Operation-Index-Driven Evaluation of Source-Grid-Load-Storage Distribution Networks via Digital Dynamic Modeling

This paper proposes an operation-index-driven intelligent generation method for distribution network simulation models, integrating model library predefinition, multi-level equivalent modeling, and multi-agent collaboration into an automated pipeline. Three agents collaborate through a unified message bus, task queue, and device model library as follows: Monitor Agent performs time-series data acquisition, national-standard threshold evaluation, and topology verification to trigger modeling tasks; Energy-flow-Model Agent uses a hierarchical model library and LLM (Large Language Model) to automatically match standardized models, parses CIM topology, and applies Thevenin/Ward/three-phase admittance matrix multi-level equivalent modeling to construct Energy Flow Network (EFN) computation graphs; Device-Model-Data Agent supplies missing impedance parameters via a public device standard library. The pipeline automatically generates 15 min granularity current-state assessment models synchronized with operating conditions. Validation on eight consecutive days of 10 kV feeder field measurements (96-time sections/day, 91 transformer areas) shows three-phase voltage mean MAPE of 2.52% and 98.9% correct trigger rate with no false activations or missed detections, achieving end-to-end automation from archive parsing to model generation.