Abstract: An accurate and reliable deformation prediction model is essential to ensure the safe operation of water dams. However, previous monitoring models lack consideration of the multidimensional spatial or temporal correlation characteristics of massive monitoring data, and fail to effectively reflect the overall or regional deformation patterns of a water dam. This paper adopts a hierarchical agglomerative clustering and projection pursuit method to take into account the integrated distances of measuring points, and deeply explores correlation information in the massive monitoring data of a dam displacement field. A fused deformation sequence reflecting the deformation characteristics of multi-measuring points in the partition is obtained. We develop a new Gaussian process regression optimized by the northern goshawk algorithm to construct a fused deformation prediction model for multi-measuring points in a partition, and construct a confidence interval of the prediction results by the Lajda criterion. The influence of different kernel functions on the model prediction accuracy is examined through an engineering example. A comparative analysis verifies that our method is more accurate and applicable than the conventional models and can achieve a reliability estimation of the prediction results. It is a useful tool for the safety monitoring of dam deformation. Keywords: dam deformation prediction; clustering partition; projection pursuit method; northern goshawk optimization; Gaussian process regression

Key words: dam deformation prediction, clustering partition, projection pursuit method, northern goshawk optimization, Gaussian process regression