Spatial association-considered real-time risk rate assessment of high arch dams using observed displacement and combination prediction model

Abstract

Structural health monitoring is an important approach to diagnose dam safety. The main purpose of this paper is to improve the accuracy of displacement prediction and real-time risk rate assessment of high arch dams by utilizing the deformation spatial association of multiple displacement monitoring points. Two strategies are integrated for the former. First, in addition to the traditional objective of minimizing the fitting error, deformation spatial association is also determined as an optimization objective for machine learning models. Second, five frequently used models are adopted as sub-models, and the overfitting degree is then used as an evaluation index to select optimal sub-models that included in the combination prediction model. On this basis, probability density analysis is conducted for displacement residuals, and a quantification function is defined to assess the real-time risk rate of a single monitoring point. The real-time risk rate of a high arch dam is then calculated according to the risk rates of multiple monitoring points, during which the spatial association are taken into account by the Copula function. Research results of the Jinping-I arch dam indicate that the deformation spatial association is very beneficial for improving the accuracy of dam safety diagnosis, where the prediction mean square error of four machine learning models decreases with an average rate of 66.92%, 45.37%, 35.45% and 73.12%, respectively, and the combination prediction method can further improve the displacement prediction accuracy by 23.79%. The observation data-based high-risk period of arch dams mainly occurs in operation scenarios that the rising or dropping rate of reservoir water level changes sharply. The joint risk rate can effectively avoid the accidental influence of a single monitoring point, and it exhibits an exponential decrease with the increasing number of used monitoring points. For the Jinping-I arch dam, a minimum number of 10 monitoring points is required.