混合效应模型在道面状态转移概率估计中的应用分析

为研究混合效应Logistic模型确定马尔可夫状态转移概率矩阵的方法及效果,以道面结构厚度、道面使用时间、状态等级和交通量等因素变量为固定效应,以截距为随机效应,建立混合效应Logistic模型.采用多个机场积累的道面实测PCI值为数据源,估计模型参数并作实例分析.结果表明:①应用混合效应Logistic模型可分析多因素对道面使用性能的影响,获得非齐次状态转移概率矩阵,实现道面使用性能马尔可夫动态预测,显著改善概率预测的精度;②混合效应Logistic模型中随机效应能够反映道面数据内不可观测的异质性,降低模型残差的相关性,提高固定效应参数估计的可靠性;③通过对混合效应Logistic模型随机效应参数bi的估计,可得到数据源内任意个体道面的状态转移概率矩阵,克服传统方法多次建模及数据不足的困难,从而实现对特定道面的性能预测.

Application of Linear Mixed Effects Model to Estimating Pavement Markov Transition Probabilities

In order to study the approach and effectiveness by using mixed effects logistic model to estimate transition probability matrices for pavement deterioration modeling, a mixed Logistic model is used to establish a dynamic relationship between pavement transition probabilities and explanatory variables such as pavement age, thickness, traffic level and random intercepts. A case study is made on the application of the model with real data. The comparison results show that: The impact of pavement types, environmental factors, traffic loading, and other relevant factors can be directly considered and a non homogeneous transition probability matrix, which varies with time and yield better predictions, is derived by using mixed logistic model. Unobserved heterogeneity which comes from measurement errors and unobserved factors across different individual pavement sections is captured by random effects, and then bias and inconsistency of estimates are reduced to an acceptable small level. Different individual pavement transition probability,which can be used to predict a given pavement performance, is obtained by estimating random effects parameters in the mixed logistic model, especially when the size of the data set is insufficient or work of modeling individuals is heavy, inefficient traditional approach cannot provide unbiased, consistent, and efficient model estimators.