Research On Modeling Of Air Quality Decision Support System Based On Artificial Intelligence

Air pollution is an issue across the world.It not only directly affects the environment and human health,but also influences the regional and even global climate by changing the atmospheric radiation budget,resulting in extensive and serious impacts.There is serious air pollution in China,so the government has attached great importance to pollution control and launched a series of plans from environmental management and scientific research,focusing on the cause and treatment,early-warning technology,health impact,etc.Therefore,the research about air quality forecast and environmental impact estimation has great application value.In recent years,with the swift growth of computing power and data,artificial intelligence has been developing rapidly and successfully applied to many fields.This provides a new pathway for the research on air pollution,namely,to introduce artificial intelligence to study air pollution.Air pollution and artificial intelligence are current research hotspots.Based on the intersection of air quality forecast and early-warning,exploring their interdisciplinary research is undoubtedly significant in theory and application.However,the research has just started,and thus there is still much room for improvement.For instance,considering the insufficient capacity of single machine learning model for data mining,it's necessary to improve the ability of model to process the highly non-stationary time series of pollutant.Previous research always simulates only one pollutant,which limits the application and validation of the model.In addition,there are some problems in air quality evaluation and atmospheric environmental impact estimation,which restricts the targeted and effective prevention and control of air pollution.Therefore,aiming at above shortcomings,a complete air quality decision support system is established based on artificial intelligence,which includes three subsystems,namely air quality forecast,air quality evaluation and environmental impact estimation.The experimental results demonstrate that the system has better prediction performance and provides accurate air pollution information and decision support,which shows a good application prospect.(1)Based on traditional machine learning methods,a combined model of machine learning for air quality forecast(ICEEMDAN-WOA-ELM)is established to make up for the deficiency of previous models for data mining.Firstly,missing data of pollutant concentration are handled by Cubic Spline Interpolation in the preprocessing module.Secondly,in the forecast module,the original sequence is decomposed into some Intrinsic Mode Functions(IMF),and then each IMF is learned and forecasted by the Extreme Learning Machine optimized by the Whale Optimization Algorithm,and all the forecast results of IMF are synthesized to obtain the final result.Lastly,four statistical indicators are applied to assess the performance of model in the assessment module.The experiment tests the performance of model for one-step prediction of daily average concentration of six conventional air pollutants in Beijing-Tianjin-Hebei region.The results show that the accuracy of ICEEMDAN-WOA-ELM for all cities and pollutants is better than other benchmark models including Autoregressive Integrated Moving Average model,General Regression Neural Network,Extreme Learning Machine,etc.Signal decomposition algorithm and optimization algorithm can significantly improve the performance of artificial neural network.(2)We further apply advanced deep learning methods and integrate the features of pollution source and meteorological condition to establish a combined model of deep learning for air quality forecast(TSTM),which improves the forecast accuracy for heavy pollution weather.Firstly,the feature engineering module applies Expectation Maximization algorithm to deal with missing data and Min-Max algorithm to normalize the input data,and completes the selection and construction of features based on domain knowledge,correlation coefficient and Sliding Time Window algorithm.Secondly,the forecast module adopts the CNN-Bi LSTM-Attention in the pollution source submodule and the meteorology submodule to learn and forecast corresponding features respectively,and obtains the final results through the fusion submodule(Convlstm).Lastly,the performance assessment module can assess the forecast performance of model for pollutant concentration,air quality and heavy pollution weather.The experiment tests the performance of model for multistep prediction of hourly concentration of six conventional air pollutants in Beijing-Tianjin-Hebei air pollution transmission channel("2+26"cities).The results show that the error of TSTM is positively correlated with pollutant concentration and forecast step,and the forecast correlation is more easily affected by random deviation when the concentration is lower.The accuracy of TSTM for pollutant concentration,air quality and heavy pollution weather is better than other benchmark models including Radial Basis Function network,Deep Belief Network,Elman neural network,etc.The results based on two independent test sets are similar,and TSTM has good robustness and generalization ability.(3)AQFCE,a fuzzy comprehensive evaluation model for air quality,is established based on fuzzy math to overcome the shortcomings of Air Quality Index(AQI)and improve the air quality evaluation and early-warning.Firstly,the preprocessing module adopts the least squares piecewise polynomial fitting/linear regression algorithm to deal with missing data/reversal data of pollutant concentration.Secondly,the evaluation module applies the concepts of membership degree,factor weight and fuzzy operator,and proposes a new calculation grid and weighting algorithm to obtain the important information of chief pollutant,relative risk and air quality level.Lastly,the early-warning module can release detailed early warning information of air pollution according to evaluation results.The experiment compares and analyzes the evaluation results of AQI and AQFCE under different space,time,tasks and objects.The results show that AQFCE can accurately reflect the status and change trend of air quality.The evaluation results of the two models are similar for the daily report,and the comprehensive strategy of AQFCE is more reasonable than the conservative strategy of AQI.Affected by incomplete evaluation standards,there are problems and even antinomies in the results of AQI,while AQFCE is effective.(4)PDHE,a systematic atmospheric environmental impact estimation model,is established based on statistical learning to reveal the impacts of air pollution on environment and human beings at different levels,and provide decision support for the government on pollution control.Firstly,at the nation level,Spearman's rank correlation coefficient is used to estimate the change trend of comprehensive status of air quality,and then the effectiveness of pollution control policy is assessed and analyzed.Secondly,at the province level,the air pollution disaster risk index(APDRI)is proposed to estimate the trend,level and driving factors of pollution disaster risk in each province.Thirdly,at the city level,the general model of health risk assessment is applied to estimate the number of cases for health endpoints attributed to air pollution exposure.Lastly,according to the results of health risk estimation,the corresponding economic losses are estimated by the comprehensive application of the value of statistical life(VSL),the cost of illness method(COI)and benefit transformation method(BT).The experiment carries out environmental impact estimation of PM_2.5 in China from 2015 to 2019.The results show that air quality has been improved significantly because of the effective implementation of the policy,and industrial source and domestic source are the main emission sources of PM_2.5.The disaster risk generally shows a downward trend,but there are notable regional differences.The disease burden due to PM_2.5 exposure is heavy.PM_2.5 concentration and population density are the main causes of health burden,and the economic burden in underdeveloped cities is usually greater.