Research On Air Quality Prediction Based On Multiple Model

With the continuous improvement of science and technology in China,industrial technology has been upgraded.But the rapid industrialization has brought about many problems such as overcapacity,high consumption of urban resources and over density of population have led to serious environmental pollution.Deterioration of air quality is aggravated.The air quality has seriously affected people’s health and public hazards.Most pollution events are caused by various severe haze phenomena.For example,the people’s health travel and quality of life has seriously affected by haze and public property is damaged in Beijing,Tianjin and Hebei.The main pollutants are PM_2.5,PM₁₀,NO₂,SO₂,etc.,where the main polluting particles are PM_2.5.To improve air quality and reduce air quality degradation,the changing patterns of PM_2.5concentrations and pollution levels are better understood.The effective and intelligent prediction methods are used to predict atmospheric PM_2.5concentration.The reliable and effective information is provided for people’s travel and prevention,which has become a hot issue in current research.This research topic relies on the platform of Beijing University of Technology campus air monitoring system.The air pollution particulate matter and meteorological conditions data are used to collect as research samples.The multiple model theoretical ideas and intelligent prediction methods are used in atmospheric pollutant concentration prediction.The main research work and innovation has several points in the thesis as following:1.Research on the prediction of pollutant concentrations that affect air qualitySeveral different models were developed to predict pollutant concentrations affecting air quality for the problems of long training time,high data demand,overfitting and falling into local minima of traditional neural networks.The long and short memory（LSTM）network is used to predict NO₂concentration,the particle swarm and least squares support vector regression（PSO-LSSVR）method are combined to predict PM_2.5concentration and the online sequence extreme Learning Machine（OS-ELM）method is used to predict PM_2.5concentration,respectively.In this thesis,the advantages of each model are used to apply and compare with traditional neural networks.The simulation results show that the speed and practicality of predicting pollutant concentrations is improved by the proposed method,and avoids the disadvantages of traditional neural network prediction models in terms of long training time,large sample size and the tendency to fall into local miniaturization.The simulation results show that the proposed method improves the accuracy,speed and practicality of predicting pollutant concentrations,and avoids the disadvantages of traditional neural network for long training time,large sample size and the tendency to fall into local minimization.An interactive multiple model（IMM）method based on time series is proposed for the PM_2.5concentration under different air quality levels.The model is established by the PM_2.5concentration data in the corresponding different air quality levels separately,which is combined with each Kalman filter.A hybrid model form of time series and Kalman filter is established.Then the prediction model is switched by Markov chain to achieve accurate prediction of PM_2.5concentration under different air quality levels.The simulation results show that the proposed IMM method can overcome the drawback of inaccurate prediction due to the influence of a single model under different air quality levels,thus the accuracy of PM_2.5concentration prediction is improved.The nonlinearity,uncertainty and stochasticity always exist in atmospheric environment with incorrect or unknown a priori noise,which is resulted in a decrease in the prediction accuracy of pollutant concentrations.Therefore,a combined support vector regression and adaptive unscented Kalman filter（SVR-AUKF）method is proposed to predict PM_2.5concentration and noise estimation.The method is based on the state equation framework of support vector regression model,which is combined with an adaptive unscented Kalman filter equation to predict PM_2.5concentration with incorrect or unknown system noise or measurement noise.The simulation results show that the PM_2.5concentration is accurately predicted with the incorrect or unknown noise.Which is estimated adaptively.The prediction accuracy and robustness of PM_2.5concentration are improved in the case of unknown and incorrect noise,and closed to the actual environment.There are differences in the best models established in different environments.Even the chosen best model in the same environment can be changed with changes of PM_2.5concentration.Establishing the corresponding prediction model is an important research problem to accurately predict the PM_2.5concentration.Thus,a multiple models adaptive unscented Kalman filter（MMAUKF）method is proposed to predict atmospheric PM_2.5concentration.According to the change of PM_2.5concentration in different time periods of the day,which is divided into three time periods.The PM_2.5concentration prediction models are established in the three time periods,and the models are combined with each adaptive unscented Kalman filter to predict the PM_2.5concentration in the corresponding time periods.Then the Bayesian fusion method is used to fuse the predicted PM_2.5concentration in three time periods to obtain value of different time periods of final day.The simulation results show that the PM_2.5concentration can be accurately predicted by the proposed method,and the accuracy and robustness of prediction is improved.2.The research on the dispersion of pollutant concentrations that affecting air qualityAiming at the concentration of sulfur dioxide（SO₂）discharged from wet flue gas desulfurization of a thermal power plant.A simulation research based on the pollutant concentration dispersion is proposed in this thesis.According to the wet flue gas desulfurization process,a Gaussian plume dispersion model of discharged SO₂is established,and the SO₂concentration dispersion is simulated according to the set conditions of the surrounding environment.The simulation results show that there is a direct relationship between SO₂concentration dispersion and wind speed,and the law of SO₂concentration dispersion can be known well,which can provide a certain reference for the monitoring of air pollution condition and environmental in this area.