Neural network-based repair strategy for seismic resilience of building group considering indirect economic losses

The growing importance of urban seismic resilience highlights the need for effective strategies to minimize earthquake-induced losses of building groups. Urban seismic capacity assessment requires rapid and accurate prediction of buildings’ seismic responses. However, in large urban building groups, the design information of many buildings is often difficult to obtain completely. In addition, post-earthquake interruption of economic activities will cause significant indirect losses to the city. Based on this background, this paper developed a convolutional neural network (CNN) model using structural characteristic parameters and ground motion parameters, by which the seismic response of structures in a building group can be rapidly predicted. On this basis, the repair priority of buildings was determined according to the repair efficiency value, and an optimal repair strategy considering indirect losses was proposed to minimize the post-earthquake indirect losses of building groups. Taking 908 buildings in Shanghai as an example, the effectiveness of the model is verified, and a comparison is made with the existing conventional repair strategies. Results showed that the proposed neural network accurately predicted the damage states of buildings. Compared with other existing strategies, the proposed optimal repair strategy effectively reduced the indirect economic loss. Further, the increased repair resources can reduce the indirect economic loss and repair time, but the reduction ratio decreases with the further increase of resource allocation.