关键词: 混凝土坝变形预测, 变量筛选, 极限学习机, 平均影响值, 反向逐变量剔除法

Abstract: Traditional statistical models are of weak generalization capability and are prone to introduce high-dimensional variables,which will negatively affect the output of neural network-based prediction models and increase the risk of overfitting.It is necessary to build a data-driven model with appropriate dimensionality to accomplish accurate monitoring of dam deformation.In this paper,extreme learning machine(ELM)is selected as the base prediction model,and a variable selection method based on mean impact value(MIV)-ELM model is proposed to eliminate redundant information in the initial variable set,thus reducing the model's complexity and improving the prediction accuracy.Analysis results demonstrate that compared with traditional prediction models,HST-MIV-ELM not only has the highest prediction accuracy and robustness,but also has strong scalability.The study provides a reliable theoretical basis for the construction of dam safety monitoring system.

Key words: deformation prediction for concrete dam, variable selection, extreme learning machine, mean impact value, reverse variable-by-variable elimination method