Research And Application Of A Hybrid Model Based On Secondary Denoising And Multi-objective Optimization For An Air Pollution Early Warning System

With the continuous development of China’s social economy,the consumption of fossil fuels and the holdings of motor vehicles have increased year by year,and they resulting in the increasingly serious air pollution in most cities in China,and the haze weather have increased significantly,which had a serious impact on the social,economic,public health and ecological environment.In this context,society’s demand for air quality related information is increasing.Most studies on air pollution focus on the health effects of poor air quality.Although researches on the forecasting of air pollutants concentrations have increased in recent years,most of these studies focused on the forecast of individual pollutant concentration and improved the forecasting accuracy.Few studies aim to analyze air pollution and establish early warning system from an application perspective.In this paper,an air pollution early-warning system was constructed,its aim to forecasting air pollutants concentrations and to evaluation air quality based on the forecasting concentrations.The results of the early warning system can be used to guide human production and life,avoiding more harm from air pollution.From the framework of the early warning system,it contains two modules:air pollution forecasting module and air quality evaluation module.And a novel hybrid forecasting model was proposed in forecasting module by integrating secondary denoising idea,multi-objective optimization algorithm,and a new forecasting model.The accuracy of forecasting is improved by reducing noise information in the original sequence and optimizing the parameters of forecasting model.In the second module,we proposed an air quality fuzzy synthetic evaluation system,which was based on the hourly concentration of PM2.5,PM10,SO2,NO2,CO and O3.In this paper,the main work can be summarized as the following:First of all,establish an air pollution early warning system,which consisting of two modules:high-precision air pollution forecasting and air pollution comprehensive evaluation.The system forecasts the concentration of various pollutants,and then evaluates the air quality by using the established fuzzy synthetic evaluation system.The evaluation results can not only meet the needs of the air quality supervision department,but also meet the needs of human daily life.Besides,from the perspective of dynamic reconstruction of time series and combined with empirical orthogonal functions that can to reduce the noise and identify the components with trends and periodic in the original sequence.The reconstructed sequence forecasted by using a new forecasting method,the extreme learning machine based on L2,1 norm and random Fourier transform(L2,1RF-ELM).And used multi-objective optimization algorithm to to find the optimal parameters in the forecasting method so as to improve the accuracy of pollutant concentration prediction,and provides reliable data support for constructing air pollution early warning system.At last,the fuzzy comprehensive evaluation method was used to determine three major pollutants in the experimental area,and the air quality of the city was comprehensively evaluated according to the concentration of major pollutants.In order to test the performance of the early warning system,which proposed in this paper,this study selects three metropolises in China,Beijing,Shanghai and Guangzhou for experiments.The numerical simulation results show that:(a)the hybrid forecasting model proposed in this paper is an efficient,accurate and scientific forecasting model,which performs well in air pollution forecasting experiments;(b)According to the fuzzy comprehensive evaluation results,the overall air quality of the three study cities in the first half of 2017 were:Beijing-mild pollution,Shanghai-good and Guangzhou-good;(c)singular spectrum analysis and ensemble empirical mode decomposition method smoothed the pollution concentration data,reduces the influence of high frequency signals on the forecasting of air pollution concentrations,and proves the effectiveness of data preprocessing to improve the accuracy of air pollution forecasting by analyzing the value of the forecasting error indexes;(d)In this paper,the latest meta-heuristic optimization algorithm is used to optimize the parameters of L2,1RF-ELM.The results show that the optimization algorithm can improve the prediction accuracy of L2,1RF-ELM model.Based on the large number of literatures,this paper systematically summarizes this study and aims to innovation from the following aspects:At first,as one of the newly proposed heuristic intelligent algorithms,the multi-objective ant lion optimization algorithm(MOALO)is seldom application in the research of air pollution prediction.In this paper,this algorithm is applied to air pollution forecasting,and the effectiveness of multi-objective ant lion optimization algorithm to improve the forecasting accuracy is verified.Second,create an early warning system that considers multiple pollutants.The system consists of two modules:the forecasting module combines the secondary denoising idea,the new meta-heuristic optimization algorithm and a new forecasting method.And in the air quality evaluation module,the fuzzy comprehensive evaluation method was used.Based on the air pollution data of three metropolises of China,this paper verifies the performance of the proposed air quality early warning system.The experimental results show that the proposed early warning system has good performance in three study areas,such as high prediction accuracy and accurate evaluation results.