Probabilistic Shear Strength Model of RC Columns Based on Gaussian Process Regression with Anisotropic Compound Kernel Function
A probabilistic shear strength model for reinforced concrete(RC)columns is proposed based on Gaussian process regression(GPR)with an anisotropic mixed kernel function to address the limitations of traditional models,which often exhibit low nonlinear approximation ability and poor generalization performance.A new anisotropic mixed kernel function is developed using the additivity and autocorrelation properties of the Matern and Rational Quadratic kernel functions,while the automatic relevance determination function is introduced to account for the effects of various feature parameters.The probabilistic shear strength model for RC columns is established by integrating the anisotropic mixed kernel function with the Gaussian process regression algorithm.The posterior distribution of model hyperparameters is obtained using Bayesian infer-ence,and the hyperparameters of the probabilistic shear strength model are determined through the maximum likelihood estimation method.The effectiveness of the proposed model is validated by comparing it with traditional kernel functions,machine learning models,and mechanical mod-els using 91 sets of experimental data.The analysis results indicated that the deterministic prediction indices RMSE and MAE of the proposed mod-el are reduced by approximately 16%and 19%,respectively,compared to traditional kernel functions.In contrast,the probabilistic prediction in-dices NLPD and MSLL are reduced by about 15%and 23%,respectively.Compared to traditional machine learning models,the RMSE and MAE of the proposed model are decreased by 38%and 39%,respectively.The proposed model demonstrated high prediction accuracy and generalization per-formance and effectively quantified the uncertainty in the shear strength of RC columns.
万方数据
