Air quality historical correlation model based on time series.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Science

Background:

• Accurate air quality prediction is crucial for human health and social development.
• Existing models often neglect temporal dynamics and emission-pollutant relationships.
• Air Quality Index (AQI) is a time series with inherent temporal features.

Purpose of the Study:

• To develop an improved air quality prediction model using historical data and pollutant emissions.
• To leverage temporal features of AQI data more effectively.
• To enhance prediction accuracy and stability compared to traditional methods.

Main Methods:

• Utilized Gaussian Hidden Markov Model (GHMM) for time series analysis.
• Employed traversal method for optimal hidden state selection in GHMM.
• Applied Multi-day Weighted Matching and Fixed Training Set Length for GHMM optimization.
• Implemented direct and indirect prediction modes for AQI forecasting.
• Integrated historical correlation model with a prior meteorological correlation model.

Main Results:

• The optimized GHMM with indirect prediction mode showed improved accuracy (MAE=13.59, RMSE=17.59).
• Integrating historical and meteorological correlation models further boosted prediction accuracy (MAE=11.59, RMSE=14.87).
• GHMM demonstrated strong temporal feature analysis capabilities, enhancing prediction stability.

Conclusions:

• The Gaussian Hidden Markov Model significantly improves air quality prediction accuracy and stability by capturing temporal dynamics.
• Combining historical emission data with meteorological factors leads to more robust and accurate air quality forecasts.
• This integrated approach offers a more comprehensive method for understanding and predicting air quality.