基于CNN-LSTM的兰州市PM_2.5浓度预测研究

Prediction of PM_2.5 Concentration in Lanzhou City Based on CNN-LSTM

摘要: 对PM_2.5浓度的精准预测能够减少很多健康问题的出现，从而提高人们的生活质量。现有的研究大多建立机器学习模型或单个的深度学习模型来预测PM_2.5浓度，且往往只关注单要素的影响而忽略多个要素的综合影响。基于此，本文建立CNN-LSTM组合模型，使用污染物浓度及气象数据对未来短期单日PM2.5浓度做预测，并将结果与单个LSTM模型作对比。结果证实，相比于单个的LSTM模型而言，CNN-LSTM组合模型预测效果更佳。

Abstract: Accurate prediction of concentration can reduce the occurrence of many health problems and im-prove people’s quality of life. Most of the existing studies have established machine learning models or single deep learning models to predict concentration, and often only focus on the impact of single factor and ignore the comprehensive impact of multiple factors. Based on this, this paper adopted the method of combining one-dimensional convolution and long-term and short-term network (CNN-LSTM), using the pollutant concentration and meteorological data to predict the future short-term single-day concentration, and the results were compared with a single LSTM model. The results showed that the combined CNN-LSTM model has better prediction effect than the single LSTM model.