Deep Clustering for Interpretable Electric-Carbon Potential Analysis in Smart Grid Resource Management

The emergence of modern smart grids introduces a more distributed and diverse structure compared to traditional grids. The widespread adoption of distributed energy resources (DER), energy storage systems, and flexible loads, along with the implementation of virtual power plant (VPP) technology, enables the centralized coordination and control of these assets. This paper proposes a deep clustering-based load analysis method aimed at efficient resource classification and effective scheduling. First, a comprehensive electric-carbon feature system is developed for resource evaluation, leveraging a robust Multi-Criteria Decision-Making (MCDM) method to provide an objective evaluation standards. Next, a deep clustering algorithm is employed to capture low-dimensional temporal dependencies within multi-modal data, enhancing clustering efficiency and enabling real-time online classification. Finally, network visualization within the algorithm offers insights into the distribution of high-quality and high-potential resources. The proposed method is validated on a dataset of 500 resources, with experimental results confirming its effectiveness and practicality.