A High-Frequency Monitoring Method for Urban Carbon Emissions Based on Vertical Federated Deep Learning and Multi-Source Heterogeneous Data Fusion Methods

Abstract

Carbon emissions are the main culprit of global warming. Accurate carbon emission forecasting helps government departments formulate effective carbon emission reduction policies and helps the carbon emission market develop orderly. Implementing an accurate and effective carbon emission monitoring model requires the collaboration of many parties because carbon emission-related data involves many sectors and industries. However, for the relevant characteristics of carbon emission monitoring, due to the different collection and storage standards of various departments, poor maintenance environment, lack of data, data loss, and abnormal severe, resulting in high frequency and high precision carbon emission monitoring. As privacy protection and data security issues are gradually taken seriously by government departments and related enterprises, the inability or unwillingness to share carbon emission-related data among enterprises or even among various departments within enterprises has created an increasingly severe data silo phenomenon. In addition, how effectively breaking the data barriers between various sectors is an urgent problem in grasping carbon emission change changes accurately. Therefore, this paper proposes a carbon emission monitoring model for key urban sectors based on vertical federated deep learning and multi-source heterogeneous data fusion and sharing. The experimental results show that the model accurately predicts carbon emission change trends in various application scenarios under the data availability and invisibility of each participant.