Application Of Tensorflow-based Recurrent Neural Network Model In Shanghai Air Quality Prediction

In recent years,air pollution has deteriorated,seriously hampering the social development of our country.Effective monitoring of air quality and accurate prediction of air pollutant concentrations are important for China's economic development and national health.In addition,because the air quality is time-series,it is affected by many difficult and non-deterministic factors.The efficiency and accuracy of the commonly used regression prediction methods are relatively low,while the cyclic neural network model can effectively utilize the ability of long-distance dependent information in time series data to perform relevant predictive analysis on such data.This paper will attempt to use the hourly mean data of the Shanghai Air Quality Monitoring Station to build a dynamic prediction model based on Google's open source Tensorflow machine learning framework and a long-short-term memory?LSTM?cyclic neural network,using timing-based The back-propagation?BPTT?algorithm gradually updates the network weights for network training,It is expected to establish the₂,₂,₃,CO,₁₀ and_2.5 air pollutants of the hourly average level of the six pollutants concentration prediction model.The main work and contributions of this paper are as follows:The gray correlation analysis method and principal component analysis method are used to screen the input variables in the atmospheric pollutant concentration prediction model respectively,and select the pollutants that have a great influence on the concentration of the prediction monitoring factors and establish relevant prediction models.Then based on the test data,the prediction models are evaluated using the evaluation indicators such as MAE,MSE,and²for the two prediction models.It is finally verified that the circulatory neural network based on Tensorflow training can be applied to the concentration prediction of air pollutants more accurately.