摘 要

近年来随着国内越来越多的城市爆发雾霾问题，有关空气污染对人体健康状况的影响得到人民普遍的关注，如何在现代化城市建设中保持环境的协调发展已经成为政府的工作重心、社会的关注焦点。在针对性地对现有的空气污染问题进行整治前，必须从科学的角度认识空气污染与空气质量问题，并从中找出影响空气质量的核心指标参数，从而进行全面的、合理的治理策划。目前国内关于空气质量预测模型的研究并没有绝对统一的评价体系，而是受到很大程度的主观因素影响，因此很多时候研究人员不能很好地确定空气污染物指标和空气质量之间的非线性关系，使得相应的空气质量预测精度偏低，且难以大范围推广。

支持向量机（SVM）方法是一种以统计学习理论为理论基础的机器学习技术，以系统结构风险最小化原则为预测思路，从上世纪被提出以来，几十年来得到了国内外学者的深入研究和不断发展。为了有效解决其他预测方法存在的“过学习”问题，支持向量机引入了核函数、松弛变量以及基于结构风险化最低准则等概念，从而很好地解决了非线性的数据分类问题。目前支持向量机方法已经广泛应用于金融、生物信息识别、建筑科学等学科的线性不可分问题。

本文将粒子群优化算法（PSO）与支持向量机方法进行结合，构建了PSO-SVM预测模型；理论上，通过使用粒子群优化算法对支持向量机的核心参数进行参数寻优，可以有效保证支持向量机的核心参数的准确度，从而不仅可以大大缩短PSO-SVM预测模型的计算运行时间，更能大大提高预测模型的整体预测效果。本论文为验证PSO-SVM预测模型的实际预测精度，将针对典型的地下建筑—武汉市汉口火车站地下车站的空气质量问题进行实证分析，分别使用传统的支持向量机模型、基于遗传算法（GA）的支持向量机模型以及基于粒子群优化算法的支持向量机模型对该地下区域的空气质量问题进行回归预测，通过对比、分析相应结果，从而确定出对地下建筑空气质量预测精度最高的预测模型。

本论文将粒子群优化算法以及支持向量机的理论研究结合于当下实际民生热点，不仅从理论上验证了基于粒子群优化算法的支持向量机预测模型的精确度，更是未来城市地下建筑的空气质量实时调控的很好的理论研究基础，从而使得本论文的研究具备一定的实际指导意义。

关键词：机器学习；支持向量机；粒子群优化算法；地下建筑；空气质量预测

Abstract

With the haze problem continued to outburst in domestic city, the impact of air pollution issue for the human health is obtaining more and more attention; nowadays, having a coordinated development for the environment in the modern urban construction becomes one of the central work for the government, as well as the focused issue for the society. It is necessary to study the air pollution issue scientifically before regulating the problem, from which the core index parameter for affecting air quality could be recognized, then the reasonable regulation for improving the air quality would be obtained. At present, there is not an uniform framework for establishing an evaluation model for the air quality, which is usually affected by the subjective factors of the researchers and will be resulting that the non-linear relation between the air pollutant concentration and the air quality grade is hard determined, then the corresponding air quality forecast has large problems with low accuracy and efficiency.

The support vector machine method is a modern machine learning technique which is firstly proposed in last century and has a rapid development and improvement then, it is based on the statistical learning theory in principle and followed with the minimizationprincipleof the structural risk of the system. At present, the support vector machine method has been widely applied in various nonlinear problems of many fields, to avoid the“curse of dimensionality”, a nonlinearkernelfunctionis introduced into the support vector machine method so that the imposed variables from the low dimensional feature spacecanbe mapped into a high dimensional feature space, thus the nonlinear problem is turned into solving linear equations; the support vector machine method has a very outstanding generalization ability than other traditional statistical method, therefore it has beenwidely used in theareassuch as finance, biological information identification, architecture sciencenowadays.

In this paper, the traditional support vector machine method is combined with the particle swarmoptimizationalgorithm, thus a support vector machine monitoring and pre-warning model based on the particle swarm optimization algorithm is established, which is simply short for PSO-SVM model; through the parameter optimization of the particle swarm optimization, the precision of the core parameter of SVM model can be effectively ensured in theory, which will largely shorten the consuming time and effectively improve the precision of whole monitoring and pre-warning model; To verify the actual pre-warning precision of the PSO-SVM model mentioned in this paper, regressive prediction of the air qualify grade for the typical underground construction, which is the underground station in the Hankou railway station of Wuhan city, is researched respectively with the traditional SVM model, the SVM model basing on the genetic algorithm and the PSO-SVM model, by comparing with the corresponding results, the most effective and highest precision model for predicting air quality grade of the underground construction is obtained.

The support vector machine method isstudiedand combined with the focus issue of human livelihood in this paper, through which the accuracy of the PSO-SVM monitoring and pre-warning model is effectively verified, and this paper will have a significant theoretical guidance meaning forthereal-time regulation of the air quality of the underground construction in the future.

Keywords:Machine Learning; Support vector machine;Particle swarmoptimizationalgorithm;Underground construction;Air quality prediction