[PM₁₀ Concentration Forecasting Model Based on Wavelet-SVM]

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Context:

• Taiyuan faces significant air pollution challenges, primarily from PM10, due to its status as a heavy industrial center reliant on coal.
• Accurate prediction of PM10 concentration is crucial for effective air pollution prevention and emergency response strategies.
• The complex and multifaceted sources of PM10, including industrial emissions, vehicle exhaust, and fugitive dust, make accurate source determination challenging.

Purpose:

• To develop an efficient and accurate forecasting model for PM10 concentrations using only historical time-series data.
• To overcome the limitations of traditional Support Vector Machine (SVM) models in handling simple data structures and incomplete information for time-series forecasting.
• To enhance the generalization ability of PM10 forecasting models by incorporating advanced data processing techniques.

Summary:

• A novel wavelet-Support Vector Machine (SVM) model was developed by decomposing one-dimensional PM10 time-series data into high-dimensional representations using wavelet transform.
• This wavelet-SVM model utilizes the decomposed low and high-frequency series as input features, enhancing the model's ability to process complex temporal patterns.
• The model was trained and validated using PM10 concentration data from four monitoring stations in Taiyuan.

Impact:

• The wavelet-SVM model demonstrated significantly higher accuracy in PM10 concentration prediction compared to traditional SVM methods.
• The enhanced model excels at capturing abrupt changes or mutational points in PM10 concentrations, providing more reliable data for atmospheric pollution warnings.
• The study revealed inherent patterns within PM10 concentration time series, leading to improved prediction accuracy for concentration variations and offering valuable insights for air quality management.