基于Wavelet-LSTM模型的北京空气污染物浓度预测  

Forecasting the Air Pollutant Concentration in Beijing Based on Wavelet-LSTM Model

作　　者：刘炳春[1] 来明昭 齐鑫 王辉 LIU Bingchun;LAI Mingzhao;QI Xin;WANG Hui(School of Management,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学管理学院

出　　处：《环境科学与技术》2019年第8期142-149,共8页Environmental Science & Technology

基　　金：天津市教委社会科学重大项目(2017JWZD16)

摘　　要：文章为了达到精准预测北京市空气污染物浓度目的,应用小波分解变换(wavelet transform)和长短期神经记忆网络(long short-term memory,LSTM)相结合的方法,建立Wavelet-LSTM空气污染物浓度预测模型,对北京市6项空气污染物浓度预测。研究首先通过小波分解变换将日空气污染物浓度的历史时间序列分解为不同频率并重新组合为高维训练数据集合;其次使用高维数据集训练LSTM预测模型,重复试验调整参数,获得最优预测模型。研究结果表明,组合模型对于污染物浓度预测比传统LSTM模型的预测精度和稳定性更高。In order to forecast the concentration of air pollutants accurately in Beijing, the wavelet transform and long shortterm neural memory network(LSTM) are combined to establish the Wavelet-LSTM model for predicting the concentration of six air pollutants in Beijing. Firstly, the historical time series of daily air pollutant concentration are decomposed into different frequencies and recombined into high-dimensional training data sets by wavelet transform. Secondly, the LSTM prediction model is trained with high-dimensional data sets, and the parameters are adjusted by repeated experiments to obtain the optimal prediction model. The results showed that the combined model is more accurate and stable than the traditional LSTM model in predicting pollutant concentration.

关 键 词：长短期神经记忆网络 小波变换 空气污染物浓度 预测 

分 类 号：X823[环境科学与工程—环境工程]