Forecasting of PM2.5 Concentration Based on Variational Mode Decomposition and Deep Learning

This repository provides the source code for the study titled "Forecasting of PM2.5 Concentration Based on Variational Mode Decomposition and Deep Learning". The proposed framework integrates VMD-based signal decomposition, entropy-based clustering, and a hybrid TCN-BiLSTM-Attention model for accurate PM2.5 forecasting. The dataset used in this study is the Beijing Multi-Site Air Quality dataset from the UCI Machine Learning Repository. Detailed dataset descriptions are provided in the README file within the data directory. Project Structure:- README.md: overall project description and usage instructions- requirements.txt: list of required Python dependencies- data/: dataset description and access instructions- outputs/: directory for storing generated figures- src/: main source code - run.py: main entry point for executing the full pipeline - pipeline.py: data processing and workflow pipeline - models.py: model architecture definitions (TCN-BiLSTM-Attention) - optimization.py: hyperparameter optimization (GWO and JADE) - utils.py: utility functions - visualization.py: plotting and result visualization All experiments can be reproduced by installing the required dependencies and running:python src/run.py --station tiantan or python src/run.py --station changping