The Prediction of Concrete Dam Displacement Using Copula-PSO-ANFIS Hybrid Model

The concrete dam displacement monitoring model is an essential part of dam health. Due to the complicated nonlinear mapping relationship between concrete dam displacement and various environmental quantities, as well as conventional statistical models, neural networks, and machine learning methods fail to consider each input's fuzzy uncertainty factors. Therefore, the model's prediction accuracy is usually affected by selecting impact factors and modeling methods. This paper uses the Copula theory to perform nonlinear correlation tests on displacement influencing factors for the above problems. Furthermore, on this basis, this paper proposes a hybrid model, which uses an adaptive neuro-fuzzy inference system (ANFIS) to establish a regression model and uses the particle swarm optimization (PSO) algorithm to find the optimal parameters of the model. This paper takes a roller-compacted concrete gravity dam as an example. It explores the effect of two clustering methods (subtractive clustering and fuzzy C-means clustering) on the ANFIS model's performance based on the dam's measured data. The results show that the MAPE in the subtractive clustering is about 26% less than the fuzzy C-means clustering in the test stage. Finally, this paper compares the prediction results of the Copula-ANFIS-PSO model with the other five conventional methods. The analysis of the results of six models with four error indicators shows that the error in the Copula-ANFIS-PSO model is about 46% less than other models. It provides a new method for concrete dam displacement monitoring.