Risk assessment of spatially distributed assets with generalized seismic fragility models based on machine learning

Urban areas consist of a diverse array of spatially distributed assets, encompassing both infrastructure systems and portfolio buildings, both of which are crucial to enhancing the sustainability and resilience of urban. Addressing the seismic risk within such settings requires a nuanced approach that considers the spatial distribution of assets and the correlation among ground motion intensity measures (GMIMs). This study undertakes finite element modeling of different structural types. Recognizing the distinct sensitivities of GMIMs to these structural types, 5 machine learning (ML) methods are implemented to construct prediction models for seismic responses based on 24 GMIMs. The generalized fragility model and vulnerability model are developed for different structural types. Moreover, the spatial cross-correlation of 24 GMIMs is integrated into a loss assessment framework using geostatistical techniques and principal component analysis. A virtual city is taken as a case study, demonstrating that the generalized fragility model based on ML enhances the accuracy of regional loss assessment and improves traffic connectivity reliability compared to traditional fragility models that consider a single GMIM. In summary, this study presents a comprehensive framework for regional seismic risk assessment, leveraging ML alongside spatial analysis to enhance prediction accuracy and reliability, thus offering valuable insights for urban planning and disaster mitigation efforts. • ML-based fragility model using 24 GMIMs for RC frames, shear walls, highways, and bridges. • PCA and geo-statistics for spatial cross-correlation fields of 24 GMIMs. • Seismic connectivity reliability of transport networks accounting for spatial cross-correlation impacts.