Research Of Prediction Model On PM_2.5 Concentration Based On Data Analysis

Nowadays,in pace with the high-speed expansion of economy in China,industry,transportation,service industries have brought the great material needs for the people'lives.Meanwhile,the pollution of many cities are becoming increasingly serious due to the dense population of the city,the aggregation of industries and the overload of transportation and so on.Among them,the smog weather has been concentrated in many cities,which have a serious impact on people's live and environment.Importantly,PM_2.5 has become the major factors of air pollution.Thus,understanding the variation,the level of pollution and the affective factors of atmospheric PM_2.5 concentration and exploring the suitable prediction model of atmospheric PM_2.5 concentration have the reference frame of people'life and environmental management department.The data collected by the monitoring station of Beijing University of Technology is taken as the experimental sample.The statistical methods are used to study the variation of PM_2.5concentration and related influencing factors in different season and period.Moreover,the machine learning methods are established to predict the PM_2.5 concentration.The primary contents are as follows:1.Under the collected historical data,the change trend of atmospheric PM2.5concentration in different season and period are analyzed.The results show that PM_2.5concentration is higher in autumn and winter,and has a significantly higher trend at night than during the day.At the same time,from the perspective of pollutants and meteorological factors,the relevant factors of atmospheric PM_2.5 concentration are studied to find the reliable model input variables.The results show that PM_2.5 has a strong positive correlation with PM₁₀,NO₂,SO₂ and temperature,while has a strong negative correlation with relative humidity.Therefore,the influence factors of PM₁₀,NO₂,SO₂,temperature and relative humidity should be fully considered when the prediction model of PM_2.5 concentration is established.2.A prediction model of PM_2.5 concentration based on BP_Adaboost neural network is proposed.Firstly,the gray relation analysis is used to remove the redundant variables.Secondly,the PSO algorithm is optimized to the weight and threshold of BP_Adaboost neural network.Finally,the simulation results show that the prediction performance is better than other prediction models.However,it is found that PM_2.5concentration is highly uncertain and non-stationary through studying the original PM_2.5 concentration and simulation results.PM_2.5 concentration may be abrupt at some time and the prediction performance of single model is insufficient,which can cause the larger prediction error.So the accuracy of the single prediction model needs to be further improved.3.A prediction model of PM_2.5 concentration based on CEEMD and multi-model is proposed.Firstly,PM_2.5 concentration has strong nonlinear and non-stationary characteristics,and it is decomposed by the complementary ensemble empirical mode decomposition method.Secondly,four univariate and multivariate PM_2.5 concentration prediction models are established respectively.Then,the final PM_2.5 concentration prediction results are obtained through the adaptive weight adjustment strategy.Finally,the simulation results show that the prediction accuracy of the proposed model is significantly better than the single model.However,the proposed model needs to predict the each IMF separately,which has a certain computational complexity.PM_2.5concentration has a strong randomness and it is expected to contain more uncertainty information.Therefore,in view of the above problems,this paper will continue to study the PM_2.5 concentration interval prediction model.4.A interval prediction model of PM_2.5 concentration based on ENN is proposed.Firstly,in view of the high computational complexity in the above modeling process,the SE is proposed to optimize it for reducing the complexity in the modeling process.Secondly,the quality evaluation index of the interval prediction model is proposed.And a novel optimization criterion of PSO algorithm is used to optimize the Elman neural network weight?and threshold b.Then,multi-input and double-output Elman neural network is used to achieve the PM_2.5.5 concentration interval prediction.Finally,the simulation results show that the CEEMD-SE-PSO-ENN model significantly improves the reliability and effectiveness of prediction results compared with ENN model and CEEMD-SE-ENN model.