Decoding the grammar of heritage: a machine learning framework for sustainable and adaptive reuse of Venice's built fabric

Abstract
Venice faces a cumulative conservation problem: Sea-level rise is expected to increase in the coming decades, and together with hydrodynamic stress and material degradation, conventional building-level interventions will be inadequate. The current digital methods of heritage protection in coastal cities still face two methodological deficiencies: the disconnection between large-scale urban environment analysis and small-scale building material study, and a reactive rather than proactive mode that limits adaptive ability. To fill these deficiencies, the present study establishes a cross-scale framework based on a Perception-Cognition-Generation paradigm and selects Venice as the case. Hydrological simulation and multi-objective optimisation are used to guide changes in the spatial distribution of buildings in urban areas to regulate tidal flow and increase sunlight for old towns. At the architectural level, CNNs and GNNs are employed to extract material attributes and implicit structural logic from built fabrics, thus forming a computable, low-carbon knowledge base. The two-level outputs are connected by the Wave Function Collapse (WFC) algorithm to build a generative decision-support system that can generate adaptive regeneration plans according to the historical authenticity, structural safety and material circularity of the site. According to the above experimental data, the new framework will have a recycling rate of 68%, reduce energy consumption by 5%, and lower embodied carbon emissions by 35%—42%. The increase in the accuracy of recognizing tacit construction knowledge by one-time operation for the machine learning part surpassed that achieved by previous rule-based methods. These results are to be regarded as decision-support evidence rather than automatic conservation instructions. Reliability is also affected by the hydrological calibration, image annotation quality, graph construction and expert review.