An Empirical Analysis of the Long Short Term Memory and Temporal Fusion Transformer Models on Regional Air Quality Forecast

Real-time meteorological observation systems are of great importance in practical engineering use. Monitoring environmental data 24/7 not only provides a healthy and green living environment but also enables the prediction of future living conditions based on existing data, thus acting as a detection and warning. At this stage, weather monitoring stations usually use conventional detection, which often results in prediction deviations, thus bringing a more significant impact on weather detection in the whole region. Therefore, the prediction strategy of meteorological factors should be further strengthened and better defined. This study aims to fill this gap and compare the advantages and disadvantages of linear regression time series models, long and short-term memory models, and Temporal Fusion Transformer models theoretically and experimentally. Ultimately, we seek to provide researchers with information to aid them in choosing a prediction algorithm based on their available data volume.