基于LSTM网络的PM2.5浓度预测

A PM2.5 Concentration Prediction Method Based on LSTM Networks

近年来PM2.5问题引起了人们的广泛关注。提出一种基于长短期记忆(LSTM)网络的PM2.5浓度预测方法。通过获取过去24 h的PM2.5、PM10、CO、SO_(2)、NO _(2)、O_(3)等环境数据的变化情况,构建一个基于LSTM门控循环单元的深度学习网络,用于未来24 h的PM2.5浓度预测。通过对比不同网络层数和不同LSTM细胞数量对预测效果的影响,选取最优的网络结构,实现对区域(南京市)未来24 h PM2.5浓度的预测。预测结果表明:基于LSTM网络的PM2.5浓度预测方法是可行的,在一定程度上提高了对真实值的拟合效果,反映PM2.5浓度的变化趋势。

In recent years,the issue of PM2.5 has aroused widespread concern.This paper proposes a PM2.5 concentration prediction method based on long short-term memory(LSTM)networks.By acquiring the changes of environmental data such as PM2.5,PM10,CO,SO_(2),NO_(2)and O_(3)in the past 24 hours,a deep learning network of gated cyclic units of LSTM networks has been constructed for the prediction of PM2.5 concentration in the next 24 hours.By comparing the influence of different network layers and different LSTM cell numbers on the prediction effect,the optimal network structure is selected to realize the prediction of PM2.5 concentration in the region(Nanjing)in the next 24 hours.The analysis and prediction results show that the PM2.5 concentration prediction method based on LSTM networks is feasible,and to a certain extent,the true value is fitted to reflect the variation trend of PM2.5 concentration.