基于多元统计和智能算法的上海市空气质量指数评价分析

发布时间：2018-04-16 11:01

本文选题：空气质量指数 + C4.5　；参考：《兰州大学》2016年硕士论文

【摘要】：本文首先对上海市2014年空气质量指数进行时间序列分析,分析是否具有明显的“星期效应”、“季节性”、“春节效应”,使用小波变换进行分解,用低频信号来反映它总体变化趋势。利用描述性统计分析和模糊综合评价法来寻找出空气质量中的首要污染物PM2.5。然后做PM2.5与其他污染物的相关系数及关联规则分析,做主成分分析目的是把多个指标提取三个主成分,得到PM2.5与三个主成分的多元统计分析都通过模型的验证。最后使用人工智能算法进行评价分类预测,决策树C4.5和LVQ神经网络建立了污染物与空气等级的智能评价模型,通过仿真实验表明C4.5在空气质量评价分类预测较为准确;支持向量机算法采用交叉验证来选取正则化因子c和参数g,比较不同核函数选取以及不同的归一化实现方式对预测分类精确度的影响;因SVM分类器参数难以选择的问题,本文提出基于粒子群算法PSO全局优化搜索SVM中的正则化因子c和内核参数g,并对比BPNN。最后对比全部智能算法的分类预测精度,得出C4.5和PSO-SVM在空气质量分类预测精度最好这一结论。
[Abstract]:In this paper, the 2014 air quality index of Shanghai is analyzed in time series, and whether there are obvious "week effect", "seasonality" and "Spring Festival effect" is analyzed, which is decomposed by wavelet transform.A low frequency signal is used to reflect its general changing trend.The main pollutant PM2.5 in air quality was found by descriptive statistical analysis and fuzzy comprehensive evaluation.Then the correlation coefficient and association rules between PM2.5 and other pollutants are analyzed. The purpose of principal component analysis is to extract three principal components from multiple indexes and obtain that the multivariate statistical analysis of PM2.5 and the three principal components are verified by the model.Finally, the artificial intelligence algorithm is used to evaluate the classification and forecast. The decision tree C4.5 and LVQ neural network are used to establish the intelligent evaluation model of pollutant and air grade. The simulation results show that C4.5 is more accurate in air quality assessment.Support vector machine (SVM) algorithm uses cross validation to select regularization factor c and parameter g, and compares the effects of different kernel function selection and different normalization implementation methods on the accuracy of prediction classification, because the parameters of SVM classifier are difficult to select.This paper presents a global optimization search for regularization factor c and kernel parameter g in SVM based on particle swarm optimization (PSO).Finally, by comparing the classification and prediction accuracy of all intelligent algorithms, it is concluded that C4.5 and PSO-SVM have the best prediction accuracy in air quality classification.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：X823

【相似文献】