Feature Extraction And The Research Of Related Analysis Based On The Data Of PM_2.5 Pollution In Beijing

The damage of PM2.5 is very extensive, and mainly reflected in two aspects. The first is that PM2.5 can enter the human lung, respiratory system and even cardiovascular system, which will cause harm to human health.On the other hand, PM2.5 is an important reason for the formation of haze weather, which has brought the serious influence to people’s social life.Therefore, it is of great significance to research the PM2.5 pollution. Based on the PM2.5 data, precursor pollution data and meteorological data in Beijing, this article analyzes and studies the distribution feature of PM2.5 in Beijing and the relationship between the concentration of PM2.5 and meteorological factors.Then, the principal component analysis method which is one of the multivariate statistical methods is applied in the process of the research about the PM2.5 pollution. The main aspects of this article include the following:1. The distribution of PM2.5 pollution in Beijing is comprehended through the aspects of the situation of pollution exceeding, daily distribution, and seasonal distribution by utilizing the PM2.5 pollution data in Beijing. It makes use of the feature extractionthat the data is analyzed to reveal the interaction between PM2.5 and relative humidity, wind speed, sunshine time, and the local wind field.The results of the study found that, during the period from 2009 to 2013 the average annual pollution level isabout 100μg·m-3;the contamination level of PM2.5 in autumn and winter significantly higher than that in spring and summer;twin peaks in the morning and eveningappear in dailydistribution.In the weather of high humidity and no wind, particulate matter is easy to accumulate which results in the formation of haze weather and the reduction of sunshine hoursIn addition, the special terrain of Beijing that this city is surrounded by mountains on three sides makes the particles quickly spread in the northwest wind.2. According to the literature researches and the data, this article determines ten variables including SO2、NO2、CO、O3、PM10、PM2.5、relative humidity、air temperature、air pressure and wind speed as the based variables in this heavy pollution of PM2.5 prediction model. Then the modeling data are preprocessed:linear interpolation method is used to supplement missing data about PM2.5 and the precursor pollutants, and for the meteorological data, expectation maximization method is additionally applied to replenishthe missing data besidesthe linear interpolation.3. Based on the method of principal component analysis and the cross validation methods, the method of SPE statistical control limits is put forward to detect the abnormal data and delete them in training data which ensures the effectiveness of the training data and obtains the prediction model of PM2.5 se- vere pollution. By this way, it makes the prediction rate of the model is greatly increased. Compared with complex numerical prediction model, the model is convenient and the forecast accuracy is higher. In the end, the conclusion is th-at five is the best number of principal components in this model, the main ele ment of the cumulative variance contribution rate in this model is about 98%, and the correct prediction rate can reach more than 90%. This method of abnor-mal data identification and PCA training is not reported in current literature.