Application of Artificial Intelligence in Hydraulic Engineering

1. Introduction to the Special Issue

Water conservancy projects have always been essential throughout the development of human society, including the development and utilization of water resources, the construction and management of water conservancy facilities and flood prevention and control [1]. With the continuous development and popularization of artificial intelligence technology, its application in water conservancy engineering has been increasingly emphasized. Introducing artificial intelligence technology can improve the efficiency and precision of water conservancy projects, lead to the more effective use and protection of water resources, and enhance the safety and intelligence of water conservancy facilities [2]. The application of intelligent algorithms in water conservancy engineering has become an important research area to solve significant scientific problems in the engineering field [3]. Their advantages in regression, classification, clustering and dimensionality reduction have led to their wide use in optimization design [4], structural simulations [5], safety monitoring [6] and water conservancy project evaluations [7]. In traditional research methods, both experiments and numerical simulations are limited by time and cost. However, with the advancement of sensors and measurement technology, a large amount of data has accumulated for use in water conservancy project safety monitoring. Intelligent algorithms have become a powerful tool for mining information, constructing data associations and monitoring data quickly and accurately. Combining intelligent algorithms with traditional computational techniques such as geotechnical testing, nondestructive testing and numerical simulations has been essential for understanding various laws and mechanisms in water conservancy projects.

Therefore, this Special Issue of “Water” aims to deeply understand the application value of intelligent algorithms in water conservancy projects and their important role in improving a project’s safety and efficiency and to provide new ideas and methods for the development and progress of these projects. This has great significance for improving the safety of water conservancy projects and the development of human society.

Since the call for papers was announced in January 2023, 15 original papers have been accepted for publication after a rigorous peer-review process (contributions 1–15); these papers can be classified into several areas: data-driven intelligent model studies, intelligent algorithms for optimal structural design, nondestructive testing (NDT) sensor applicability studies and simulations and experiments for structural characterization. To better understand this Special Issue, we summarize the highlights of the published papers below.

2. Overview of the Contributions to this Special Issue

Regarding data-driven intelligent model research, Shao et al. comprehensively analyze the effects of water level load transfer, rockfill flow and soil properties on settlement during operation (contribution 1). The original position coordinates are replaced by space parameters which are more consistent with the deformation characteristics, such as the elevation of the upper fill, the thickness of the rockfill materials, the distance between the measuring point and the panel, etc. A multi-monitoring point (MMP) model is combined with the XGBoost model, which has a high prediction accuracy for concrete-faced rockfill dams (CFRDs). Then, the importance of the selected factors is determined. This work has a certain reference value for dam safety monitoring. A new method for inverse analyses of permeability coefficients is proposed by Zhao et al. (contribution 2). They constructed an inversion sample set of a permeability coefficient using the combination of a finite element model and orthogonal test design. The random forest (RF) algorithm proxy model is established, and the Harris Hawk optimization (HHO) algorithm is used to determine the optimal value of the permeability parameters in the project area. This method is used to explore and verify the distribution of natural seepage fields in P hydropower stations. An artificial neural network (ANN) is proposed by Saya et al. (contribution 3) to predict future water consumption as a function of some environmental parameters, and the Copernicus Climate Change Service (C3S) is used to determine the water demand trend in the next ten years. Finally, future consumption is predicted based on the ANN model, the continuity equation of the tank is solved through integral discretization and the time series of tank volume changes and the total number of crisis events are obtained, which are used to estimate the optimal capacity of small city reservoirs. In terms of research on structural safety detection methods, an intelligent detection method for concrete dam surface cracks based on two-stage migration learning is proposed by Li et al. (contribution 4). First, two-stage transfer learning across domains and within domains transfers relates domain knowledge to the target domain so that the model can be adequately trained with small data sets. Second, the segmentation capability is enhanced using the residual network 50 (ResNet50), as a UNet model feature extraction network to enhance crack feature information extraction. Finally, multilayer parallel residual attention (MPR) is integrated into the jump connection path to improve the focus on critical information for clearer fracture edge segmentation. This concrete dam surface crack detection method exhibits a high efficiency and accuracy and a strong robustness. A convolutional neural network (CNN) model is used by Peng et al. (contribution 5) to achieve accurate segmentation of coarse-grained soil CT images, exceeding the accuracy of traditional methods and uncovering a new method for soil particle size analyses. The validity of the CNN model is proven by a three-dimensional (3D) model reconstructed from segmented images, highlighting its accuracy and potential for automation in soil particle analyses. New empirical formulas for the ideal particle size and discount coefficient in coarse-grained soil are also revealed and verified. In terms of risk analysis and comprehensive model agent evaluation methods, a feedback, rolling and adaptive operational decision-making mechanism is proposed by Han et al. (contribution 6). The relationship between the work cycle and multiple time scales is considered according to the change in time scale and the dynamics of the work process. The risk transmission mechanisms of urban river ecological governance engineering projects are examined by Xu et al. (contribution 7). Prediction is combined with operation to adapt to dynamic changes in multiple time scales during operation. Based on existing research, the internal mechanism, the influencing factors of risk transmission and the dynamic evolution process of risk are considered. Based on the theory of infectious disease dynamics and the susceptible exposure to infection recovery susceptibility (SEIRS) risk transmission model, a risk transmission delay model of an urban river ecological governance project under a scale-free network is established. This model provides a basis for the effective supervision and control of project participants and theoretical support for indirect and post-event supervision. In-depth literature reviews and brainstorming are used by Xu et al. (contribution 8) to identify 63 risk elements in urban river ecological management projects. An expert survey method is used to identify the correlation between risk factors, and the risk factors are taken as network nodes. Then, the relationships between these nodes are used as network edges (i.e., paths) to construct a complex network model. The risk network’s overall characteristic parameters and local characteristic parameters are analyzed using the network visualization tool. According to the parameter characteristics, the risk transmission characteristics of urban river ecological governance projects are analyzed, and the internal relationship of risk transmission in complex networks is revealed. A corresponding comprehensive evaluation index system based on interval number theory is built by Chen et al. (contribution 9). Expert evaluations and the improved analytic hierarchy process (AHP) are combined to propose a reasonable allocation method for index weights in addition to a comprehensive evaluation model. Comprehensive evaluation criteria are formulated. Taking a roller compacted concrete (RCC) dam in China as an example, the applicability of the comprehensive evaluation method based on interval number theory is illustrated.

Regarding the research on intelligent algorithm optimization structure design, a noise reduction method combining the improved adaptive singular value decomposition algorithm (ASVD) and the improved complete ensemble EMD with adaptive noise (ICEEMDAN) is proposed by Wang et al. (contribution 10). A Hankel matrix is constructed based on collected discrete-time signals. After performing SVD on the Hankel matrix, the ASVD algorithm automatically selects the practical singular values to filter out most of the background white noise and retain helpful frequency components with a similar energy in the signal. Then, the ICEEMDAN is combined with the Spearman correlation coefficient method to further to filter out residual white noise and low-frequency water flows. Based on the practically measured vibration signals of a floodgate at a large hydropower station, the results show that the ASVD-ICEEMDAN method exhibits a good noise reduction performance and feature information extraction abilities for floodgate vibration signals and can provide support for operational mode analysis and damage identification for practical structures under complex interference conditions. The characteristics of low-specific-speed centrifugal pumps are simulated and analyzed using CFD by Ke et al. (contribution 11) by considering that the inlet width of the splitter blade is more significant than the inlet width of the primary blade. The influence of the splitter blade’s geometric parameters on the flow field is studied in orthogonal experiments, and the geometric parameters of the separator blade are optimized by using the artificial fish swarm algorithm. Finally, the accuracy of the algorithm is verified by experiments. This study provides a reference for the shape selection of splitter blades.

In terms of the applicability study of nondestructive detection sensors, Yu et al. (contribution 12) conduct model tests and quantitative analyses on the leak monitoring performance of a heated temperature sensing cable, and the temperature change curves in different media are obtained. The ability of the heated temperature sensing cable to identify leaks in soil media with additional moisture content is verified. Finally, through a field simulation test at a manganese slag reservoir, it is verified that the cable can be used for leak monitoring in an actual karst depression reservoir basin. A volumetric meter using flow control valves that are ordinarily already present in buildings’ hydraulic installations is proposed by Gonçalves et al. (contribution 13). A system is developed for the electromechanical and thermal characterization of the sensor. Computational simulations of the sensor are performed in Ansys^® (2021R2) software, and the electronic circuit is designed in LTSpice^® (v. 17.0.30.0) software. An artificial neural network is used to estimate the flow and volume from trapezoidal pulses.

In terms of simulations and tests of structural states, a process model of the impact of construction load on a reservoir cutoff wall in a reconstruction section of an expressway is established using FLAC3D two-dimensional and three-dimensional numerical simulation methods by Sun et al. (contribution 14). The stress–strain state of the expressway reconstruction section and the nearby reservoir cutoff wall is simulated in detail, directly reflecting the overall deformation of the reservoir cutoff wall and the interaction and differential deformation between the wall structures. The influence of the construction load on the safety and stability of the cutoff wall of the reservoir is evaluated, and various advanced mechanical behaviors of the cutoff wall are predicted. A simulation method for the mechanical behavior of tight rock is proposed by Tang et al. (contribution 15) by combining physical experiments with numerical analyses and considering fluid mechanics. The validity of the proposed method is verified by comparing the numerical and physical results of triaxial shear tests on sandstone under seepage conditions. Based on this verified method, the stability of the surrounding tight sandstone underground water sealed oil and gas storage caverns during excavation is analyzed.

3. Conclusions

The guest editors envision that the papers published in this Special Issue will be of interest to researchers, designers and practitioners for the safety monitoring and management of reservoirs and dams and will help identify further lines of research. We also hope that readers will find the material in this Special Issue both interesting and stimulating as they explore the application of AI in hydraulic engineering from the perspective of monitoring models, optimization algorithms, nondestructive sensors, numerical simulations and experiments. The research results and methods introduced in this Special Issue, including data-driven models and mechanisms, structural design optimization, nondestructive testing sensors and structural safety simulations, are of great research significance. These technological contributions can help related scholars and project managers analyze and manage the safety of the major structures in reservoir dams.