A Hybrid Generative AI and Optimization Approach for Resilient Emergency Department Layouts
Year: 2026
Type: conference-paper
Abstract: Effective spatial organization in hospital emergency departments (EDs) is critical for delivering timely patient care and minimizing operational inefficiencies. Instead of relying on human judgment or traditional optimization methods, which may not adapt well to changing healthcare systems, this work proposes a hybrid method of combining a large language model (LLM) with an optimized algorithm to automate and improve hospital department layouts. The framework incorporates key spatial and operational criteria, including patient flows between rooms, inter-room distances, room priority levels and the most important are the health crises constraints. The LLM acts as a constraint-aware symbolic generator used to initialize the optimization stage that produces clinically plausible room-function reassignments from a structured JSON representation of the layout, sanitary and operational constraints, and scientific knowledge retrieved via Retrieval-Augmented Generation (RAG). These LLM-generated proposals are then refined by Particle Swarm Optimization (PSO) and Simulated Annealing (SA), which optimize flow-weighted distance costs, reassignment penalties, room priorities, and feasibility requirements. Our contribution focuses on integrating LLM-based reasoning with optimization techniques to improve solutions to the healthcare facility layout problem (FLP). By comparing widely used optimization algorithms for healthcare facilities, we evaluate the effectiveness of our integrated approach. Our experimental results from our simulated ED with realistic hospital constraints indicate increased spatial and functional efficiency over the existing baseline layout. The simulated emergency department in this study is inspired by the organization of a hospital emergency unit in Roanne.
Institutions: Université Jean Monnet, Sidi Mohamed Ben Abdellah University
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201874792