Research On Air Quality Prediction Based On Ensemble Learning And Nonparametric Method

With the acceleration of urbanization and industrialization in China,there have an increasing consumption of energy which causes most cities in China are faced with air pollution and seriously threatens the health of residents.Therefore,it is an urgent task to construct an effective air quality prediction system to guide public life and industrial production.However,most of the existing studies only focus on the deterministic prediction of air pollutants which is no sufficient to provide useful uncertain information for the prevention and control of air pollution.This paper proposes an air quality prediction model integrating deterministic prediction and probabilistic prediction.The deterministic prediction module aims to build an accurate point prediction model with strong generalization ability.Therefore,six major pollutants are firstly selected as the research object in this paper and cities with different features are determined as study area by clustering method.Secondly,an efficient preprocessing method is used to fully extract the potential characteristics of the original time series by decomposing it into several components.Thirdly,an adaptive incremental extreme learning machine is used to predict each component and the final prediction result is obtained by summing all of these results.Finally,fuzzy evaluation method is established to transform the prediction results into intuitive air quality grade information.The experimental results of eighteen data sets from three cities show that the deterministic prediction model based on decomposition integration established in this paper has higher accuracy and stronger generalization ability than the benchmark model.In order to obtain the uncertainty information of air pollutants,a probability prediction model based on nonparametric method is further constructed in this paper.Firstly,taken point prediction errors as input,a series of quantiles is obtained by through quantile regression neural network and the prediction interval are constructed based on quantiles and point prediction results.Secondly,the probability density function of pollutants is obtained by kernel density estimation method using the series of quantiles as input.Finally,the experimental results of all pollutants show that the probability prediction model proposed by this research can obtain more accurate prediction interval and probability density function of pollutant concentration compared with other benchmark models.