Research On Fine Particulate Concentration Prediction Model Based On Correlation Analysis And Deep Learning

For the past few years,with the increase of population and the acceleration of urbanization and the increase of various industrial and transportation activities,concentrations of fine particles in air pollution in a large number of cities around the world are increasing.The main pollutants in air pollution include fine particulate matter such as PM2.5 and PM10,and its concentration is directly related to the degree of air pollution.Fine particulate concentration prediction is an effective way to provide early warning of air pollution and support clean industrial production,but existing methods have a weak ability to capture long-term dependencies and complex relationships from time series data of air pollution,and most of the fine particulate concentration prediction models are based on single-site data,but because the change of fine particulate concentration at the target monitoring site will be affected by the pollutant concentration of adjacent monitoring sites,the correlation between the data needs to be analyzed.In order to accurately and efficiently predict massive fine particulate concentration data,this thesis takes multi-site air pollution time series data as the research object,establishes a corresponding fine particulate concentration prediction model,and constructs a prototype system for fine particle concentration prediction based on correlation analysis and deep learning.The main tasks are as follows:1.In view of the fact that the existing time-series based air pollutant concentration prediction models may have long time lags and do not fully consider the correlation between monitoring sites and spatially,a method for predicting fine particulate matter concentrations based on correlation analysis is proposed.In this method,the three sites with the highest correlation with pollutant concentration in the target site were screened out by the Pearson coefficient,and then the air pollutant data of these sites were fused with the data of the target site,and finally the fine particulate matter concentration prediction was predicted by combining the corresponding deep learning prediction model.The experimental results show that the proposed prediction method for fine particulate matter concentration based on correlation analysis improves the prediction accuracy of PM2.5 and PM10 fine particulate concentration in the next 6 hours.2.Aiming at the problem that the existing deep learning prediction model has a complex structure and cannot effectively use the past and future characteristics in a specific time range,a fine particulate matter concentration prediction model based on correlation analysis and BiGRU(Bi-directional Gated Recurrent Unit)is proposed.The model first selects the three sites with the strongest correlation with the prediction target of the target site through the aforementioned correlation analysis based on the correlation analysis prediction method,and then uses the multivariate time series data composed of the air pollutant data related to these sites and the relevant data of the target site as the data used for the training of the BiGRU model.The experimental results show that the proposed prediction model of fine particulate concentration based on correlation analysis and BiGRU can effectively reduce the error of PM2.5 and PM10 fine particulate concentration prediction results compared with other deep learning prediction models.3.In summary,the proposed fine particulate concentration prediction model is designed and implemented based on the browser and server architecture,and uses the front-end and back-end separation method to design and implement the fine particulate concentration prediction prototype system.This system realizes the two fine particulate concentration prediction models proposed earlier,and realizes the real-time fine particulate concentration prediction of the current target site.In addition,the system also realizes the functions of dynamic data display,background operation and maintenance personnel information management,log management,large-screen display of fine particle concentration data,etc.,which can more efficiently assist background management personnel to understand the current changes in air quality in a timely manner,and can also test the feasibility of the current fine particle concentration prediction model.