Neighborhood-Aware Optimization for Smart Building Predictive Maintenance
Auteurs : Anas Hossini (DAVID), Leïla Kloul (DAVID), Maël Guiraud (LINEACT), Benjamin Cohen-Boulakia (David)
Conférence : Communications avec actes dans un congrès international - 23/11/2026 - International conference on system reliability and safety
Predictive maintenance of smart buildings must
keep the Quality of Service (QoS) above its required level at
minimal long-term cost. This constrained optimization problem is
commonly tackled with neighborhood-based metaheuristics such
as Simulated Annealing (SA), yet the neighborhood itself—how
candidate solutions are generated at each iteration—is rarely
studied. We show that this choice strongly impacts the quality of
the resulting strategies. We build a reliability model that captures
component aging and inter-system fault propagation, and derive
three problem-aware neighborhoods from it, which we compare
under identical conditions on the lighting system of a real smart
building. The neighborhood alone shifts performance as much as
changing the optimization algorithm, and a poorly designed one
can bias the very signal used to evaluate the strategies. A hybrid
reinforcement learning–SA algorithm inherits these gains.