Agentic Multi-Objective Retrofitting Framework (AMORF) for Buildings

Hindesh Akash, Yang Zhao

Keywords: Machine Learning, Multi-Objective Optimization, Building Retrofitting, Agentic Systems

Abstract: Retrofitting legacy buildings for energy efficiency is critical for sustainability, yet balancing economic feasibility, energy savings, and occupant comfort remains challenging. This research proposes a Phased Retrofitting System framework with Reinforcement Learning \& Agentic AI characteristics, integrating economic assessment with adaptive HVAC control. Algorithmic developments decide which locations to fit and what to fit. The strategic agent decides on synchronization with existing systems while optimizing daily energy use, jointly minimizing energy consumption, carbon emissions, and costs while maintaining thermal comfort. Simulations show higher energy savings and minimal comfort violations compared to baselines. This approach offers a scalable solution for building portfolios, aligning with urban decarbonization goals.

Submission Number: 27