基于高斯过程回归的建筑物震陷量预测模型

城市抗震防灾系统是一个复杂开放巨系统，系统中由于灾情的动态演化导致的建筑物震陷量形成机理也日趋复杂。本文根据高斯过程理论和贝叶斯规则，对训练样本进行的“归纳推理学习”，即综合先验信息，调整各随机变量的后验分布，进而提出基于高斯回归过程的建筑物震陷量非线性预测模型。采用EP算法获得预测样本潜在函数的近似后验高斯分布，并对其超参数和协方差函数的选择进行了探讨，利用LSSVM模型、PLS模型和MLR模型等统计模型对建筑物实测震陷样本进行预测训练，通过模型的交叉验证分析及建模参数详细对比分析，验证了预测模型的科学性和可靠性，可为城市抗震防灾决策提供借鉴。

Prediction of Building Settlements Due to Earthquake Based on Gauss Process Regression Model

Urban earthquake disaster prevention system is a complex open giant system, the formation mechanism of building settlement due to earthquake is increasingly complex in virtue of the disaster evolution. According to Gauss process theory and Bayesian rule, a nonlinear prediction model of building seismic subsidence based on Gaussian regression process is proposed in this paper, and the inductive reasoning learning of training samples is carried out by which synthesize the prior information and adjust the posterior distribution of the random variables. Furthermore, the EP algorithm is used to obtain the approximate posterior Gauss distribution of the potential function of the predicted samples, and the selection of its hyper parameter and covariance function is discussed. Moreover, the building subsidence samples also has been trained by several statistical models such as LSSVM model, PLS model and MLR model. Therefore, the scientificity and reliability of the prediction model are verified through the cross validation analysis and the detailed comparative analysis of modeling parameters, which can provide relevant decision support for urban earthquake resistance and disaster prevention planning.