Objectives For the structure of the lashing bridge of an ultra-large container ship, the complicated design and severe load environments lead to higher requirements for reliability. Aiming at the problems of the poor efficiency and low accuracy of large ship structure reliability analysis, this paper proposes an improved gradient boosting decision tree-Monte Carlo (GBDT-MC) method.
Methods First, an approximate model of the improved gradient boosting decision tree (GBDT) is established through the Python library, fewer sample points are generated through experiment design and the sample points near the failure surface are screened. The SMOTE algorithm is then used to synthesize new sample points and participate in finite element calculation, as well as being combined with the original sample points to form a training set. The trained approximate model is used to predict the sample point information generated by the Monte Carlo (MC) method, thereby completing the structural reliability analysis. Finally, the feasibility and accuracy of the improved GBDT-MC method is verified by two examples and applied to the reliability analysis of the structure of the lashing bridge of an ultra-large container ship.
Results The calculation results show that the failure probability error under the effect of static lashing force is 3.5% and the calculation time of the improved GBDT-MC method is 2.55 h, but the MC method requires 416.7 h. Therefore, within the allowable calculation error range, the improved GBDT-MC method can greatly reduce the calculation time of reliability analysis.
Conclusions This improved GBDT-MC method significantly improves calculation accuracy and shortens the calculation time, which can provide support for the optimization design of high reliability index structures.
DownLoad: