A hybrid model integrating principal component analysis, fuzzy C-means, and Gaussian process regression for dam deformation prediction

Dam behavior prediction model is a fundamental component of dam structural health monitoring systems. As the most intuitive monitoring indicators, deformation is commonly used to reflect the dam behavior change. The selection of input variables and training samples determines the performance of dam deformation predictive models. In this paper, a novel hybrid model integrating principal component analysis (PCA), fuzzy C-means (FCM), and Gaussian process regression (GPR) are proposed to predict dam deformation. Specifically, PCA is utilized to extract the main information of original thermometer data as temperature variables, while FCM is used to divide the samples into several categories according to the similarity of the environmental monitoring data. Then, the samples in each category are used to train GPR models with five commonly used covariance functions based on influencing factors, respectively. In the test phase, FCM is used to determine what category the samples in the test set belong to, and then, the corresponding trained GPR model is utilized to predict dam deformation. The proposed hybrid model is fully demonstrated and validated by monitoring data collected from a multiple-arch concrete dam in long-term service. Various benchmark models with or without FCM analysis are selected as comparison models. Experimental results show the proposed novel model outperforms the other comparison methods in terms of all evaluation indicators. This indicates fuzzy clustering analysis can effectively improve the performance of the prediction model, and the proposed hybrid model can predict future dam deformation with high accuracy and efficiency.