Regional Collaborative Prediction of Air Pollutants Based on CNN-BiLSTM Model

Abstract

Abstract The development of industry has brought serious air pollution problems. It is very important to establish a high-precision and high-performance air quality prediction model and take corresponding control measures. In this paper, based on four years of air quality and meteorological data in Tianjin, China, the relationship between various meteorological factors and air pollutant concentrations are analyzed, the abnormal data are detected and preprocessed. A hybrid deep learning model consisting of convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) is proposed to predict pollutant concentrations, and the effects of three different database input modes are compared. In addition, the Bayesian optimization algorithm is applied to obtain the optimal combination of hyper-parameters for the proposed deep learning model, which makes the model have higher generalization ability. Furthermore, based on the air quality data of multi stations in the region, a regional collaborative prediction method is designed, the concept of strongly-correlated station (SCS) is defined, and the results of collaborative prediction are modified using the idea of SCS, effectively improving the accuracy of prediction.