Page 84 - 水利学报2021年第52卷第1期

全专业委员会 2014 年会暨“大坝安全监测技术现状与展望”学术交流会论文集 . 中国水力发电工程学会，
2014 .
［ 35］ 常广，鄢素云，王毅 . 零相位数字滤波器在非平稳信号处理中的应用［J］. 北京交通大学学报，2011，35
（6）：49-56 .
［ 36］ 李双平，张斌 . 基于小波与谱分析的大坝变形预报模型［J］. 岩土工程学报，2015，37（2）：374-378 .
［ 37］ LIU W J，PAN J W，REN Y S，et al . Coupling prediction model for long-term displacements of arch dams based
on long short-term memory network［J］. Structural Control and Health Monitoring，2020 . doi：10.1002/stc.2548.
［ 38］ GOODFELLOW I，BENGIO Y，COURVILLE A . Deep learning［M］. Cambridge，Massachusetts：MIT Press，
2016 .
［ 39］ GAMSE S，HENRIQUES M J，OBERGUGGENBERGER M，et al . Analysis of periodicities in long-term dis⁃
placement time series in concrete dams［J］ . Structural Control and Health Monitoring，2020 . doi：10.1002/
stc.2477.

Safety monitoring model of hydraulic structures and its optimization based
on deep learning analysis

1
3
1
2
REN Qiubing ，SHEN Yang ，LI Mingchao ，KONG Rui ，LI Minghao 1
（1. State Key Laboratory of Hydraulic Engineering Simulation and Safety，Tianjin University，Tianjin 300354，China；
2. China Three Gorges Corporation，Beijing 100038，China；
3. Northwest Engineering Corporation Limited，PowerChina，Xi’an 710065，China）
Abstract： With the development of automation technology for safety management of hydraulic structures，
big data characterized by richness， diversity and complexity has gradually become a significant feature of
safety monitoring system of hydraulic structures. The commonly used mathematical models of safety monitor⁃
ing （three conventional models and shallow learning algorithms） are difficult to extract the deep underlying
information automatically from large amounts of data， i.e. the shallow model is incompatible with big data
mining and analysis. Deep learning algorithm is composed of multiple nonlinear mapping layers， which can
learn the essential characteristics of input data layer by layer and complete the high-level abstraction， but
it also has some problems such as poor engineering applicability. To address this issue，this paper summa⁃
rizes the features of safety monitoring big data，introduces long-term short-term memory （LSTM），and pro⁃
poses an optimized deep analysis model for safety monitoring of different types of hydraulic structures. The
model takes competitive learning mechanism as the core， adopts digital filtering， limited interval and roll⁃
ing iteration to improve LSTM from three aspects of front-end processing， network structure and epitaxial
prediction. It also achieves optimization modeling through random search and step verification. Combining
with engineering projects，several groups of measured data of different effect quantities were selected as typ⁃
ical application scenarios， and the effectiveness of the proposed method has been verified and evaluated
through simulation and comparison experiments. The results indicate that compared with the shallow model，
the deep model is more suitable for safety monitoring big data processing in most scenarios，so as to pro⁃
vide decision support for the safe operation of hydraulic structures.
Keywords：hydraulic structure；safety monitoring；deep learning；long short-term memory networks；intelli⁃
gent analysis

（责任编辑：杨 虹）

— 80 —