关键词:
Sensitivity analysis
摘要:
Here, to aocurately predict load-bearing capacity of steel plate-RC composite (PRC) coupling beam, regression trainings of PRC coupling beam test data were conducted using support vector regression (SVR) algorithm, extreme gradient boosting (XGBoost) algorithm and particle swarm optimization-support vector regression (PSO-SVR) algorithm, respectively. In addition, effects of data feature parameters on load-bearing capacity of PRC coupling beam were analyzed using Sobol sensitivity analysis method. The results showed that average absolute percentage errors of prediction models based on SVR, XGBoost and PSO-SVR are 5. 48%, 7. 65%, and 4. 80%, respectively, so the load-bearing capacity prediction model based on PSO-SVR has the highest prediction accuracy, stronger robustness and generalization ability;coupling beam feature parameters of steel plate ratio p, cross section height h and beam span to height ratio In/h have the largest impact on load-bearing capacity of PRC coupling beam, the sum of their global impact indexes is over 0. 75;steel plate ratio p is the single factor having the largest impact on load-bearing capacity of PRC coupling beam, its first-order sensitivity index and global sensitivity index are 0. 342 3 and 0. 362 0, respectively;the study results can provide a reference for design and application of PRC coupling beam in practical engineering. © 2025 Chinese Vibration Engineering Society. All rights reserved.