A Deep Learning-based Air Quality Index Prediction Model Using LSTM and Reference Stations: A Real Application in Taiwan

Air pollution has attracted significant attention as a public concern, prompting government efforts towards air quality monitoring. With the accumulation of data, the feasibility of utilizing artificial intelligence (AI) to aid in air quality prediction has grown. Air quality prediction plays a pivotal role in policy formulation and environmental management. We propose a neural network model framework constructed using LSTM (Long Short-Term Memory) model and fully connected model. It employs pollutant and meteorological data from target stations, neighboring stations, and outpost stations as feature factors. By utilizing the previous 24 hours monitoring data, it predicts the hourly air quality index (AQI) for the next 12 hours. Experimental results demonstrated the effectiveness of this approach in enhancing prediction accuracy and providing early forecasts of long-range transport of air pollutants. This prediction model is applied within the Taiwan Air Quality Monitoring Network [1] and the Environment Info Push app [2] both maintained by the Ministry of Environment of Taiwan (MOENV), allowing citizens to access information of environment at any time and receive activity recommendations based on real-time monitoring values.