PDF(1849 KB)
基于PCA-BP神经网络的PM2.5季节性预测方法研究
张怡文, 郭傲东, 吴海龙, 袁宏武, 董云春
南京林业大学学报（自然科学版） ›› 2020, Vol. 44 ›› Issue (5) : 231-238.
PDF(1849 KB)
PDF(1849 KB)
基于PCA-BP神经网络的PM2.5季节性预测方法研究
Seasonal prediction of PM2.5 based on the PCA-BP neural network
,
【目的】分季节预测PM2.5浓度值,利用PCA方法对数据进行降维,分析季节及气象因素对PM2.5的影响,在提高预测准确率的同时降低时间复杂度。【方法】以合肥市2014—2017年的PM10、SO2、CO2、CO、O3浓度值,以及同时段的气象因素值,对PM2.5浓度进行预测。数据分析中发现PM2.5在不同季节浓度差异较大,故本研究选择分季节进行预测;为了提高预测准确率,加入如风力、温度、湿度、气压等气象因素进行预测,同时采用主成分分析(PCA)的方法进行数据降维,将降维后的数据再输入BP神经网络模型进行预测。【结果】实验采用3组实验进行对比:5种污染物指标(PM2.5-5)预测PM2.5、加入气象因素的综合12项指标(PM2.5-12)预测PM2.5、对综合指标进行PCA处理后的(PM2.5-PCA)预测PM2.5。实验结果表明:4个季节的PM2.5浓度值有较大变化,均方根误差(RMSE)的差值较大;采用PM2.5-PCA的方法,在任何季节的RMSE均有降低,相关系数(r)均有所提高。【结论】PM2.5浓度具有季节性特征,采用季节性预测方法可以提高预测准确率;同时采用PCA方法进行降维,可以在保证准确率的同时降低预测时间复杂度。
【Objective】 We predict the concentration of PM2.5 during different seasons and use the Principal Component Analysis (PCA) method to reduce the dimensionality of the data, while also improving the accuracy of the prediction and reducing the time complexity, to serve as a reference for travel by people and government decision-making. 【Method】 The PM2.5 concentration was forecast based on the values of PM10, SO2, CO2, CO and O3 concentration in Hefei from 2014 to 2017, and the meteorological factors during the same period. The data analysis found that the concentration of PM2.5 varies greatly across seasons; therefore, this study is focused on the forecasting during different seasons. To improve the accuracy of forecasting, influencing meteorological factors such as wind, temperature, humidity and air pressure were added to the forecasting. The PCA method was used for data dimensionality reduction, and then, the data were input into the BP neural network model for prediction. 【Result】 The experiment used three groups of assessments for comparison: five types of pollutant indicators to predict PM2.5 (PM2.5-5), addition of twelve comprehensive indicators of meteorological factors to predict PM2.5 (PM2.5-12), and use of comprehensive indicators which were processed by PCA (PM2.5-PCA) to predict PM2.5. The experimental results showed that the PM2.5 concentration in the four seasons had large changes, and the difference in Root Mean Square Error(RMSE) is large, when using the PM2.5-PCA method. We found that if the RMSE is reduced in any season, the correlation coefficient (r) value is increased. 【Conclusion】 The PM2.5 concentration value has seasonal characteristics, and the seasonal prediction method can improve prediction accuracy. Moreover, adopting the PCA method to reduce data dimensionality can ensure accuracy and decrease the time complexity at the same time.
PM2.5 / neural networks / prediction / principal component analysis(PCA)
| [1] |
李令军, 王占山, 张大伟 , 等. 2013—2014年北京大气重污染特征研究[J]. 中国环境科学, 2016,36(1):27-35.
|
| [2] |
The inorganic main elements, trace elements and PAHs were determined from selected PM(1), PM(2.5) and PM(10) samples collected at the Nordic background station in Virolahti during different seasons and during the wildfire episodes in 2006. Submicron particles are those most harmful to human beings, as they are able to penetrate deep into the human respiratory system and may cause severe health effects. About 70-80%, of the toxic trace elements, like lead, cadmium, arsenic and nickel, as well as PAH compounds, were found in particles smaller than 1 microm. Furthermore, the main part of the copper, zinc, and vanadium was associated with submicron particles. In practice, all the PAHs found in PM(10) were actually in PM(2.5). For PAHs and trace elements, it is more beneficial to analyse the PM(2.5) or even the PM(1) fraction instead of PM(10), because exclusion of the large particles reduces the need for sample cleaning to minimize the matrix effects during the analysis. During the wildfire episodes, the concentrations of particles smaller than 2.5 microm, as well as those of submicron particles, increased, and also the ratio PM(1)/PM(10) increased to about 50%. On the fire days, the mean potassium concentration was higher in all particle fractions, but ammonium and nitrate concentrations rose only in particles smaller than 1.0 microm. PAH concentrations rose even to the same level as in winter.
|
| [3] |
杨孝文, 周颖, 程水源 , 等. 北京冬季一次重污染过程的污染特征及成因分析[J]. 中国环境科学, 2016,36(3):679-686.
|
| [4] |
张晓茹, 孔少飞, 银燕 , 等. 亚青会期间南京大气PM2.5中重金属来源及风险[J]. 中国环境科学, 2016,36(1):1-11.
|
| [5] |
刘小生, 李胜, 赵相博 . 基于基因表达式编程的PM2.5浓度预测模型研究[J]. 江西理工大学学报, 2013,34(5):1-5.
|
| [6] |
DUNEAD,
|
| [7] |
彭斯俊, 沈加超, 朱雪 . 基于ARIMA模型的PM2.5预测[J]. 安全与环境工程, 2014,21(6):125-128.
|
| [8] |
贺祥, 林振山 . 基于GAM模型分析影响因素交互作用对PM2.5浓度变化的影响[J]. 环境科学, 2017,38(1):22-32.
|
| [9] |
刘杰, 杨鹏, 吕文生 , 等. 基于气象因素的PM2.5质量浓度预测模型[J]. 山东大学学报(工学版), 2015,45(6):76-83.
|
| [10] |
|
| [11] |
沈剑波, 雷相东, 李玉堂 , 等. 基于BP神经网络的长白落叶松人工林林分平均高预测[J]. 南京林业大学学报(自然科学版), 2018,42(2):147-154.
|
| [12] |
王嫣然, 张学霞, 赵静瑶 , 等. 北京地区不同季节PM2.5和PM10浓度对地面气象因素的响应[J]. 中国环境监测, 2017,33(2):34-41.
|
| [13] |
姚达文, 刘永红, 丁卉 , 等. 气象参数对基于BP神经网络的PM2.5日均值预报模型的影响[J]. 安全与环境学报, 2015,15(6):324-328.
|
| [14] |
杨笑笑, 汤莉莉, 张运江 , 等. 南京夏季市区VOCs特征及O3生成潜势的相关性分析[J]. 环境科学, 2016,37(2):443-451.
|
| [15] |
陈刚, 刘佳媛, 皇甫延琦 , 等. 合肥城区PM10及PM2.5季节污染特征及来源解析[J]. 中国环境科学, 2016,36(7):1938-1946.
|
| [16] |
|
| [17] |
|
| [18] |
肖小兵, 刘宏立, 马子骥 . 基于奇异谱分析的经验模态分解去噪方法[J]. 计算机工程与科学, 2017,39(5):919-924.
|
| [19] |
XIEX,
In this paper, a novel Gabor-based kernel principal component analysis (PCA) with doubly nonlinear mapping is proposed for human face recognition. In our approach, the Gabor wavelets are used to extract facial features, then a doubly nonlinear mapping kernel PCA (DKPCA) is proposed to perform feature transformation and face recognition. The conventional kernel PCA nonlinearly maps an input image into a high-dimensional feature space in order to make the mapped features linearly separable. However, this method does not consider the structural characteristics of the face images, and it is difficult to determine which nonlinear mapping is more effective for face recognition. In this paper, a new method of nonlinear mapping, which is performed in the original feature space, is defined. The proposed nonlinear mapping not only considers the statistical property of the input features, but also adopts an eigenmask to emphasize those important facial feature points. Therefore, after this mapping, the transformed features have a higher discriminating power, and the relative importance of the features adapts to the spatial importance of the face images. This new nonlinear mapping is combined with the conventional kernel PCA to be called
|
| [20] |
关蓓蓓, 郑思俊, 崔心红 . 城市人工林空气负离子变化特征及其主要影响因子[J]. 南京林业大学学报(自然科学版), 2016,40(1):73-79.
|
| [21] |
污染物数据采集[EB /OL].http://www.pm25.in/.2013-08-06/2018-2-20.
|
| [22] |
气数据采集[EB /OL].http://www.weather.com.cn/.2008-05-16/2018-5-10.
|
/
| 〈 |
|
〉 |
