Democratizing Microgrid Optimization: An LLM Agent for Dispatching Mobile Chargers to Construction Electric Vehicles

Daniela Rojas Lozano, Yuanyuan Shi

Keywords: LLM agents; world models; MILP; microgrids; mobile charging; optimization; human-AI collaboration.

TL;DR: MCS Dispatch AI Assistant, it's an agentic framework that reframes optimal scheduling of dispatching Mobile Charging Stations (MCS) to Construction Electric Vehicles (CEVs) as a conversational problem

Abstract: Optimally dispatching Mobile Charging Stations (MCS) to Construction Electric Vehicles (CEVs) requires a mathematical formulation that unifies logistical routing constrained by dynamic vehicle work schedules with the physical limits of battery dynamics and energy transfer. This integration poses a significant barrier for site managers aiming to decarbonize their operations. To bridge this gap, we introduce and demonstrate MCS Dispatch AI Assistant, an agentic framework that reframes scheduling as a conversational problem. Our system employs a multi-agent pipeline to translate high-level goals into mathematically precise, solver-ready inputs. This architecture includes an Understanding Agent for parsing natural language, a Validation Agent for ensuring physical feasibility, and a Recommendation Agent for completing the problem specification with expert defaults. The framework treats the formal MILP optimization model as a specialized tool: the primary agent calls this tool to reason about the system’s complex physical and economic constraints, guaranteeing the final output is both feasible and optimal. We demonstrate our system in a real-world use case at the UCSD microgrid, showcasing a human AI collaboration that democratizes access to optimization, minimizing both operational costs and environmental impact.

Submission Type: Demo Paper (4-9 Pages)

Submission Number: 118