A new combination model using decomposition ensemble framework and error correction technique for forecasting hourly PM2.5 concentration | JoVE Visualize

Area of Science:

Environmental Science
Data Science
Artificial Intelligence

Background:

Particulate Matter (PM2.5) poses significant health risks, necessitating accurate concentration forecasting.
Existing forecasting models often struggle with complex data patterns and achieving high accuracy.
The need for robust models that can handle the non-linear and dynamic nature of PM2.5 concentrations is critical.

Purpose of the Study:

To develop and validate a novel hybrid model (AIVMD-RBF-IOWA-LSTM-EC) for precise hourly PM2.5 concentration forecasting.
To enhance forecasting accuracy and robustness by integrating adaptive decomposition, ensemble learning, and error correction techniques.
To address limitations in existing combination models, particularly when individual model complementarity is poor.

Main Methods:

Agreement Index Variational Mode Decomposition (AIVMD) adaptively determines the mode number for decomposing PM2.5 data into simpler Intrinsic Mode Functions (IMFs).
Radial Basis Function (RBF) neural networks and Long Short-Term Memory (LSTM) networks are employed for forecasting each IMF component.
An RBF-based error correction model and Induced Ordered Weighted Averaging (IOWA) operators are utilized for combining predictions and optimizing weight allocation, ensuring complementary forecasting.

Main Results:

The proposed AIVMD-RBF-IOWA-LSTM-EC model demonstrated superior performance in short-term PM2.5 concentration forecasting compared to traditional models.
The adaptive decomposition and ensemble framework effectively reduced data complexity and improved prediction accuracy.
The error correction and IOWA operator integration enhanced model robustness and handling of poor single-model complementarity.

Conclusions:

The AIVMD-RBF-IOWA-LSTM-EC model offers a significant advancement in accurately forecasting hourly PM2.5 concentrations.
The hybrid approach effectively combines the strengths of decomposition, ensemble learning, and error correction for improved air quality prediction.
This model provides a valuable tool for environmental monitoring and public health protection strategies.