A probabilistic deep reinforcement learning approach for optimal monitoring of a building adjacent to deep excavation

During deep excavations, structural health monitoring (SHM) of nearby buildings is crucial to ensure safety. The authors propose a probabilistic deep reinforcement learning (PDRL) method to optimize the monitoring to minimize costs and mitigate excavation-induced risks. 1) A Bayesian-bi-directional general regression neural network, built as a probabilistic model, describes the relationship between ground settlement in the foundation pit and the safety of nearby buildings; 2) A double deep Q-network, which can capture the features of the excavation, is trained to form a decision loop for continuous learning of monitoring strategies; 3) The proposed PDRL method is then applied to a deep excavation in No. 14 Shanghai Metro. The method can estimate the probability of damage while updating the state of the nearby building. According to the PDRL strategy, if the quality of the monitoring data can be trusted, monitoring of the adjacent buildings begins in the middle stage rather than the commencement of excavation. If the uncertainty level of the data rises, the starting day can be shifted earlier. The PDRL method is strong enough to address the uncertainties in the environment and the model, helping to optimize the monitoring plan towards achieving cost-effectiveness and risk mitigation.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01918597
  • Record Type: Publication
  • Files: TRIS
  • Created Date: May 16 2024 4:35PM