龙溪口防护堤变形在线监控模型构建与优化

龙溪口库区防护堤的地质条件与运行环境复杂，如何利用原型监测数据实现其安全运行状态的在线监控是亟待解决的关键问题。针对传统监控模型在非线性特征的监测序列方面存在泛化性差、精度低等问题，基于长短时记忆神经网络(LSTM)算法，以水位、温度、时效为输入量，变形为输出量，构建防护堤LSTM在线变形监控模型。通过对比分析不同模型参数对精度的影响规律，揭示学习率>分块尺寸>最大迭代次数>隐藏层单元数的多参数敏感性规律，并提出相关参数的建议取值。工程应用表明，该模型精度高，适用性和稳定性强。

Construction and optimization of online monitoring model for deformation of protective embankment in Longxikou

The geological conditions and operating environment of the protective embankment in the Longxikou reservoir area are complex,and how to use the original monitoring data to achieve online monitoring of its safe operation status is a key issue that needs to be urgently solved.In response to the problems of poor generalization and low accuracy in traditional monitoring models for monitoring sequences with strong nonlinear features,a long short-term memory (LSTM) online deformation monitoring model for the protective embankment based on LSTM algorithm is built,with water level,temperature,and time as inputs and deformation as outputs.By comparing and analyzing the influence of different model parameters on the accuracy,the multi-parameter sensitivity law of learning rate>block size>maximum number of iterations>number of hidden layer units is revealed,and the recommended values of relevant parameters are proposed.Engineering applications show that the model has high accuracy,strong applicability,and stability.