Abstract: The living load of high-rise industrial rack is usually much greater than the dead weight of structure, so the shear stiffness of upright frame is crucial for the overall lateral stability of the rack. Due to the variety of the size and the connection of structural components, it is still difficult to build a general analytical model at present. Taking the upright frame in Z-pattern as an example, based on the physical simulation and data-driven method, a cognitive prediction model on shear stiffness of upright frame is proposed. Firstly, based on the mechanical test, the finite element simulations of shear stiffness of upright frame were made to obtain the shear stiffness data sets of different upright frame structures. Secondly, the XGBOOST algorithm was used to train the intelligent prediction model of shear stiffness. In contrast with the convolutional neural network and long short-term memory network, the prediction results of the XGBOOST algorithm are more efficient and closer to the finite element method. On this basis, the results of prediction model are interpreted and understood with the visual method of SHapley additive exPlanation (SHAP). By the importance score analysis, it is found that the shear performance of the upright frame is mainly determined on the design of the bracing configuration, and the distance between two uprights has relatively limited influence on the shear performance which could be flexibly configured according to the user’s requirements. What’s more, the SHAP summary plots and dependency plots showed that the shear stiffness of upright frame is significantly increased by enlarging the dimensions of structural members under the separation configuration of bracing members.

Key words: data-driven, shear stiffness, XGBOOST, cognitive prediction, upright frame