A Novel Air Quality Prediction Method based on GAF and Dense Net

Abstract

Data-driven air quality prediction methods mostly use convolutional neural networks, but the problem of gradient disappearance occurs when the number of network layers increases during network training, and the direct use of air quality-related data as network input results in incomplete feature extraction. Model-based and signal-based air quality prediction methods have problems such as difficult modeling and tedious signal analysis; To overcome these issues, a Gramian Angular Field (GAF) and densely connected convolutional network (Dense Net)-based air quality prediction technique is presented. GAF converts one-dimensional time series of air quality data into two-dimensional images, preserving correlation information between the time series data; the two-dimensional images are fed into Dense Net, which performs feature extraction on the two-dimensional images, improving feature information utilization and achieving accurate air quality prediction. Experiments using data from the UCI air quality data set show that the method can accurately predict future air quality with an MSE error of only 0.0236, demonstrating that the method proposed in this paper can better extract multidimensional time-series feature information and achieve higher air quality prediction accuracy. Further, the method proposed in this paper is expected to be further extended to business big data forecasting, such as stock forecasting, default forecasting, quantitative trading, and other regression problems.