A novel hybrid prediction model for PM_2.5 concentration based on decomposition ensemble and error correction

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Particulate Matter (PM2.5) concentration is a key indicator of air pollution.
• Accurate PM2.5 prediction is crucial for urban air quality management.
• The nonlinear nature of PM2.5 data presents significant prediction challenges.

Purpose of the Study:

• To develop a novel hybrid model for accurate PM2.5 concentration prediction.
• To address the nonlinear characteristics of PM2.5 data using advanced decomposition and optimization techniques.
• To enhance urban air monitoring and control systems.

Main Methods:

• Proposed an improved Variational Mode Decomposition (IVMD) method to decompose PM2.5 data into intrinsic mode functions (IMFs).
• Developed a hybrid Cuckoo Search (CS) and Chimp Optimization Algorithm (ChOA) to optimize an outlier-robust extreme learning machine (ORELM), creating the ChOACS-ORELM model.
• Implemented an ensemble-based error correction (EC) model to further refine prediction accuracy.

Main Results:

• The proposed IVMD-ChOACS-ORELM-EC model demonstrated high prediction accuracy for PM2.5 concentrations.
• Achieved a correlation coefficient of 0.9999 between predicted and actual PM2.5 values across four Chinese cities.
• The model significantly outperformed existing comparison models in prediction performance.

Conclusions:

• The novel IVMD-ChOACS-ORELM-EC model offers a robust and accurate solution for PM2.5 concentration prediction.
• This approach effectively handles the complexities of nonlinear air pollution data.
• The findings support the application of this model for improved urban air quality management.