Real-time hourly ozone prediction system for Yangtze River Delta area using attention based on a sequence to sequence model

The Yangtze River Delta (YRD) area is becoming increasingly polluted with ground level ozone, making the prediction of ozone particularly important. This study uses a deep learning approach to forecast ozone concentrations over the YRD region of eastern China. We propose an attention-based sequence to sequence model for ozone concentration prediction, which addresses the dynamic, spatial, temporal, and nonlinear characteristics of multivariate time series data by gated recurrent unit based encoder-forecaster architecture. Through multivariate time series forecasting experiments for ozone concentration, we show that the proposed model is easier and performs better than the weather research and forecasting model with chemistry based forecasting system. Furthermore, we show that the predicted ozone concentration can be matched with the ground truth value under single-timestep and multi-timestep forward forecasting conditions. The experiment results show that the seq2seq model is capable of reliably predicting ozone concentration with a high level of accuracy. The root mean square error of 1- h ozone forecast is 12.40 μg/m³ and the mean absolute error of 1- h ozone forecast is 9.27 μg/m³ on the test dataset. •Predicted ozone concentration over Yangtze River Delta region.•Proposed an attention-based sequence to sequence model.•Proposed model is fast, simple, efficient, and accurate.