School of Science Department of Mathematics 45 Improvement of the Air Quality Forecast by Using Deep-learning Technique Supervisor: FUNG, Jimmy Chi Hung / MATH Student: LIU, Haoran / MATH-SF Course: UROP1100, Summer This study aims to understand and improve the application of Long Short-Term Memory (LSTM) frameworks for air quality prediction. Due to rapid industrialization and economic development, air pollutants have significantly impacted human health and social stability. Accurate forecasting of these pollutants can help address these threats. LSTM frameworks, which could track historical results, have been identified as a suitable solution for this challenge. However, some problems, such as weak correlation between future air pollution conditions and past ground observations still exist. This study focuses on understanding LSTM and related forms, with the goal of refining their structure to develop more accurate and efficient machine learning-based models for air quality prediction. Improvement of the Air Quality Forecast by Using Deep-learning Technique Supervisor: FUNG, Jimmy Chi Hung / MATH Student: LIU, Jianmeng / COSC MENG, Zeyuan / COMP Course: UROP1100, Fall UROP1100, Fall Deep learning frameworks can effectively predict air pollution data for individual sites by decoding time series data. Previously exists a method in Sun et al. (2022) used an LSTM network combined with a broadcasting layer, which incorporates a learnable weight decay parameter designed for point-to-area extension. We tried some improvements on that model by combining more features into the broadcasting layer, as well as making the weight-decay parameter chronologically sensitive. To validate the proposed deep-learning framework and keep consistency, PM2.5 and O3 forecasts for the next 48 h were obtained using WRF-CMAQ and ground observation data as the inputs. Although the results were not as good as the previous experiment, they allowed us to find a new direction for further research. Extension of the original model from the GBA area to the whole China and surrounding areas was also attempted, and we spent a lot of time processing the observation as well as WRF-CMAQ prediction data. Up to now, the data processing is basically completed, and we leave the comparison of the effect between models in future works. Improvement of the Air Quality Forecast by Using Deep-learning Technique Supervisor: FUNG, Jimmy Chi Hung / MATH Student: TIAN, Hangyu / SSCI Course: UROP1100, Summer In recent years, environmental pollution has become a new major problem faced by human beings. How to reasonably predict and control pollutants is very essential for the sustainable development of cities. The formation and the destruction of PM2.5 are complicated it is hard to predict the distribution of the pollutant. In this study, we will combine the meteorological and air pollutant data from observation stations with a deep learning model to predict the concentration of PM2.5 over a wider area in the future 48h. By using a more advanced model and AI approach, our study provides more accurate data as well as the ability to predict pollutant concentrations over a wider area.
RkJQdWJsaXNoZXIy NDk5Njg=