Double decomposition and optimal combination ensemble learning approach for interval-valued AQI forecasting using...

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Accurate air pollutant concentration forecasting is crucial for predicting environmental risks.
• Existing models struggle with massive, redundant hourly data or lack detail in daily averages.
• Interval-valued time series present unique challenges due to nonlinearity and irregularity.

Purpose of the Study:

• To propose a novel double decomposition and optimal combination ensemble learning approach for interval-valued air quality index (AQI) forecasting.
• To address the limitations of existing models in handling complex, high-volume air quality data.
• To improve the accuracy and reliability of environmental risk predictions.

Main Methods:

• A double decomposition strategy: first, seasonal decomposition based on the Chinese calendar, followed by reconstruction of seasonal series.
• Second decomposition using Bivariate Empirical Mode Decomposition (BEMD) to break down interval-valued signals into intrinsic mode function (IMF) components.
• Interval Multilayer Perceptron (iMLP) to model both lower and upper bounds of AQI, combined via an optimal ensemble method.

Main Results:

• The proposed model demonstrated superior forecasting performance across different datasets and forecasting horizons.
• Empirical studies confirmed the model's effectiveness in handling complex interval-valued AQI time series.
• The approach significantly outperformed other considered models in AQI prediction.

Conclusions:

• The double decomposition and ensemble learning approach offers a robust solution for interval-valued AQI forecasting.
• This method effectively reduces data complexity and captures fluctuations for more accurate environmental risk assessment.
• The proposed model provides a significant advancement in air quality forecasting technology.