A deep learning model for PM2.5 concentration prediction

In recent years, air quality has been deteriorating. It not only seriously threatened people's health but also affected environmental quality and social development. Therefore, it is particularly urgent to build a set of accurate air quality prediction and monitoring mechanisms. PM2.5 concentration has always been one of the key factors in monitoring due to its long transportation distance and strong hazard. This paper proposes a bidirectional nested long-term short-term memory network (BiNLSTM) based on deep learning to predict PM2.5 concentration changes. The training sequence of the BiNLSTM model is composed of a forward nested long short-term memory network (NLSTM) and a backward NLSTM, which allows it to obtain forward and backward features at the current time point. Due to the nested structure of NLSTM units, BiNLSTM can better capture the information of long-term historical data. Experiments show that BiNLSTM can better handle the nonlinearity and instability information in PM2.5 concentration compared to other machine learning models and time series models, and it is better than other models in terms of prediction precision and stability.