發(fā)布時間：2018-05-04 17:08

  本文選題：BP神經(jīng)網(wǎng)絡(luò) + MIV�。� 參考：《昆明理工大學(xué)》2017年碩士論文

【摘要】：近年來,空氣污染日漸成為一個嚴(yán)峻的問題。空氣質(zhì)量惡化對人身健康和環(huán)境存在巨大的或者潛在的危害。因此,大氣污染物濃度預(yù)報非常重要,它不僅對人們的日常生活有所幫助,而且對政府制定相關(guān)政策具有指導(dǎo)意義。2013年,國務(wù)院頒布了《大氣污染防治行動計劃》,要求各地建立監(jiān)測預(yù)警體系,京津冀、長三角、珠三角及其他省、副省級市、省會城市均包括在內(nèi)的城市或區(qū)域開展空氣質(zhì)量預(yù)報預(yù)警的工作。通過研究昆明市的污染物濃度預(yù)測模型,有助于昆明市空氣質(zhì)量預(yù)報預(yù)警工作的開展。以統(tǒng)計模型和機器學(xué)習(xí)模型為代表的非機理模型在污染物濃度預(yù)報中應(yīng)用廣泛,其中BP神經(jīng)網(wǎng)絡(luò)以其較強的非線性擬合能力在污染物濃度預(yù)測中廣泛應(yīng)用。本文利用BP神經(jīng)網(wǎng)絡(luò)結(jié)合變量篩選的方法建立了 SO2,NO2,O3,CO,PM10,PM2.5等6種污染物的濃度預(yù)測模型,并選取2014-1-1至2015-11-28時段,昆明市區(qū)6個環(huán)境監(jiān)測點6種污染物濃度的監(jiān)測數(shù)據(jù)建立了昆明市污染物均濃度預(yù)測模型。采用平均影響值(Mean Impact Value,MIV)的方法篩選出分別對6種污染物日濃度值有主要影響的變量,作為BP神經(jīng)網(wǎng)絡(luò)的輸入變量,利用建立的預(yù)測模型分別對6種污染物的日濃度進(jìn)行預(yù)測,并討論MIV的方法在濃度預(yù)測中應(yīng)用的可行性。(1)通過變量篩選的結(jié)果可以看出,前一日的其他污染物濃度對預(yù)報對象的濃度有較大影響;(2)BP神經(jīng)網(wǎng)絡(luò)模型的預(yù)測結(jié)果較好,預(yù)測的濃度水平和變化趨勢與實測值的變化吻合度較高。標(biāo)準(zhǔn)化平均偏差NMB均小于18,標(biāo)準(zhǔn)化平均誤差NMB均小于40,剩余標(biāo)準(zhǔn)差RMSE均小于30,相關(guān)系數(shù)R多大于0.6;(3)利用MIV方法對輸入變量篩選,有助于BP神經(jīng)網(wǎng)絡(luò)模型預(yù)測精度的提高,個別模型如關(guān)上監(jiān)測點N02,CO,碧雞廣場SO2,龍泉鎮(zhèn)S02,呈貢新區(qū)SO2,東風(fēng)東路S02、03的預(yù)測模型并不能提高預(yù)測精度;(4)各污染物的IAQI分指數(shù)的準(zhǔn)確率較高,可以達(dá)到70%以上,首要污染物的準(zhǔn)確可以達(dá)到50%左右,各點位的AQI均可達(dá)到65%以上。
[Abstract]:Air pollution has become a serious problem in recent years. The deterioration of air quality has great or potential harm to physical health and environment. Therefore, it is very important to predict the concentration of air pollutants. It is not only helpful to people's daily life, but also has a guiding significance for the government to make relevant policies for.2013 years, the State Council The action plan for the prevention and control of air pollution has been promulgated, which requires the establishment of monitoring and early warning system in all parts of the city, the Beijing Tianjin Hebei, the Yangtze River Delta, the Pearl River Delta and other provinces, the sub provincial cities and the provincial capital cities to carry out the air quality prediction and early warning in the cities and regions, which are included in the city and the provinces. The study of the pollutant concentration prediction model in Kunming will help the air quality in Kunming. The non mechanism model, represented by statistical model and machine learning model, is widely used in the prediction of pollutant concentration. The BP neural network is widely used in the prediction of pollutant concentration with its strong nonlinear fitting ability. In this paper, the method of BP neural network network combined with variable selection is used to establish SO2, NO 2, O3, CO, PM10, PM2.5, and other 6 kinds of pollutant concentration prediction model, and select the 2014-1-1 to 2015-11-28 period, the monitoring data of 6 pollutants concentration in 6 environmental monitoring points in Kunming City, establish the prediction model of the average concentration of pollutants in Kunming city. The average influence value (Mean Impact Value, MIV) is used to screen the daily concentration of 6 kinds of pollutants respectively. The variable which has the main influence on the degree value, as the input variable of the BP neural network, uses the established prediction model to predict the daily concentration of the 6 pollutants respectively, and discusses the feasibility of the application of the MIV method in the concentration prediction. (1) through the results of variable selection, it can be seen that the concentration of other pollutants on the previous day has the concentration of the forecast object. It has great influence; (2) the prediction results of BP neural network model are better. The predicted concentration level and change trend coincide with the measured values. The standard average deviation NMB is less than 18, the standard average error NMB is less than 40, the residual standard difference RMSE is less than 30, the relative number R is more than 0.6; (3) the MIV method is used to screen the input variable sieve. Selection is helpful to improve the prediction accuracy of BP neural network model. Some models, such as monitoring point N02, CO, BBI square SO2, Longquan town S02, Chenggong New Area SO2, Dongfeng East Road S02,03 prediction model, can not improve the prediction accuracy. (4) the accuracy of IAQI sub index of each pollutant is higher than 70%, and the primary pollutant is accurate. At about 50%, the AQI of each point can reach more than 65%.

【學(xué)位授予單位】：昆明理工大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：X51

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 孫f，

本文編號：1843862