Predicting structural deterioration of large-scale building clusters using snapshot data: an integrated Markov-LSTM model

To ensure a safe environment for occupants, predicting long-term structural deterioration of buildings is critical. However, existing models have limited capability to predict structural deterioration with mathematical tractability and accuracy, especially for large-scale building clusters. To address this gap, this study aims to establish a new integrated Markov-LSTM model, combining the strengths of model-driven and data-driven methods, for enhanced structural deterioration prediction. Specifically, the proposed two-stage inhomogeneous Markov chain allows the deterioration process to be tractable through the derivation of analytical transition probabilities. To further improve the accuracy, long short-term memory (LSTM) is employed to predict the residuals calculated from Markov-based predictions and true values. The performance of the proposed model is evaluated through two case studies, using snapshot data of large-scale building clusters. The results demonstrate significant improvements over benchmark models, with the reduction of mean absolute error (MAE) by an average of 0.1780 (and 0.3292), mean squared error (MSE) by 0.1421 (and 0.5717), and mean absolute percentage error (MAPE) by 4.7778% (and 13.2736%) in Case 1 (and Case 2). This study contributes to research and practice in structural deterioration prediction by providing both mathematical tractability and accuracy, focusing on large-scale building clusters, and supporting more effective condition-based maintenance.