基于整体经验模态分解和随机森林的城市PM2.5预测

发布时间：2018-07-05 00:17

本文选题：整体经验模态分解 + 随机森林回归　；参考：《长春工业大学》2017年硕士论文

【摘要】：近年来,PM2.5已经严重影响到了人们的生活和出行,大气环境污染问题较为严重。我国一些大中型城市的年均PM2.5浓度已经远远高于世界卫生组织环境空气质量的指导值10ug/m3,使得患心血管和呼吸系统疾病的人逐年增加。PM2.5作为形成雾霾的主要污染物,已成为城市大气污染治理的首要目标。因此,研究PM2.5的变化趋势,掌握其变化规律,建立合理精确的预测模型,对其进行短期预测,不仅能提醒人们及时采取有效措施,保护身体健康。而且对调控一些社会活动和对重度环境污染带来的危害进行提前预警具有一定的现实意义。本文以构建合理的PM2.5预测模型为目的,采用整体经验模态分解法对北京市PM2.5时间序列进行处理,研究了北京市PM2.5的波动规律及其周期性变化。首先,采用整体经验模态分解(EEMD)方法对PM2.5数据进行分解,对PM2.5的变化趋势、波动和周期性变化进行分析;其次,运用主成分分析回归(PCR)方法、支持向量机回归(SVR)的预测方法、自回归滑动平均(ARIMA)的方法、随机森林(RF)回归等方法对北京市PM2.5进行短期预测;最后,提出基于整体经验模态分解和随机森林的PM2.5预测方法,即通过对北京市PM2.5数据进行EEMD分解,对分解后得到的固有模态函数和PM2.5的影响因子进行随机森林建模。通过比较ARMA、SVR、PCR、RF和EEMD-RF这五种方法对PM2.5预测的精确度,以探究最佳的PM2.5预测模型。
[Abstract]:In recent years, PM2.5 has seriously affected people's life and travel. The average annual concentration of PM2.5 in some large and medium-sized cities in China is much higher than that of the WHO's guiding value of ambient air quality, which makes the number of people suffering from cardiovascular and respiratory diseases increase year by year as the main pollutant to form haze. It has become the primary target of urban air pollution control. Therefore, studying the change trend of PM2.5, mastering its changing law, establishing a reasonable and accurate prediction model and forecasting it in the short term can not only remind people to take effective measures in time to protect their health. Moreover, it is of practical significance to regulate and control some social activities and to forewarn the harm caused by heavy environmental pollution. In order to construct a reasonable prediction model of PM2.5, the global empirical mode decomposition (EMD) method is used to deal with the PM2.5 time series in Beijing, and the fluctuation law and periodic variation of PM2.5 in Beijing are studied. Firstly, the global empirical mode decomposition (EEMD) method is used to decompose PM2.5 data, and the variation trend, fluctuation and periodicity of PM2.5 are analyzed. Secondly, the prediction method of support vector machine regression (SVR) is presented by principal component analysis (PCR). The methods of autoregressive moving average (Arima) and random forest (RF) regression are used to predict PM2.5 in Beijing in the short term. Finally, a forecasting method of PM2.5 based on global empirical mode decomposition and stochastic forest is proposed, which is based on EEMD decomposition of PM2.5 data in Beijing. The natural mode function and the influence factor of PM2.5 are analyzed by stochastic forest modeling. In order to explore the best prediction model of PM2.5, the accuracy of PM2.5 prediction was compared between ARMA-SVRN PCRN RF and EEMD-RF.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X831

【参考文献】