Optimized deformation monitoring models of concrete dam considering the uncertainty of upstream and downstream surface temperatures

Abstract

Measured data-driven deformation monitoring models of concrete dams are reliable scientific approaches for analyzing the operational state of dam structures. Considering the complexity of dam structural behavior, the existing models are established on the basis of a series of assumptions, which also contribute to the uncertainty of the analyzing results. In view of the fact that classical deformation monitoring models are insufficient in effectively accounting for the hysteresis effect of upstream and downstream surface temperature on the structural behavior of dams, this research analyzes the construction theory of deformation models and adopts the chi-square distribution to determine the weight of antecedent surface temperature. A model taking the weighted air and water temperature measurements as variables for thermal deformation and accounting for the hysteresis effect of surface temperature is then proposed. Meanwhile, to make up for the weakness of the proposed model to be applied in situations where little recorded water temperature data is available, a variant form of the proposed model is also presented. For further improvements in the prediction performance, a Long Short-Term Memory (LSTM) network-based model optimized by Genetic Algorithm (GA) is introduced. An engineering example demonstrates that the proposed model efficiently lowers the uncertainty aroused by surface temperature, that is, it quantitatively considers the hysteresis effect and, to some extent, separates the temperature component mixed with the hydrostatic component. And the GA-LSTM network significantly promotes the prediction performance of the monitoring model, which provides a new method for dam deformation prediction.