Research on Building Disaster Governance Strategies in the Guangdong–Hong Kong–Macao Greater Bay Area Driven by AI Digitalization—Based on a Stochastic Evolutionary Game Model

The Guangdong–Hong Kong–Macao Greater Bay Area has high building density and diverse project types, and building disaster governance features cross-regional, multi-stakeholder and strongly uncertain characteristics. To reveal the multi-agent strategic interactions after embedding AI digital technologies into the governance process, this paper constructs a three-party stochastic evolutionary game model among public regulators, construction firms and AI technology providers, and introduces multiplicative Gaussian white noise with boundary degradation into the replicator dynamics. The study finds that (1) under baseline parameters, the system evolves toward the state “coordinated strong regulation–AI-compliant governance–high-quality supply”; noise below the local mean-square stability threshold does not change the direction of recovery near the equilibrium but enlarges short-term fluctuations of stochastic trajectories; (2) there is synergistic transmission in the initial strategies of the three parties, with firms’ compliance probability linking both the regulatory and technology sides; (3) governance performance benefits, cross-regional coordination gains, firms’ digitalization gains and high-quality technology subsidies each form positive incentives at different links, while firms’ AI retrofit costs directly depress compliance returns; (4) the mechanism shown on the original parameter surface indicates that higher AI retrofit costs first suppress firms’ AI-compliant governance and, via firms’ demand transmission, affect high-quality technology supply, whereas the response of public regulators is relatively weak. The findings provide mechanism-level hypotheses for phased incentives, cross jurisdictional data collaboration, technology quality assurance, and adaptive regulation. Because the parameters are dimensionless and uncalibrated, project or jurisdiction specific policy magnitudes require empirical estimation and validation.