贝叶斯框架下的大坝变形交互式时变预测模型及其验证

大坝变形是同一时刻内外多重风险因素综合作用的结果,应用时序分析方法挖掘历史监测数据潜在规律是变形预测的常用方法,现有时变预测模型不仅参数配置难度高,且难以融入专业知识,导致预测效果并不理想。本文提出一种耦合自动预测算法与大坝专业知识的交互式变形预测模型。该模型在贝叶斯框架下,以加法模型为基础重构各时序分解项作为模型底层,根据仿真结果甄选模型参数缺省值进行自动预测,通过结合参数化检测与直观参数配置实现交互式建模,并借助拟合可视化和统计指标准确反映预测误差来源,从而进一步修正参数以提高模型适用性。基于上述流程协同构建的大坝变形循环预测体系,以某混凝土坝多测点长期变形监测数据为例,对模型的准确性、鲁棒性和灵活性进行了有效验证与分析,为大坝变形安全预测与分析提供了新的模型和手段。

Prediction model for interactive time series evolution and its verification of dam deformation under Bayesian framework

Dam deformation is a result of the combined action of multiple risk factors at the same time. It is a common method for predicting deformation by using time series analysis to excavate the potential pattern of historical monitoring data. The existing prediction model for time series evolution of dam deformation is not only difficult to configure parameters,but also hard to integrate expertise,which leads to poor prediction. The solution presented in this paper is an interactive deformation prediction model (IDPM),which combines automatic prediction procedure and background knowledge of dam engineering field. Each decomposition item of the traditional additive model is refactored as the underlying structure of IDPM under Bayesian framework. The default values for model parameters are selected on the basis of numerical simulation to achieve automatic prediction. The artificial custom modeling is also realized by combining parameterized detection and intuitive parameters configuration. By means of visual fitting and statistical indicators to accurately reflect the source of prediction errors,model parameters are further modified to improve the practical applicability. In addition,taking a concrete dam as an example,the prediction circulation system of dam deformation composed by the above processes is used to effectively verify and analyze the accuracy, robustness and flexibility of IDPM. The model proposed would provide a novel method for prediction and analysis of the dam deformation safety.