2022 Volume 3 Issue J2 Pages 488-497
In this study, we constructed air pollution (NO2) forecasting model. The proposed model by LSTM used the future data of weather (temperature and wind speed), traffic volume and background air pollution for input data. The future data of weather were given by mesocale numerical model (MSM) by Japan Meteorological Agency that corrected to the field observation data. The traffic volume were estimated with hourly periodical change patterns of historical data. The background air pollution were estimated with the futer data of weather that calculated in this study. The proposed model greatly improved the forecast accuracy compared to the model using only historical data. We confirmed the effectiveness of introducing future data for input data. Using the proposed model, we were able to estimate the time series change in the daily average value, which is an evaluation index for environmental standards. It was shown that using the estimation results as a decision index can lead to more appropriate provision of detour guidance information for the purpose of avoiding traffic concentration.
