framework of swarm intelligence algorithm is proposed, which improves the generalization performance of traditional ensemble learning method. Correspondingly, the PM2.5 forecast and warning model is constructed. The implementation of this project will promote the theoretical research and practical application of multifractal and swarm intelligence optimization algorithm, which provides scientific support for the prevention and control of air pollution .

PM2.5作为首要空气污染物引起了广泛的关注，如何对其进行有效地分析和预报成为当前的研究热点。针对空气污染系统高维、非线性等特点，本课题拟引入多重分形理论与协同群体智能优化算法对其进行挖掘。首先，将协同进化引入萤火虫群优化算法，并结合MapReduce编程模式将其并行化处理，进而提出基于协同进化萤火虫算法的分形属性选择方法，提高传统分形属性选择方法的精度和效率，解决了PM2.5气象环境影响因素选取的问题；其次，研究基于多重分形理论的PM2.5空气污染时空演变机制，以及PM2.5与影响因素之间的分形相关性；最后，构建文化框架下的萤火虫和鱼群算法，设计信仰空间、种群空间，使用文化框架下的多群体智能算法进行选择性集成学习，提高传统集成学习的泛化性能，从而构建PM2.5空气污染预测预警模型。本项目的实施将对多重分形和群体智能优化算法的理论研究和实际应用起到推动作用，为空气污染的防治提供科学支持。

空气污染日益严重，影响到民众的日常生活。PM2.5作为首要空气污染物引起了广泛的关注，如何对其进行有效地分析和预报成为当前的研究热点。针对空气污染系统高维、非线性等特点，本项目对空气污染物影响因素的选取、空气污染物时空演化特征、空气污染物预测预警进行了研究。本项目的主要成果包括：1、研究了基于群体智能算法与多重分形维数的空气污染影响因素选择。首先对鲸鱼算法进行优化，提出了基于混合自适应策略的鲸鱼优化算法，并对其进行Spark并行化处理；然后提出基于群体智能优化算法的分形属性选择方法，使用改进后的群体智能优化算法作为搜索策略，多重分形维数作为属性子集评价准则，并将于应用于空气污染属性选择中。2、研究了基于多重分形的空气污染时空演化特征。首先使用基于EMD的MF_DFA方法分析了单个PM2.5空气污染时间序列的Hurst指数，分析PM2.5时间序列的特性；然后使用Coupling-DFA分析PM2.5与相关影响因素之间的关系。3、研究了基于群体智能算法的空气污染预测预警方法。首先，提出了文化框架下的粒子群算法与鲸鱼算法；然后提出了基于Kappa测度与人工蜂群的改进随机森林方法；最后将上述方法应用于空气污染预测预警中。此外，还研究了图像分割技术在空气污染中的应用。在本项目的资助下，项目成员发表学术论文13篇，申请发明专利3项目，培养研究生3人。项目研究成果深化了对群体智能算法以及分形理论的理解，扩宽了群体智能算法的应用范围，较好地达成了项目的预期目标。