A Particulate Matter Concentration Prediction Model Based on Long Short-Term Memory and an Artificial Neural Network

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Rapid industrialization increases particulate matter (PM) globally, posing risks to human health and ecosystems.
• Accurate prediction of PM concentrations is crucial for timely public health interventions and environmental management.
• Existing PM prediction models often focus on current levels, lacking sufficient advance forecasting capabilities.

Purpose of the Study:

• To develop a predictive model for forecasting future trends (increase or decrease) in PM2.5 concentrations.
• To address the need for advance warning systems for high PM events.
• To specifically model PM2.5 originating from anthropogenic volatile organic compounds.

Main Methods:

• Development of an hourly model selection algorithm integrating Long Short-Term Memory (LSTM) and Artificial Neural Network (ANN) models.
• Utilized historical data for training and validating the predictive models.
• Comparative analysis of the proposed algorithm against standalone LSTM, ANN, and Random Forest models.

Main Results:

• The developed algorithm demonstrated superior performance, achieving a higher F1-score compared to individual LSTM, ANN, and Random Forest models.
• The model effectively predicts the direction of change (increase or decrease) in PM2.5 concentrations.
• The hourly model selection approach enhances predictive accuracy for air quality forecasting.

Conclusions:

• The proposed algorithm offers a more effective method for predicting future regional PM2.5 concentration levels.
• This predictive capability can significantly aid in proactive air quality management and public health advisories.
• The study highlights the potential of advanced machine learning techniques for environmental monitoring and forecasting.