Improvement Of Air Ouality Forecast Performance Based On Machine Learning Algorithm

The 19th National Congress of the Communist Party of China sets the goal of winning the battle to protect our blue skies.Because the factors affecting air pollution are complex,the accurate forecast of air quality is the necessary condition to solve air pollution.In this context,according to the characteristics of time series of air quality and meteorological data,this study proposed a long short memory(LSTM)algorithm based on time series,combining with the WRF meteorological numerical model of air quality forecast model.In order to improve the accuracy of air quality forecast,the model fully considers the relevant factors of air quality(such as meteorology and environment).This paper analyzes the temporal and spatial characteristics of the air quality observation data of Zhengzhou in 2016 and 2019,and the results show the air quality in Zhengzhou City in 2019 was better than that in 2016.Except for the poor air quality in winter,the air quality in other seasons changed at an annual average level.The study shows that the air quality of scenic spots is obviously better than that of central urban areas.The whole city presents a decreasing trend from northeast to southwest,and the urban areas show the characteristics of spreading to the periphery.This paper analyzes the correlation between meteorological factors and the concentration of pollutants in historical observation data,and the temporal and spatial correlation among various air quality monitoring stations in Zhengzhou.Based on the observation data,meteorological feature parameters of the MCIP module in the WRF model is added to the hybrid model.Through gray correlation analysis and expansion factor method test,the input feature library of the hybrid forecast model is obtained,and this input feature library is used for model training and forecasting.This paper compares and analyzes the predicted results of the two models based on the measured pollutant concentration data from the air quality monitoring stations,and evaluates the model based on the conventional model evaluation indicators.The results show that the simulated and predicted data of the two models are compared with the actual measurement.The forecasting accuracy of the AQI of the WRF-CMAQ model and the WRF-LSTM model are 69.1%and 75.9%respectively.The forecast accuracy of the 6 pollutants is improved except for the SO2 forecast.WRF-LSTM models are more than 10%higher than WRF-CMAQ.In conclusion,the prediction and analysis precision of the hybrid model clearly proposed in this paper is better than that of the traditional prediction model,which can provide a certain technical reference and basic theory application for improving the accuracy of Henan air index weather forecasting system software and formulating energy saving and emission reduction countermeasures in the future.