基于VAE和INGO-ResNet-Informer的双阶段混凝土坝位移预测模型

A two-stage concrete dam displacement prediction model based on VAE and INGO-ResNet-Informer

摘要

【目的】位移是内外环境双重作用下大坝运行态势和结构性态最直观可靠的反映,建立科学合理的大坝安全监控模型对保障其安全运行具有重要意义。【方法】针对混凝土坝位移监测数据呈现复杂非线性、多样且时变的特征,为提高大坝性态分析的准确性,提出一种基于变分自编码器(VAE)、改进北方苍鹰优化算法(INGO)、残差神经网络(ResNet)和Informer的双阶段混凝土坝位移预测模型。首先,通过VAE对温度监测数据进行潜在空间建模与非线性特征提取;然后,利用改进ResNet对多源输入影响因子进行深层表征,引入Informer模型捕捉数据中的长时间依赖信息,并采用多种策略改进得到的INGO算法对模型关键超参数进行自适应调节,实现对坝体位移的精确建模。【结果】以某混凝土拱坝位移监测数据为例,将所提模型与传统预测模型进行对比,结果显示,所提模型可以准确预测大坝位移变化趋势,在各测点位移数据测试集上的RMSE、 MAE和R²均为最优。【结论】所提模型在预测精度、稳定性及残差抑制能力等方面均具有显著优势,可为混凝土坝安全监测与运行管理提供技术支撑。

Abstract

[Objective] Displacement is the most intuitive and reliable reflection of the dam' s operating status and structural behavior under the dual effects of internal and external environments. Establishing a scientific and reasonable dam safety monitoring model is of great significance to ensure its safe operation. [Methods] To address the complex nonlinear, diverse, and time-varying characteristics of concrete dam displacement monitoring data, a two-stage concrete dam displacement prediction model based on Variational Autoencoder(VAE), Improved Northern Goshawk Optimization(INGO) algorithm, Residual Neural Network(ResNet) and Informer is proposed to improve the accuracy of dam behaviour analysis. Firstly, VAE is used to perform latent space modeling and nonlinear feature extraction on temperature monitoring data. Then, the improved Res Net is utilized to deeply characterize the multi-source input impact factors, the Informer model is introduced to capture the long-term dependency information in the data, and the INGO algorithm, enhanced with multiple strategies, adaptively tunes the key hyperparameters of the model to achieve precise modeling of dam displacement. [Results] Taking the displacement monitoring data of a concrete arch dam as an example, the proposed model is compared with the traditional prediction model. And the result demonstrate that the proposed model can accurately predict the dam displacement trend, achieving the best RMSE, MAE and R² values on the displacement data test set of each measuring point. [Conclusion] The proposed model exhibits significant advantages in prediction accuracy, stability and residual suppression ability, providing technical support for the safety monitoring and operation management of concrete dams.

关键词

混凝土坝位移预测 / 变分自编码器 / 残差神经网络 / Informer / 改进北方苍鹰优化算法 / 影响因素 / 预测模型

Key words

concrete dam displacement prediction / variational autoencoder / residual neural network / Informer / improved northern goshawk optimization algorithm / influencing factor / prediction model

基金资助

国家重点研发计划(2022YFC3005504)

长江勘测规划设计研究有限责任公司生产类自主创新科研项目(CX2023S056)