传统基于单一模型的混凝土大坝变形预测方法预测精度低,噪声稳健性差,泛化能力弱。为解决该问题,提出一种基于贝塔先验主成分分析(BP-PCA)与水循环算法(WCA)优化支撑向量机(SVM)相结合的混凝土大坝变形组合预测方法。首先利用所提BP-PCA模型对变形数据进行多尺度降噪分解,将复杂非线性、非平稳随机过程分解为一系列结构简单的主分量;然后利用WCA优化的SVM(WCA-SVM)对每个主分量分别建立预测模型;最后将多个主分量的预测结果综合叠加得到最终预测结果。以我国中部地区某混凝土大坝变形监测数据开展试验,结果表明,所提BP-PCA模型能够有效挖掘数据中隐含的趋势性和规律性信息,BP-PCA-WCA-SVM模型能够获得较高的预测精度,预测结果的相对误差为1.07%,误差均方根为0.065。相对于Kalman滤液、SVM、CNN 3种方法,所提模型预测性能提升均超过62%,并且具有更强的噪声稳健性和泛化能力。

Traditional single-model prediction methods suffer from issues like low accuracy, susceptibility to noise, and limited generalization capability. To address these challenges, we propose a novel approach for predicting concrete dam deformation by integrating the Beta Prior Principal Component Analysis (BP-PCA) and the Water Cycle Algorithm (WCA). Initially, the BP-PCA model decomposes deformation data into multiple scales, effectively reducing noise. This decomposition transforms the intricate nonlinear and non-stationary stochastic process into a set of principal components with simplified structures. Simultaneously, it enhances noise robustness by suppressing noise during the decomposition process. Subsequently, we employ the Water Cycle Algorithm optimized Support Vector Machine (WCA-SVM) to construct prediction models for each principal component. Finally, we integrate the prediction outcomes from multiple principal components to derive the final prediction result. The relative prediction error is minimized to 1.07%, with a root mean square error of 0.065. Compared to the three methods included in the comparative analysis, our approach yields over 62% improvement in prediction performance, demonstrating superior noise robustness and generalization capability.