Forecasting mechanism for energy transition in Chinese cities based on configuration perspective and TCN-FECAM-MTransformer

We propose a modeling framework that incorporates a combination of configuration effects, fixed effects, and deep learning models to explore the impact of non-high-carbon emission pathways on urban energy transition. In the configuration identification stage, the dynamic qualitative comparative analysis (DQCA) method extracts the configuration pathways of the ‘economy-energy-environment’ complex system that lead to both high and non-high carbon emissions. In the analysis stage, the fixed effects model identifies valuable configurations and addresses temporal heterogeneity. In the prediction stage, we employ a deep learning model based on a spatio-temporal convolutional network and a frequency-enhanced channel attention mechanism—TCN-FECAM-Mtransformer—to enhance prediction accuracy and robustness. The Shapley additive explanation (SHAP) method quantifies the contributions of configuration conditioning variables and other factors to the predictions made by the TCN-FECAM-Mtransformer model; the results show that there are six pathways within the ‘economy-energy-environment’ complex system for realizing non-high-carbon emissions. Fixed effects models significantly improve predictive accuracy, with the TCN-FECAM-Mtransformer model outperforming traditional machine learning and deep learning models. SHAP analysis indicates that population size (32.90 %), forest cover (32.50 %), energy production (6.30 %), and energy consumption (5.10 %) have significant impacts on the energy transition. The framework demonstrates high accuracy, making it a valuable reference for formulating relevant policies.