A hybrid prediction model for PM_2.5 concentration based on high-frequency and low-frequency IMFs with EMD...

Area of Science:

• Environmental Science and Engineering
• Atmospheric Science
• Data Science and Machine Learning

Background:

• North China's heavy industry causes severe air pollution, with Particulate Matter (PM2.5) as a primary pollutant contributing to haze.
• High PM2.5 concentrations significantly disrupt daily life and industrial activities, necessitating accurate prediction for mitigation.
• Effective air pollution control nationwide relies on robust regional prediction models, particularly for areas like North China.

Purpose of the Study:

• To develop and validate an advanced hourly prediction model for PM2.5 concentrations in six major North China cities.
• To enhance prediction accuracy by leveraging the time-frequency characteristics of PM2.5 time-series data.
• To provide a practical tool for air pollution prevention and control strategies in the region.

Main Methods:

• Utilized hourly PM2.5 data from Beijing, Tianjin, Shijiazhuang, Taiyuan, Jinan, and Zhengzhou.
• Employed Empirical Mode Decomposition (EMD) to decompose non-linear, non-stationary PM2.5 data into intrinsic mode functions (IMFs) and residuals.
• Integrated Long Short-Term Memory (LSTM) for high-frequency subsequence prediction, Autoregressive Integrated Moving Average (ARIMA) for low-frequency subsequence prediction, and Support Vector Machine (SVM) for final result integration (Hybrid-EMD_HL model).

Main Results:

• The Hybrid-EMD_HL model demonstrated significantly improved PM2.5 concentration prediction accuracy compared to single prediction models.
• The model achieved a directional indicator (DA) greater than 0.69 in multiple experiments, indicating superior direction prediction capabilities.
• The hybrid approach effectively mines data information and captures inherent time-series characteristics, enhancing model adaptability.

Conclusions:

• The proposed Hybrid-EMD_HL model offers a substantial advancement in hourly PM2.5 forecasting for North China.
• The model's enhanced prediction accuracy and directionality are crucial for effective air quality management and public health protection.
• This research provides a valuable methodology for predicting air pollutants in complex urban environments.