| 153 | 0 | 65 |
| 下载次数 | 被引频次 | 阅读次数 |
针对变形监测序列非平稳、非线性、多测点协同变化特征显著，以及多测点直接耦合易引入冗余干扰、削弱有效特征的问题，提出一种融合神经基展开时间序列分析(N-BEATS)、细粒度特征软选择机制与双向长短期记忆网络(BiLSTM)的混合预测模型。该模型首先利用N-BEATS将监测序列解耦为趋势项与季节项特征；随后通过细粒度软选择机制筛选高相关特征分量，实现多测点变形信息的高质量融合；最后借助BiLSTM捕捉时序依赖并完成预测。工程实例表明，该模型能够有效挖掘多测点变形数据的时空互补特性，较传统方法在预测精度上具有明显优势，具备良好的工程应用价值。
Abstract：Considering that deformation monitoring series are typically nonstationary, nonlinear and exhibit strong inter-sensor co-variability, direct coupling of multi-point measurements may cause redundancy and obscure effective features. To address these issues, a hybrid forecasting model integrating neural basis expansion analysis for time series(N-BEATS), a fine-grained feature soft-selection mechanism, and bidirectional long short-term memory network(BiLSTM) is proposed. The model first employs the N-BEATS encoder to decompose deformation series into independent trend and seasonal feature components. A fine-grained soft-selection mechanism is then applied to emphasize highly relevant components, enabling high-quality fusion of multi-point deformation information. Finally, the fused features are fed into a BiLSTM network to capture temporal dependencies and generate predictions. An engineering case demonstrates that the proposed model can effectively exploit the spatiotemporal complementarity of multi-point deformation data and achieves superior predictive performance compared with conventional models, indicating strong potential for practical engineering applications.
[1] 张孟昕，陈波，刘伟琪，等.SSA-XGBoost与时空特征选取的大坝变形预测模型[J].水力发电学报，2024,43(1):84-98.(ZHANG M X,CHEN B,LIU W Q,et al.Dam deformation prediction model selected by SSA-XGBoost with temporal and spatial features[J].Journal of hydroelectric engineering,2024,43(1):84-98.(in Chinese))
[2] 漆一宁，苏怀智，姚可夫，等.耦合时序特征分解筛选的大坝变形分析模型[J].水力发电学报，2023,42(7):56-68.(QI Y N,SU H Z,YAO K F,et al.Dam deformation analysis model based on characteristic decomposition screening of coupling time series[J].Journal of hydroelectric engineering,2023,42(7):56-68.(in Chinese))
[3] 周煜，王煜，谭雪怡，等.基于RIME-LSTM-Transformer的混凝土坝变形预测模型[J].水电能源科学，2025,43(12):168-171,181.(ZHOU Y,WANG Y,TAN X Y,et al.Concrete dam deformation prediction model based on RIME-LSTM-transformer[J].Water resources and power,2025,43(12):168-171,181.(in Chinese))
[4] 张芷玥，郑浩然，漆一宁，等.基于时变效应——能量熵降噪的BiLSTM组合混凝土坝变形预测模型[J].水电能源科学，2025,43(8):147-151.(ZHANG Z Y,ZHENG H R,QI Y N,et al.BiLSTM composite concrete dam deformation prediction model based on time-varying effect-energy entropy noise reduction[J].Water resources and power,2025,43(8):147-151.(in Chinese))
[5] 仵凡，郑浩然，漆一宁，等.基于VAE-GRU和残差注意力的大坝位移深度学习预测模型[J].水电能源科学，2025,43(8):152-156.(WU F,ZHENG H R,QI Y N,et al.Dam displacement prediction model based on VAE-GRU and residual attention mechanism[J].Water resources and power,2025,43(8):152-156.(in Chinese))
[6] 吕国旭，陈波，张孟昕，等.基于VMD-AOA-BiLSTM的大坝变形预测模型[J].水电能源科学，2025,43(7):148-152.(LV G X,CHEN B,ZHANG M X,et al.Dam deformation prediction model based on VMD-AOA-BiLSTM [J].Water resources and power,2025,43(7):148-152.(in Chinese))
[7] 吴中如.水工建筑物安全监控理论及其应用[M].北京：高等教育出版社，2003:74-77.(WU Z R.Safety monitoring theory & its application of hydraulic structures[M].Beijing:Higher Education Press,2003:74-77.(in Chinese))
[8] 程旭初，刘景霞，康荣凯.基于改进STGCN与N-BEATS的风功率超短期预测[J].现代电子技术，2025,48(8):115-121.(CHENG X C,LIU J X,KANG R K.Wind power ultra-short-term prediction based on improved STGCN with N-BEATS[J].Modern electronics technique,2025,48(8):115-121.(in Chinese))
[9] 汪正华，王予，倪梓豪，等.基于多尺度冗余特征的轻量化特征融合模块[J].信息与控制，2025,54(6):801-811.(WANG Z H,WANG Y,NI Z H,et al.Lightweight feature fusion module based on multi-scale redundant features [J].Information and control,2025,54(6):801-811 .(in Chinese))
基本信息:
DOI：10.20040/j.cnki.1000-7709.2026.20252213
中图分类号:TV698.11;TV642.4
引用信息:
[1]穆永波,张彩荷,漆一宁,等.基于分解重构与自适应特征融合的混凝土拱坝变形监控模型[J].水电能源科学,2026,44(10):167-173.DOI:10.20040/j.cnki.1000-7709.2026.20252213.
基金信息:
国家重点研发计划(2024YFC3210701); 国家自然科学基金青年科学基金项目(C类)(52509161); 江苏省自然科学青年基金项目(BK20251470)
2025-12-26
2025
2026-07-24
2025-12-30
2025
1
2026-01-22
2026-01-22
2026-01-22