CEEMD-Subset-OASVR-GRNN For Ozone Forecasting

Ozone is one of the six major pollutants in air,when the concentration of ozone in the atmosphere is too high,it will lead to the deterioration of the ecological environment and adversely affect human health.In recent years,the concentration of ozone has been increasing.There are two main reasons,one is the increase in pollutants emitted by man-made activities,and another is the weather.The stronger the sun’s rays,the more ozone is produced.As people pay more and more attention to the degree of ozone pollution,it is extremely important for researchers to predict ozone concentration in a timely and effective manner.At the same time,ozone forecasting is of great significance for the prevention and control of air pollution.Relevant departments can prevent the harm to the human body and the environment caused by high ozone concentration based on effective ozone forecasts.Therefore,it is meaningful to predict ozone efficiently.This paper proposes the CEEMD-Subset-OASVR-GRNN model to predict ozone concentration.According to the literature research,it can be found that combined forecasting is a popular forecasting method in the field of air pollution forecasting.However,there is a problem of blind combination,which is to combine individual models without selection.In order to solve the above problems,this paper proposes the CEEMD-Subset-OASVR-GRNN model to study the selections of individual model and the number of individual models.The model is based on three individual model selection methods:MSE ranking,factor score and systematic clustering.In addition,it combines complementary ensemble empirical mode decomposition（CEEMD）,optimization algorithm（OA）,generalized regression neural network（GRNN）,support vector regression（SVR）to establish a combined forecasting model of average daily ozone concentration.Specifically,CEEMD is used to decompose the original time series data into three intrinsic mode functions().PSO-SVR,PSOGSA-SVR,GWO-SVR and GRNN are employed to model and predict,and the prediction results are combined to establish 100 individual models.The individual model set is obtained through the individual model selection methods,and then the CEEMD-Subset-OASVR-GRNN model is established.In this study,the ozone day data of two cities in Xiamen and Harbin are selected to verify the reliability of the proposed model.The experimental results shows that individual models cannot be combined blindly in combination prediction,and the selection of individual models by systematic clustering method can effectively improve the prediction accuracy of combined model.At the same time,the number of individual models is not the more the better.Generally,5-6 individual models are the best.Compared to the optimal individual model among 100 individual models,CEEMD-Subset-OASVR-GRNN model has smaller prediction errors and stronger generalization ability.Taking Xiamen’s ozone time series as an example,compared to the optimal individual model among 100 individual models,the prediction accuracy of CEEMD-Subset-OASVR-GRNN model is improved by 4.978%.