Research On Urban Air Quality Forecast Based On ARMA-BP Neural Network

In recent years,with the rapid development of modernization and social economy,the problem of urban air pollution has become increasingly prominent.Air pollution not only destroys the environmental ecosystem,but also poses a serious risk to human health.The research on the prediction of urban air quality development trend has become a very important topic,especially the accurate prediction of possible heavy pollution disasters in the short term future,which can minimize the impact of severe pollution.In previous studies of air quality prediction,a single model is usually used for analysis,but the single model often fails to fully extract the implicit information in the historical air quality data,which had the disadvantages of low prediction accuracy and poor stability.In order to overcome the shortcomings of a single model in some aspects,this paper proposes to combine the BP neural network and the ARMA model to build an ARMA-BP neural network combined model to predict the air quality index(AQI)of Xi’an city in the short-term,and adopt a genetic algorithm(GA)Improve the stability of the combined model.This article first crawled the historical AQI data of Xi’an from 2014 to 2020,and analyzed the change trend of the data set and its correlation characteristics.AQI data series is more complex,and it is a time series with linear and nonlinear relationships coexisting,thus this paper divides the AQI series into linear and nonlinear parts for research.In this paper,a single BP neural network model is established on the sample data to predict air quality in a short-term.The average relative error of the prediction results is 11.10%,which shows that BP neural network has certain prediction ability for air quality,but has the disadvantage of large error.By fully combining time series ARMA with BP neural network,the advantages of each can be fully utilized.In this paper,a combined ARMA-BP neural network model is established to make short-term forecasts of AQI in Xi’an city.The process is as follows:first use the ARMA model to establish a time series model to predict the linear subjects in the AQI sequence.The relative error of the prediction result is 14.28%.Although the prediction result of this model is in line with the trend of the sample measured values,compared with the BP neural network prediction model the error is larger.The BP neural network is then used to fit the residual part,that is,the nonlinear part of the AQI series,and the final output consists of the superimposed linear and nonlinear parts of the resulting prediction.The emulation results of the same sample data show that the average relative error of ARMA-BP combination model is 6.32%,and the prediction accuracy of the combined model is higher compared with that of the single model.Finally,the prediction results of air quality by the combined ARMA-BP model were analyzed,and it was found that the prediction effect of the combined model was not very stable,which was caused by the defect that the BP neural network was easy to fall into local extremes,and thus the use of genetic algorithm GA was proposed to optimize the parameters of the BP network in the combined model.Through the simulation and comparison experiments of multiple model algorithms,it is found that the average relative error of the prognostication results of the combined prognostication algorithm based on the GA improved ARMA-BP neural network is 2.47%,which significantly improves the prognostication accuracy compared with the single model and the original combined model,which fully reflects the feasibility and accuracy of the GA improved ARMA-BP combined model prediction.