GNDO-SVR: An efficient surrogate modeling approach for reliability-based design optimization of concrete dams

Dams are considered as the critical part of infrastructure of any country, and their construction requires considerable amounts of capital and labor. The optimal design of dams can lead to significant savings in time and cost. In this regard, the goal of this study is proposing an efficient framework for the probabilistic design of gravity dams using a new developed reliability-based design optimization (RBDO) approach. In this RBDO approach, the concreting volume of dam is considered as the objective function of the optimization problem under uncertainties, while the geometric parameters are considered as the decision variables. To solve the RBDO problem of the gravity dam design, a new surrogate model consists of a hybrid Support Vector Regression (SVR) based generalized normal distribution optimization (GNDO) is coupled to the Monte Carlo Simulation (MCS). The hybrid SVR-GNDO is utilized to predict the dam response in order to decrease the computational cost during the RBDO analysis. The use of GNDO-SVR instead of SVR increased the predicted R-2 from 0.85 to 0.99. The proposed method was implemented on the Pine Flat dam in California (USA). The results showed that the performance of the RBDO approach in the optimal design of a concrete gravity dam has a better safety level than the deterministic approach. This is such that the presented new hybrid model was able to reduce the failure probability of the design under uncertainty from 71% to 0.000018% by increasing the volume of concrete used in the dam by 8%.

Automated summary

This study proposed an efficient framework for probabilistic design optimization of gravity dams, using a new reliability-based design optimization approach. By coupling a hybrid Support Vector Regression-based generalized normal distribution optimization model to Monte Carlo Simulation, the research successfully optimized the design of a concrete gravity dam, increasing safety level and significantly reducing failure probability.