论文部分内容阅读
基于贝叶斯理论,以马尔可夫链蒙特卡罗方法(Markov chain Monte Carlo Simulation,MCMC法)的自适应差分演化Metropolis算法为参数后验分布抽样计算方法,建立利用时变测试数据的参数随机反分析及模型预测方法。以香港东涌某天然坡地降雨入渗测试为算例,采用自适应差分演化Metropolis算法对时变降雨条件下非饱和土一维渗流模型参数进行随机反分析,研究参数后验分布的统计特性,并分别对校准期和验证期内模型预测孔压和实测值进行比较。研究结果表明,DREAM算法得到的各随机变量后验分布标准差较先验分布均显著减小;经过实测孔压数据的校准,模型计算精度很高,校准期内95%总置信区间的覆盖率达到0.964;验证期第2～4个阶段95%总置信区间的覆盖率分别为0.52、0.79和0.79,模型预测结果与实测值吻合程度较高。
Based on the Bayesian theory, the adaptive differential evolution Metropolis algorithm based on the Markov chain Monte Carlo Simulation (MCMC) is used as the parameter sampling method for posterior distribution sampling, and the parameters of time-varying test data are established randomly Back analysis and model prediction methods. Taking a rainfall infiltration test on a natural slope in Tung Chung, Hong Kong as an example, the adaptive differential evolution Metropolis algorithm is used to stochastically back-analyze the unsaturated soil one-dimensional percolation model parameters under time-varying rainfall conditions to study the statistical properties of the posterior distribution of parameters. The predicted pore pressures and measured values of the model during the calibration period and the verification period are compared respectively. The results show that the standard deviation of the posterior distribution of each random variable obtained by the DREAM algorithm is significantly reduced compared with the prior distribution. After calibrating the measured pore pressure data, the accuracy of the model calculation is very high. The 95% confidence interval coverage during the calibration period Reaching 0.964. The coverage of 95% confidence intervals in the second to the fourth stages of the verification period were 0.52, 0.79 and 0.79, respectively. The predicted results of the model are in good agreement with the measured values.