Prediction Of Chongqing Air Quality Index Based On Combination Weight Model

In recent years,China’s economy has changed from high-speed development to high-quality development.The country is committed to building an ecological civilization and a beautiful China,and has achieved remarkable results.To achieve sustainable development,the construction of ecological civilization is still one of the key tasks during the“14th Five-Year Plan”period,and air pollution has become one of the hot issues of common concern in urban and social management.From a macro perspective,research on air quality can help provide support for precise pollution control.From a microscopic point of view,the air quality index（AQI）,as an indicator to measure the quality of air conditions,is helpful for people to judge the environmental conditions.it is of great practical significance to study and analyze it.This paper collected monthly data（96 items in total）of Chongqing’s air quality index（AQI）and six major pollutants（PM2.5,PM10,NO2,SO2,CO,O3）from January 2014 to December 2021.Among them,the training model uses the monthly data from January 2014 to May 2021（89 in total）,and the test model uses the remaining months’data（7 in total）.Mainly use SPSS software and R for modeling analysis.Firstly,considering whether the air quality index in Chongqing has the distribution characteristics in time,the system clustering method is adopted and SPSS is used to draw the pedigree diagrams of four different interclass distances.The conclusion is consistent:the air quality index of Chongqing presents obvious seasonal effect.Combined with the descriptive analysis of the air quality index,the conclusion is that the air quality of Chongqing is the worst in winter,the air quality is better in summer and autumn,and the best in spring.Secondly,the prediction of air quality in Chongqing is a focus of this study,and a single model often only contains a certain aspect of information and cannot extract sufficient information.The emergence of the combined model just solves these problems.Combined forecasting model is a method to maximize the collection of information,fully mine useful information in the data,and improve the prediction accuracy of the model.In terms of model selection,considering that the data has the linear characteristics and nonlinear characteristics,this paper adopts two commonly used single models-ARIMA model and BP neural network model..In terms of weight determination,this paper adopts the least square weight method and the MAPE method.Combined comprehensive weight method.The first step is to build a single model.Using the time series model established by training data,the model result obtained is ARIMA（1,1,1）（0,1,1）_[12].A three-layer BP neural network model was established with the structure of 6-10-1.Two models were used for prediction respectively.The second step is to establish a single weight combination model.MAPE method and least square method were used to determine the weight values of the respective combined models,and the two sets of weight values were used to combine the two single models to get two groups of new predicted values.The third step is to use the combination of the least squares weight method and the MAPE method to determine a new comprehensive weight,use this new weight value to weight the forecast results of the time series model and the BP neural network model to obtain a new set of forecast values.Finally,the best model is selected by comparing the accuracy indicators that measure the model.The results show that the prediction accuracy of the single weight combination model is higher than that of the single model,and the combined weights further improve the prediction accuracy of the combination model compared with the single weight.To sum up,this paper selected ARIMA seasonal model,BP neural network model,MAPE weight combination model,OLS weight combination model and MAPE-OLS comprehensive weight model to predict the air quality index of Chongqing,and analyzed and compared the prediction accuracy of each model.It is concluded that the prediction effect of MAPE-OLS comprehensive weight model is better than the other four models.In this paper,it solves the shortcomings that time series models are more suitable for short-term forecasting and BP neural network is more suitable for long-term forecasting,and the linear features and nonlinear features extracted from the original sequence are effectively combined.