关键词: 关键词: 大气污染现象, 烛台理论, 模式匹配, 卷积神经网络, PM2.5

Abstract: Abstract: Nowadays, the health of residents is seriously threatened by the increasing air pollution phenomenon. PM2.5 (particulate matter of 2.5 micrograms or lesss) has been condemned as the main assassin of the haze weather. Hereinabove statement, when periodically changing data of atmospheric quality is represented by the candlestick chart, different classification features can be displayed. A prediction model(CCA-VGG) of PM2.5 concentration variation direction was put forward based on Candlestick images cluster analysis and VGG extract the characteristics of pollutant transport and diffusion . Firstly, the diffusion model of pollutant concentration was defined as a series of candlestick diagrams. Through it, combined with the unique advantages of convolutional neural network in image recognition, the concentration pattern matching was carried out. In the last step, a candlestick chart with a reversal signal was used to determine the reversal trend that would result. The experimental results had been compared with several advanced methods in different time intervals. Compared with SVM、AlexNet and VGG , the accuracy of the proposed method, increased by 20.97%、16.21% and 7.09%. respectively. PM2.5 trend characteristics could be extracted according the proposed method, and the method based on the Candlestick diagram to predict the future change of pollutant concentration was verified to be effective.

Key words: Keywords: air pollution phenomenon, candlestick theory, pattern matching, convolutional neural network, PM2.5