Urban Multi-energy Consumption Prediction and Carbon Emission Collaborative Monitoring Model Based on GAN and Random Forest

Against the backdrop of global energy transition and carbon neutrality goals, this study proposes an urban multi-energy carbon emission monitoring framework integrating Generative Adversarial Networks (GAN) and Random Forest (RF). The framework addresses challenges in traditional models, such as fragmented data and insufficient nonlinear modeling, through a three-layer architecture: dynamic data augmentation via GAN, multi-energy coupling modeling based on RF, and carbon emission mapping. By leveraging GAN’s data generation capability and RF’s ensemble learning advantages, the framework forms a closed-loop system for “energy consumption forecasting—carbon emission accounting—policy simulation”. Case studies in Guangdong Province demonstrate that the model achieves significant improvements in prediction accuracy for raw coal, diesel, fuel oil, and liquefied petroleum gas, with R2 values exceeding traditional methods by 15–35%. The framework also quantifies the correlation between energy structure and carbon emissions, providing a scientific basis for formulating low-carbon policies.