Page 38 - 2025年第56卷第9期

889-897，909.
［ 19］ 宋文龙，冯天时，陈龙，等 . 基于遥感指数与深度学习的黄河冰凌遥感监测识别分析［J］ 中国水利水电科学
.
研究院学报（中英文），2024，22（1）：28-35.
.
［ 20］ RUMELHART D E， HINTON G E， WILLIAMS R J. Learning representations by back propagating errors［J］
Nature，1986，323（6088）：533-536.
/
［ 21］ HU J，SHEN L，SUN G. Squeeze-and-excitation networks［C］/ Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition. 2018.
［ 22］ WOO S， PARK J， LEE J， et al. CBAM： convolutional block attention module［J/OL］ Eprint Arxiv， 2018，
.
11211：3-19. https：/arxiv. org/abs/1807. 06521.
/
/
［ 23］ SZEGEDY C，VANHOUCKE V，IOFFE S，et al. Rethinking the Inception Architecture for Computer Vision［C］/
2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR） 2016.
.
.
［ 24］ HOCHREITER S， SCHMIDHUBER J. Long short-term memory［J］ Neural Computation， 1997， 9（8）： 1735-
1780.
［ 25］ CHUNG J，GULCEHRE C，CHO K H，et al. Empirical evaluation of gated recurrent neural networks on sequence
.
modeling［J/OL］ Eprint Arxiv，2014. https：/arxiv. org/abs/1412. 3555.
/
/
［ 26］ HE K， ZHANG X Y， REN S Q， et al. Deep residual learning for image recognition［C］/Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition. 2018.

Intelligent monitoring and early warning method for high dam discharge structure safety
based on multi-source heterogeneous data fusion

MA Bin，PENG Zhi，LIANG Chao

（Tianjin University，State Key Laboratory of Hydraulic Engineering Intelligent Construction and Operation，Tianjin 300350，China）

Abstract：The discharge structure of high dams will inevitably be damaged during the long operation and mainte⁃
nance period. It is urgent to implement effective safety monitoring and early warning to avoid local abnormalities from
expanding into safety accidents. Since the monitoring items，including air sound pressure and flow pattern images，

are very sensitive to abnormal operating states of high dams，they are monitored synchronously with the low-

frequency vibration displacement to enrich the types of monitoring data and improve effective information. A feature-
level fusion is proposed to splice the time-frequency images of vibration and sound pressure with the segmented and
cropped flow pattern images in the additional dimension，so as to retain and fuse the key features of the above multi-

source heterogeneous data as much as possible. Based on the framework of autoencoder，a deep learning network is

built，Inception and GRU modules are embedded to improve the spatial and temporal feature learning capabilities of

the model，and then the Autoencoder-Inception-GRU single-classification anomaly recognition model is proposed.
Absolute mean error percentage and Euclidean distance are used as the reconstruction error function of the model，
and 95% of their maximum values are set as the anomaly threshold. Based on the prototype monitoring experiment，a

multi-source heterogeneous database of vibration-sound-image was constructed， and the performance of the

Autoencoder-Inception-GRU model was analyzed in detail. The accuracy，robustness and generalization ability of the

proposed model were tested and investigated by case studies under various conditions. The results show that the pro⁃
posed approach achieves excellent performance，which provides key technical support for engineering application.

Keywords：high dam flood discharge；monitoring and early warning；single classification anomaly recognition；

multi-source heterogeneous data fusion；prototype monitoring
（责任编辑：鲁 婧）

— 1142 —