Optimization Application On Integrated Forecast Of Environmental Meteorological Model Based On Improved Neural Network

The air quality numerical model forecast is mainly to simulate the process of atmospheric operation through the meteorological model,and analyzes the interaction of various elements in the atmosphere to predict the concentration of various pollutants,so as to predict the air quality.However,due to the uncertainty of the initial meteorological field conditions in the numerical prediction,the error of the model itself and the chaotic characteristic of atmosphere,there are some deviations in the numerical model prediction.Multi-model integration technology is a highly efficient method to solve the problem of uncertainty in numerical prediction.It can comprehensively utilize the advantages of various environmental meteorological models to reduce the forecast error of model.Based on the analysis of the air pollution situation in Beijing-Tianjin-Hebei region,this paper studies the multi-mode integrated prediction technology based on neural network algorithm,enriches the methods of deep learning in air quality prediction,and provides scientific reference for pollution prevention and control and traffic travel.The main conclusions are as follows.The air pollution situation in Beijing-Tianjin-Hebei region is analyzed in detail from three aspects:air quality index（AQI）,primary pollutant and pollutant concentration.This paper explores the spatial distribution characteristics of the annual average value of AQI and the four-section change rule of the seasonal average value,and makes statistics on the frequency of pollution occurrence at various levels,the proportion of days of primary pollutants,and the monthly change rule of major primary pollutant days,and studies the seasonal and daily change characteristics of the concentration of pollutants in key cities.The experimental results show that,the air quality is high in the south and low in the north,and the pollution is serious in autumn and winter.PM_2.5is the main primary pollutant,and it is a persistent and regional air pollutant.Based on the numerical models,including Rapid Refresh Multi-scale Analysis and Prediction System-CHEM and CMA Unified Atmospheric Chemistry Environment,Multi-model integrated forecasting model based on recurrent neural network and long-short-term memory neural network has been constructed.The model is optimized by parameter sensitivity experiment,the prediction results are evaluated and analyzed,and real-time correction and trend analysis modules are added to improve the model structure and simulate typical heavy pollution case.The experimental results show that,the integrated forecasting model improves the forecast result of the numerical models effectively and reduces the prediction error.