基于变量修正分布的混凝土结构长期变形预测Prediction of long-term deformation for concrete structure based on revised distribution of variables

由于随机因素的影响，混凝土结构长期变形通常表现出很强的离散性，Bayesian理论提供了改善这种离散性的方法。基于Bayesian理论的混凝土结构长期变形预测方法的核心是引入短期变形构造似然函数，通过修正先验概率得到长期变形的后验概率分布。但是对于实际结构而言，在施工之前短期变形及其标准差无法获取，这就使得这种方法在开展时机和实际应用方面存在一定的限制性。为了改善这种限制性，在随机变量修正分布的基础上，结合拉丁超立方抽样技术，采用CEB‐FIP（MC90）模型建立了钢筋混凝土梁长期变形的随机分析模型。采用该模型进行混凝土梁长期变形随机分析，得到基于变量修正分布的混凝土梁长期变形预测结果，并分别与先验预测结果和Bayesian预测结果进行比较。研究结果表明，基于变量修正分布的预测结果与Bayesian预测结果十分接近，比先验预测结果不确定性降低50％左右，与试验结果吻合良好。

Prediction of long-term deformation for concrete structure based on revised distribution of variables

The long-term deformation prediction method based on Bayesian theory for concrete structures becomes weak when the prediction work should be carried out before the structure being constructed.Because there is no measured value of short-term deformation to build the likelihood function,and the standard deviation of the short-term deformation,an important part of the likelihood function,is difficult to acquire for a real structure.An alternative method of Bayesian for predicting long-term deformation of the concrete structure was provided in this paper.Prior distribution of variables were revised by taking advantage of data of variable experiment.The stochastic analysis was carried out by using Latin hypercube sampling (LHS) method based on the revised distribution of the variable.The prediction results of 4 concrete beams indicate that the uncertainty of the revised distribution decreases by nearly 50%.The revised prediction results show a good coincidence with the Bayesian prediction results and the experimental data.