Safety early warning method and intelligent monitoring system for cold region gate dam project

Abstract

This invention discloses a safety early warning method and intelligent monitoring system for dam projects in cold regions. It adopts an edge-cloud collaborative architecture, where intelligent sensors use compressed sensing technology to perceive dam load and structural response information. This information is then uploaded to a cloud computing center after being processed by an edge server for denoising, feature extraction, physical quantity calculation and conversion, and on-site discrimination. The computing center uses multi-source information fusion for intelligent discrimination or multi-field coupled numerical simulation based on model correction. The intelligent comprehensive discrimination uses TimeGAN and MAD-GAN models; the numerical simulation uses a water-thermal-mechanical multi-field coupled model for water, temperature, wind, rain, and snow loads. After the system and model stabilize, the early warning indicators, constraints, and enhanced samples are periodically updated and deployed to the edge terminal through operational condition identification, facilitating rapid early warning using relevant data and models such as TimeGAN at the edge terminal. This invention demonstrates outstanding innovation and practicality, and has good potential for transformation and application.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy project management technology in cold regions, specifically relating to a safety early warning method and intelligent monitoring system for dam and gate projects in cold regions. Background Technology

[0002] With the increasing occurrence of extreme weather events and the continuous construction of dams in high-altitude and cold regions, safety engineering and early warning systems for dams in these areas have become a major concern for researchers, design units, and operation and management personnel. Due to drastic temperature fluctuations, water-retaining structures in high-altitude and cold regions are more prone to deformation, and the deformation can be significant. Given the harsh climate and transportation conditions, early warning for non-regulating reservoirs in these regions is currently difficult, and personnel management is challenging, requiring high-level early warning capabilities. Furthermore, high-altitude and cold regions are often remote and harsh environments. If all information for water-retaining structures in such environments relies on remote processing, the dependence on the communication network between the site and the remote facility becomes too high, making it difficult to adapt to the safety management requirements and development trends of dams in remote areas, and even more difficult to achieve real-time early warning. Summary of the Invention

[0003] In view of the problems existing in the current technology, the present invention discloses a safety early warning method and intelligent monitoring system for dam projects in cold regions.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows:

[0005] Safety early warning methods for dam projects in cold regions include the following steps:

[0006] Step 1: Collect measured data on temperature, ice thickness, water level, and deformation in the dam area to form multi-sample time series data; conduct correlation and distribution tests on the sample measurements; when the number of samples is insufficient to meet the above test requirements, use the TimeGAN generative adversarial network model, which takes the influencing factors of the measured values ​​as input and the limited measured samples as the true values, to generate and supplement the samples until the sample number requirement is met.

[0007] Step 2: Select measured ice thickness data and compare it with the ice thickness selected for the design load. Ice thicknesses greater than the design ice thickness are the ice thickness extreme value sequence. When the design does not consider ice load, all measured data with ice thickness greater than 0 are taken as extreme values, thus obtaining the ice thickness extreme value sequence. Calculate the average water level based on the measured data, and use the design water level as the high water level threshold. Water levels greater than the selected design water level are considered high water levels, thus obtaining the high water level extreme value sequence. When there is no continuous measured ice thickness data, ice thickness data is calculated based on temperature data using empirical formulas.

[0008] Step 3: Based on the two sets of extreme value sequences obtained in Step 2, namely the ice thickness extreme value sequence and the high water level extreme value sequence, use the corresponding probability density estimation method to obtain their respective probability density functions according to the test results in Step 1, and perform goodness-of-fit test until the test is passed.

[0009] Step 4: Obtaining the joint probability function of ice thickness and water level; The joint probability distribution of ice thickness and water level acting on the dam is determined by the bivariate Copula function; The form of the joint probability distribution Copula function is determined based on the correlation and normality tests between variables. Under severe conditions, water level and ice thickness are positively correlated, so the Gumbel-Hougarard Copula function is selected, and the function coefficients are derived from the Kendall rank correlation coefficient.

[0010] Step 5: Determine the critical probability using the reciprocal of the flood return period in the dam project design. The flood return period is in years. Based on the joint probability density function determined in Step 4, calculate whether the joint probability density function of maximum ice thickness and high water level during the project's operational life is greater than the critical probability. If it is not greater than the critical probability, continue collecting data and return to Step 1; otherwise, proceed to the next step.

[0011] Step 6: When structural safety risks exist, it is necessary to check the completeness of monitoring items and measuring points such as structural strength, seepage, and deformation response. When the monitoring items and measuring points are complete, the TimeGAN generative adversarial network model is first used in the cloud computing center in combination with trend change, over-prediction confidence method or numerical calculation critical value method for single measuring point early warning. The output of the TimeGAN model is the measured structural response information of the dam at the measuring point, and the input is the corresponding load information, including wind direction, wind force, ice thickness, water pressure, temperature, time, snow load, etc. Then, the MAD-GAN generative adversarial network is used to conduct early warning for multiple monitoring items and multiple measuring points of the dam section and the entire dam body, as well as for local and overall dam early warning. The input of MAD-GAN is the measured response value of multiple measuring points, and the output is the judgment of the structural safety of the dam section / dam as a whole.

[0012] If the items for monitoring structural response are incomplete, proceed to the next step;

[0013] Step 7: Select the appropriate water-thermal-mechanical coupling model considering frost heave based on the dam material and structural type. This model considers the effects of ice expansion and phase change on the stress and deformation of the dam and its foundation. The water-thermal-mechanical coupling model includes: an unsteady-state heat conduction equation accompanying the phase change, using the sensible heat capacity method to describe the phase change; the water migration control equation is determined by combining the mass-conserved fluid continuity equation with the saturated-unsaturated Forchheimer's permeability law with a starting pressure gradient, combined with the temperature gradient, water diffusivity, and ice content; material strain considers temperature, humidity, stress-strain, and creep; the constitutive model adopts a fractional-order constitutive model; the material yield criterion adopts the three-shear strength criterion, and crack initiation adopts strain energy release. The calculation process employs a rate criterion; combined with the overall equilibrium method, a water-thermal-mechanical coupling model is obtained; load conditions are based on measured data from the sensing system, including air and water temperature, ice thickness, wind direction and speed, snow thickness and its distribution; during numerical simulation, the surface ice pressure, lower water pressure, and sediment pressure of the upstream / downstream reservoir water are input, while also considering the temperature stress of the dam body itself, the phase change effect of water and ice, and the frost heave force and volume force; local and overall safety early warnings for the dam are conducted based on material strength criteria, seepage failure criteria, and overall yield stability criteria; simultaneously, the calculated critical state deformation values ​​and seepage pressure values ​​are deployed to the edge end as early warning indicators, and the samples obtained during the calculation process are used for sample enhancement in the probability statistical test at the edge end.

[0014] Furthermore, in step 2, the ice thickness is calculated using the following formula:

[0015]

[0016] In the formula, h is the ice thickness; I is the freezing day; a is the empirical value under different conditions; for lake ice caps without snow cover and with wind blowing, a = 2.7 cm / ℃·d; for lake ice caps without snow cover, a = 1.7~2.4 cm / ℃·d; for lake ice caps with snow cover, a = 1.4~1.7 cm / ℃·d.

[0017] For dams in areas with frequent temperature changes, the following formula is used:

[0018]

[0019] In the formula, a(T) is a variable related to temperature.

[0020] Furthermore, when there is no actual measured temperature data at the dam site, the temperature is obtained by interpolation using temperature data from the meteorological department.

[0021] Furthermore, the generative networks in the TimeGAN and MAD-GAN models employ extended recurrent neural networks (adRNNs) with attention mechanisms.

[0022] Furthermore, steps 1-4 are completed on-site through on-site edge computing.

[0023] Furthermore, steps 5-7 are completed through a cloud computing center.

[0024] This invention also provides a smart monitoring system for the safety of dam projects in cold regions, comprising: sensors, field computing devices, edge computing gateways, and a back-end server. Sensors collect data, and multiple field computing devices form a field edge computing network, each connected to a sensor and storing relevant data on-site. The field edge computing network performs data error identification, processing, and analysis, compares and issues warnings based on the latest warning indicators issued by the cloud computing center, and uploads the warning results to the cloud computing center or relevant authorized nodes. The edge computing gateway collects data from each edge-end field computing device, processes it, and uploads it to the cloud computing center. The cloud computing center collects information from multiple edge gateways, employs multi-field coupled numerical analysis (finite element method), model correction and data assimilation, and parameter inversion to calculate and obtain deformation, seepage, and stress-strain at the structural failure critical point as single-point, local, or overall warning indicators, serving as a basis for data rationality analysis of the edge components.

[0025] Furthermore, the sensors include: several air temperature and wind speed sensors, rainfall and snowfall sensors, water level sensors, water temperature sensors, ice thickness sensors, strain sensors, crack measuring sensors, inclinometers, dam body temperature sensors, and seepage pressure sensors installed at the site of the water-retaining structure; the air temperature and wind speed sensors collect wind speed, direction, and temperature data at the site of the water-retaining structure and then store the data in an embedded field computing and storage device via wireless or wired networking; the rainfall and snowfall sensors collect rainfall and snow thickness data and then store the data in an embedded field computing and storage device via wireless or wired networking; the water level and water temperature sensors obtain the vertical water level and temperature distribution of the ice layer and water body; the ice thickness gauge measures the real-time thickness of the ice; the strain sensor, crack measuring sensor, and inclinometer monitor deformation and seepage at the site of the water-retaining structure and measure the tilt angle and displacement of the dam structure in the horizontal X and Y directions, and the collected data is stored in an embedded field computing and storage device via Zigbee or wired networking; the dam body temperature sensor obtains the internal temperature distribution of the dam, and the seepage pressure sensor obtains the internal seepage pressure when it has not yet frozen.

[0026] The beneficial effects of this invention are as follows:

[0027] 1. This invention adopts an edge-cloud collaborative early warning architecture. Intelligent sensors use compressed sensing to collect data on environmental loads and structural response elements of dams. Edge servers perform noise reduction, feature extraction, or physical quantity calculation and conversion before uploading the data to a cloud computing center. The center performs intelligent discrimination or multi-field coupling calculation analysis based on measured information. Intelligent discrimination uses a time-series-based generative adversarial network (MAD-GAN) model. Multi-field coupling analysis uses a thermo-solid-liquid coupling model considering temperature, wind, and snow loads, combined with strength and stability yield criteria. Through modal recognition, sensitive and important early warning indicators are periodically updated and deployed to the edge, facilitating rapid edge early warning and improving the intelligence level of dam safety monitoring and early warning. This invention overcomes the shortcomings of traditional data center and cloud computing models, avoids excessive reliance on the network environment, and ensures timely response and safety early warning in emergencies. This invention is highly innovative and practical, with good potential for transformation and application.

[0028] 2. The data undergoes two compression processes during collection and uploading, which greatly alleviates the problem of narrow communication channel bandwidth with the cloud computing center. At the same time, migrating the early warning from the back-end server to an embedded device near the data generation location can greatly improve the real-time performance of the early warning. Attached Figure Description

[0029] Figure 1 This invention provides an edge computing framework for the intelligent safety monitoring system of cold-region dam projects.

[0030] Figure 2 This is a schematic diagram of the hardware structure of the system of the present invention.

[0031] Figure 3 This is a schematic diagram of the data acquisition and processing flow under the edge computing framework.

[0032] Figure 4 A flowchart of the safety early warning method for cold-region dam projects provided by the present invention.

[0033] Figure 5 This is a diagram of the architecture of an extended recurrent neural network (adRNN) with an attention mechanism. Detailed Implementation

[0034] The technical solutions provided by the present invention will be described in detail below with reference to specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0035] The safety early warning method for cold-region dam projects provided by this invention has the following process: Figure 4 As shown, it includes the following steps:

[0036] Step 1: Collect measured data on temperature, ice thickness, water level, and deformation in the dam area to form multi-sample time series data. Collect measured temperature, ice thickness, upstream and downstream water levels, and deformation values ​​at measuring points for the water-retaining structure / dam area over the past five years. These values ​​should be collected and recorded in advance using various sensors (temperature and wind speed sensors, rainfall and snowfall sensors, water level sensors, ice thickness sensors, and deformation monitoring equipment such as coordinate measuring machines and inclinometers). If the data period for newly constructed water-retaining structures is less than five years, collect all measured data for the entire period and interpolate to extend it to five years. Perform correlation and distribution tests on the sample measurements. If the number of samples is insufficient to meet the above test requirements, use a TimeGAN generative adversarial network model with the influencing factors of the measured values ​​as input and the limited measured samples as the true values ​​to generate supplementary samples until the number of samples required for the following steps is met; the TimeGAN generative network uses an extended recurrent neural network (adRNN) with an attention mechanism. The training of the generative network model in this step is completed in the cloud computing center and is regularly updated and deployed at the edge, with the remaining parts completed at the edge.

[0037] Extended Recurrent Neural Network (adRNN) architecture with attention mechanism, such as Figure 5 As shown in the figure, the main parameters are calculated using the following formula:

[0038] f t =σ(W f x t +V f h t-1 +U f h t-d +b f Equation (3)

[0039] u t =σ(W u x t +V u h t-1 +U u h t-d +b u Equation (4)

[0040] o t =σ(W o x t +V o h t-1 +U o h t-d +b o Equation (5)

[0041]

[0042] Where the subscript t represents the time step; σ represents the logical sigmoid function; x tIt is the input vector; h t-1 and h t-d These represent the most recent and delayed control states; d>1 indicates inflation; W, V, and U are weight matrices; b is the bias vector, and o... t It is the output vector.

[0043] The state c is based on the most recent (c t-1 ), delayed (c t-d ) and candidates The state is calculated as follows:

[0044]

[0045] in This represents the Hadamard product (element-wise product).

[0046] Please note that c-state is a weighted combination of past c-states and new candidate states. The fusion vector f t This determines the mixing ratio of the most recent and delayed c-states, while updating the vector u t This determines the proportion of old and new information in the result state c.

[0047] Based on state c t , Figure 5 The output vector h of the unit shown t and y t (or m) t The following is confirmed:

[0048]

[0049]

[0050] Where s y s h s m and s q Representing vectors y and y respectively t , m t and The length (s in our implementation) h and s q (They are the same). When applied in this invention, the network structure and parameters are set as needed.

[0051] Step 2: Select measured ice thickness data. Compare the ice thickness with the design load. Ice thicknesses greater than the design thickness are identified as the thickest ice extreme value sequence. When the design does not consider ice load, all measured ice thicknesses greater than 0 are considered extreme values, thus obtaining the thickest ice extreme value sequence. Calculate the average water level based on the measured data. Use the design water level as the high water level threshold. Water levels greater than the selected design water level are considered high water levels, thus obtaining the high water level extreme value sequence. When continuous measured ice thickness data is unavailable, ice thickness data is calculated using temperature data (if no local measured temperature data is available, interpolation from meteorological department temperature data can be used). Ice thickness information can be used to estimate ice cover changes using the following formula:

[0052]

[0053] In the formula, h is the ice thickness; I is the number of days the ice freezes; and a is an empirical value under different conditions.

[0054] In the early stages of ice sheet formation, the thickness increases rapidly, resulting in a lower a value. In the later stages of ice sheet formation, the a value is higher. With snow cover, the a value is lower; without snow cover, the a value is higher. For lake ice sheets without snow cover and exposed to wind, a = 2.7 cm / ℃·d; for lake ice sheets without snow cover, a = 1.7–2.4 cm / ℃·d; and for lake ice sheets with snow cover, a = 1.4–1.7 cm / ℃·d.

[0055] For dams in areas with frequent temperature changes, the following formula is used:

[0056]

[0057] a(T) is a temperature-dependent variable. This step is performed by local edge calculations.

[0058] Step 3: Based on the two extreme value sequences obtained in Step 2—the ice thickness extreme value sequence and the high water level extreme value sequence—the kernel density estimation method is used to obtain their respective probability density functions according to the test results in Step 1. A goodness-of-fit test is then performed until the function passes. If the function fails, the parameters are adjusted to ensure it passes. This step is completed by on-site edge calculation.

[0059] Step 4: Obtaining the joint probability function of ice thickness and water level: A bivariate Copula function is used to determine the joint probability distribution of ice thickness and water level acting on the dam. The form of the joint probability distribution Copula function is determined based on the correlation and normality tests between the variables. Under severe conditions, water level and ice thickness are positively correlated; therefore, the Gumbel-Hougarard Copula function is selected, and the function coefficients are derived from the Kendall rank correlation coefficient. This step is completed by on-site edge calculations.

[0060] Step 5: Determine the critical probability using the reciprocal of the flood return period (in years) for the dam project design. Based on the joint probability density function determined in Step 4, calculate whether the joint probability density function of maximum ice thickness and high water level during the project's operational lifespan is greater than the critical probability. If it is not greater than the critical probability, continue data collection and return to Step 1; otherwise, proceed to the next step. This step is completed by the cloud computing center.

[0061] Step 6: When structural safety risks exist, it is necessary to conduct a completeness check on monitoring items and measuring points such as structural strength, seepage and deformation response. The completeness check shall be determined in accordance with the Technical Specification for Appraisal of Dam Safety Monitoring System (SL766-2018).

[0062] When all monitoring items and measuring points are complete, a single measuring point early warning is first performed using a generative adversarial network (GAN) model combined with trend changes, over-prediction confidence methods, or numerical calculation critical value methods. The output of the TimeGAN model is the measured structural response information of the dam at the measuring point, and the input is the corresponding load information, including wind direction, wind force, ice thickness, water pressure, temperature, time, and rain / snow load. Then, a generative adversarial network (MAD-GAN) is used for early warning of multiple monitoring items and multiple measuring points for each dam segment and finally the entire dam body, both locally and for the entire dam. The input of MAD-GAN is the measured response values ​​of multiple measuring points, and the output is a judgment on the structural safety of each dam segment / dam as a whole. The generative networks in both the TimeGAN and MAD-GAN models employ extended recurrent neural networks (adRNNs) with attention mechanisms. According to the "Technical Specification for the Identification of Dam Safety Monitoring Systems" (SL766-2018), if the monitored structural response items are incomplete, the process proceeds to the next step; if the current structural response items are complete, numerical simulation is unnecessary. This step was completed by the cloud computing center and the validity of the calculation results was confirmed by experts.

[0063] Step 7: Select the appropriate water-thermal-mechanical coupling model considering frost heave based on the dam material and structural type. This model considers the effects of ice expansion and phase change on the stress and deformation of the dam and its foundation. The water-thermal-mechanical coupling model includes: an unsteady-state heat conduction equation accompanying the phase change, using the sensible heat capacity method (solid phase increment method) to describe the phase change; the water migration control equation adopts the fluid continuity equation with mass conservation, combined with the saturated-unsaturated Forchheimer permeability law with starting pressure gradient, and determined by temperature gradient water diffusivity and ice content; material strain considers temperature, humidity stress-strain and creep; the constitutive model adopts a fractional constitutive model; the material yield criterion adopts the three-shear strength criterion, and the crack initiation adopts the strain energy release rate criterion; combining the overall equilibrium method yields the water-thermal-mechanical coupling model. The load conditions are calculated based on measured data from the sensing system, including air and water temperature, ice thickness, wind direction and speed, snow thickness and its distribution. Numerical simulations input surface ice pressure, lower water pressure, and sediment pressure from the upstream / downstream reservoirs, while also considering the dam's own temperature stress, water-ice phase transition effects, and frost heave and volume forces. Local and overall safety warnings for the dam are provided based on material strength criteria, seepage failure criteria, and overall yield stability criteria. Simultaneously, the calculated critical state deformation and seepage pressure values ​​are deployed to the edge end as warning indicators. Samples obtained during the calculation process are used for sample augmentation in probabilistic statistical tests at the edge end. This step is completed by the cloud computing center and the validity of the calculation results is confirmed by experts.

[0064] Specifically, the water-thermal-mechanical coupling model equations used in this example are as follows:

[0065]

[0066] The above equations represent the unsteady-state heat conduction equation and the moisture migration control equation, respectively, accompanying the phase transition. In the equations, T is the transient temperature of the object (°C); t is time (s); k is the thermal conductivity of the material (W / (m·°C)); and ρ is the density of the material (kg / m³). 3 ), c p q is the specific heat at constant pressure of the material (J / (kg·℃)). v The internal heat source strength of the material (W / m) 3 ), f s Let θ be the solid fraction of the node, L be the latent heat of phase change of the soil during freezing or thawing (J / kg), x and y be rectangular coordinates (m), θ be the volume content of unfrozen water, and D(θ) be the water diffusion coefficient in the material (cm). 2 / s), D TLet K(θ) be the water diffusivity under temperature gradient, and K(θ) be the water conductivity of the material (cm / s). The unsteady-state heat conduction equation uses the sensible heat capacity method to describe the phase transition. The water migration control equation is determined by combining the mass-conserved fluid continuity equation with the saturated-unsaturated Forchheimer's permeability law with an initiating pressure gradient, along with the water diffusivity under temperature gradient and ice content.

[0067] Fluid continuity equation

[0068]

[0069] This represents the mass conservation relationship, where ρ is the fluid density and v is the flow velocity in the flow field.

[0070] Forchheimer's Law of Osmosis

[0071]

[0072] J is the hydraulic gradient, a1 and a2 are empirical parameters related to the seepage medium and fluid, respectively, and v is the seepage velocity. In the formula, a2 = 0, which reduces the order to Darcy's law. Let λ be the pressure gradient, and λ be the starting pressure gradient.

[0073] This invention designs an intelligent safety monitoring system for dam and gate engineering in cold regions, based on, for example, Figure 1 The edge computing framework shown includes on-site edge devices and a cloud computing center. Data collection, simple calculations, and early warning comparisons are performed at the edge. Extracted, reduced, or compressed measured data is uploaded to the cloud computing center for processing via narrowband communication. The cloud computing center processes information from multiple gateway nodes, leveraging its powerful computing capabilities and employing multi-field coupled numerical analysis (finite element method), model correction and data assimilation, and parameter inversion algorithms. It calculates deformation, seepage, and stress-strain at the structural failure critical point as single-point, local, or overall early warning indicators, and distributes these indicators to the edge gateways, implementing steps 5-7 of the early warning method. The edge gateway devices (i.e., edge devices, including sensors) perform on-site early warnings or measurement rationality checks based on the early warning indicators, implementing steps 1-4 of the early warning method, and sending the results to relevant nodes or terminals in real time, thereby improving the intelligence level of dam safety monitoring and early warning. This method can compensate for the shortcomings of traditional data center and cloud computing models, avoid excessive reliance on the network environment, and ensure timely response and safety early warning in emergencies.

[0074] The hardware structure of the system of this invention is as follows: Figure 2As shown, the system includes sensors, field computing devices, edge computing gateways, back-end servers, and communication networks. In this example, the water-retaining structure is a dam. Multiple embedded field computing devices form a field edge computing network, which connects to various necessary sensors (such as thermometers, laser anemometers, weighing rain and snow gauges, pressure water level gauges, ice thickness sensors, coordinate instruments capable of measuring structural deformation, inclinometers, tiltmeters, water thermometers, dam body thermometers, and seepage pressure sensors) and stores relevant data on-site. The embedded system performs data error identification, processing, and analysis. Based on this, it compares and issues warnings according to the latest warning indicators issued by the cloud computing center, and uploads the warning results to the cloud computing center or relevant authorized nodes (including mobile terminals). The embedded system uses an embedded processor as its core and is designed with waterproofing, moisture-proofing, and wide temperature variation capabilities to adapt to long-term field water conservancy work environments. Typical commercial embedded devices operate normally at temperatures ranging from -20°C to +70°C. Considering the extreme temperature conditions at the site, industrial-grade embedded devices with an operating temperature range of -40°C to +85°C were selected. An IMAX6Q quad-core industrial-grade development board was used to connect the sensors and convert the physical information sensed by the sensors into data storage. The Cortex-A9 quad-core processor is based on the ARM architecture, equipped with 8GB of eMMC local storage and 1GB of DDR3 memory, offering fast storage speeds and providing multiple interfaces for connecting peripheral devices. Furthermore, a Bluetooth communication module was added to the embedded device to enable periodic data transfer at the site. Because different types of sensors are deployed in different environments, a single communication protocol is insufficient to meet actual needs. Therefore, in terms of communication methods, LoRa communication was chosen for data measurements such as air temperature over long distances within the dam area, while the lower-cost ZigBee communication was used for data collection such as water level outside the dam, which has a shorter communication distance. For deformation and seepage monitoring within the dam's corridors, a faster and more reliable mesh networking approach was adopted, sharing all node resources.Specifically, the system includes several air temperature and wind speed sensors, rainfall and snowfall sensors, water level sensors, water temperature sensors, ice thickness sensors, strain sensors, crack sensors, inclination sensors, dam body temperature sensors, and seepage pressure sensors installed at the water-retaining structure site. The air temperature and wind speed sensors collect wind speed, direction, and temperature data at the water-retaining structure site and store it via LoRa / wired networking to an embedded field computing and storage device. The rainfall and snowfall sensors collect rainfall and snow thickness data and store it via LoRa / wired networking to an embedded field computing and storage device. The water level and water temperature sensors obtain ice thickness data. The system monitors the vertical water level and temperature distribution of the water body. Ice thickness gauges measure the real-time thickness of ice, and the data is stored in an embedded field computing and storage device via Zigbee / wired networking. Strain sensors, joint sensors, and inclinometers monitor deformation and seepage at the water-retaining structure, and measure the tilt angle and displacement of the dam structure in the horizontal X and Y directions. The collected data is stored in an embedded field computing and storage device via Zigbee or wired networking. Dam body temperature sensors are used to obtain the internal temperature distribution of the dam, and seepage pressure sensors are used to obtain the internal seepage pressure before freezing, achieving real-time data acquisition. See Appendix. Figure 2 All of the above sensors can provide real-time warnings when data is abnormal.

[0075] An Ubuntu system is installed on the edge embedded device to schedule hardware resources. Based on the Linux kernel, it provides rich basic functions and reusable interfaces. The lightweight splite3 database is used for field embedded data storage. It has a small operating space, is controlled by C language code, and is written using the Source Insight editor. The code is migrated to the corresponding directory on the Ubuntu system for cross-compilation. Finally, the executable file is downloaded to the embedded device via NFS network sharing, and multi-task parallel computation is performed according to the schedule.

[0076] The edge computing gateway aggregates data from various edge computing devices, processes it briefly, and uploads it to the cloud computing center. Unlike embedded field computing and storage devices that directly connect to dam monitoring instruments, the edge computing gateway does not directly interact with the instruments. Instead, it connects to a wireless sensor network composed of embedded field computing and storage devices via a wireless interface. It connects to corresponding sensor networks through different network protocol interfaces, aggregating and reporting early warning information. Using a civilian-grade BeiDou communication card, it relays early warning information to the cloud computing center via the BeiDou short message system, providing a basis for dam safety management decisions. The core computing module of the edge computing gateway uses the same configuration as the embedded field computing and storage devices, but it has more communication interfaces for connecting to different networks.

[0077] The back-end server is a DELL T40 server, featuring an Intel i3-9100 quad-core 64-bit processor and 8GB of memory. The CentOS 7 operating system, also based on the Linux kernel, supports application virtualization technology that completely isolates applications from the operating system, resulting in a longer and more stable lifecycle. Storage is handled by a MySQL database, storing deployment parameters and received historical information from each edge node. MySQL is user-friendly, stable, and provides logging services; it supports multiple operating systems, offering strong portability and APIs for various programming languages. A Python compilation environment is added to the server, using Python 3 and PyCharm to write scripts that interact with the database. The MySQL package is installed to connect to the database, and SQL statements are used to insert the aggregated early warning information received by the BeiDou short message system into the historical database according to a set pattern. Each acquisition node is assigned a primary key in the MySQL database, and data tables are created through natural join queries. Combined with the deployment location information of the embedded devices in the database, historical early warning information for a specific area can be retrieved. Specific system equipment parameters are shown in Table 1.

[0078] Table 1 Hardware Equipment List and Key Technical Parameters

[0079]

[0080] Figure 3 This describes the data acquisition and processing flow within an edge computing framework. The first stage of data processing occurs on embedded field processing devices at the edge, filtering out monitoring instrument noise and abnormal values, performing simple analysis and calculations, and setting up an early warning model for real-time alerts. The second stage occurs at the edge computing gateway, aggregating and compressing key information from multiple embedded field processing devices, and then uploading it sequentially to the cloud computing center via low-speed satellite or public network communication such as 4G / 5G. The cloud computing center then performs calculations such as finite element analysis based on the aggregated information, distributing the early warning indicators to the embedded field processing devices at the edge via the edge computing gateway, refining the early warning model, and improving the accuracy of the early warning.

[0081] The data undergoes two compression processes during acquisition and upload, significantly alleviating the narrow bandwidth issue in communication channels with the back-end cloud computing center. Simultaneously, migrating early warnings from the back-end server to embedded devices near the data generation location greatly improves the real-time performance of the warnings. The measured edge computing algorithm is programmed in C language suitable for embedded systems. After passing indoor debugging and testing, it is deployed within the front-end embedded system. Before operation, the front-end device stores its hardware and software self-test information in local memory and sends it to the reservoir management center or the cloud server of the supervising / related unit when the communication bandwidth meets requirements or the link is idle. The reservoir management center is located between the edge and the supervising / related unit's cloud computing center. It has a public broadband network connection with the Nanjing cloud computing center and a LoRa or wired connection with the edge. Its role is to maintain the edge computing network, handle data transfer, and trace the source of hazards. Only low-bandwidth BeiDou short message communication is needed between the monitoring site and the back-end cloud computing center to meet the requirements for two-way interaction, adapting to harsh conditions such as remote mountainous areas without public broadband, and also applicable in environments where public broadband is interfered with or damaged.

[0082] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A safety early warning method for dam projects in cold regions, characterized in that, Includes the following steps: Step 1: Collect measured data on temperature, ice thickness, water level, and deformation in the dam area to form multi-sample time series data; conduct correlation and distribution tests on the sample measurements; when the number of samples is insufficient to meet the test requirements, use the TimeGAN generative adversarial network model, which takes the influencing factors of the measured values ​​as input and the limited measured samples as the true values, to generate and supplement the samples until the sample number requirement is met. Step 2: Select measured ice thickness data and compare it with the ice thickness selected for the design load. Ice thicknesses greater than the design ice thickness are the ice thickness extreme value sequence. When the design does not consider ice load, all measured ice thicknesses greater than 0 are taken as extreme values, thus obtaining the ice thickness extreme value sequence. Calculate the average water level based on the measured data, and use the design water level as the high water level threshold. Water levels greater than the selected design water level are considered high water levels, thus obtaining the high water level extreme value sequence. When there is no continuous measured ice thickness data, ice thickness data is calculated using empirical formulas based on air temperature data. Step 3: Based on the two sets of extreme value sequences obtained in Step 2, namely the ice thickness extreme value sequence and the high water level extreme value sequence, use the corresponding probability density estimation method to obtain their respective probability density functions according to the test results in Step 1, and perform goodness-of-fit test until the test is passed. Step 4: Obtaining the joint probability function of ice thickness and water level; The joint probability distribution of ice thickness and water level acting on the dam is determined by the bivariate Copula function; The form of the joint probability distribution Copula function is determined based on the correlation and normality tests between variables. Under severe conditions, water level and ice thickness are positively correlated, so the Gumbel-Hougarard Copula function is selected, and the function coefficients are derived from the Kendall rank correlation coefficient. Step 5: Determine the critical probability using the reciprocal of the flood return period in the dam project design. The flood return period is in years. Based on the joint probability density function determined in Step 4, calculate whether the joint probability density function of maximum ice thickness and high water level during the project's operational life is greater than the critical probability. If it is not greater than the critical probability, continue collecting data and return to Step 1; otherwise, proceed to the next step. Step 6: When structural safety risks exist, it is necessary to check the completeness of the structural strength, seepage, and deformation response monitoring items and their measuring points. When the monitoring items and measuring points are complete, the TimeGAN generative adversarial network model is first used in combination with trend change, over-prediction confidence method or numerical calculation critical value method for single measuring point early warning. The output of the TimeGAN model is the measured structural response information of the dam at the measuring point, and the input is the corresponding load information, including wind direction, wind force, ice thickness, water pressure, temperature, time, and snow load. Then, the MAD-GAN generative adversarial network is used to conduct early warning for multiple monitoring items and multiple measuring points of the dam section and the entire dam body, as well as for local and overall dam early warning. The input of MAD-GAN is the measured response value of multiple measuring points, and the output is the judgment of the structural safety of the dam section / dam as a whole. If the items for monitoring structural response are incomplete, proceed to the next step; Step 7: Select the appropriate water-thermal-mechanical coupling model that considers frost heave based on the dam material and structural type. This model considers the effects of ice expansion and phase change on the stress and deformation of the dam and its foundation. The hydrothermal coupling model includes: an unsteady-state heat conduction equation accompanying the phase change, using the sensible heat capacity method to describe the phase change; a water migration control equation using the mass-conserved fluid continuity equation combined with the saturated-unsaturated Forchheimer's permeability law with an initiation pressure gradient, determined by the water diffusivity and ice content based on the temperature gradient; material strain considering temperature, humidity, stress-strain, and creep; a fractional constitutive model; a three-shear strength criterion for material yielding, and a strain energy release rate criterion for crack initiation; combining these with the overall equilibrium method yields the hydrothermal coupling model; and load conditions are defined by... Based on the measured data of the sensing system, including air and water temperature, ice thickness, wind direction and speed, snow thickness and its distribution, the numerical simulation calculations input the surface ice pressure, lower water pressure, and sediment pressure of the upstream / downstream reservoir water, while also considering the temperature stress of the dam body itself, the phase change effect of water and ice, and the frost heave force and volume force. Local and overall safety early warnings for the dam are conducted according to the material strength criterion, seepage failure criterion, and overall yield stability criterion. At the same time, the calculated critical state deformation value and seepage pressure value are deployed to the edge end as early warning indicators, and the samples obtained in the calculation process are used for sample enhancement for probability statistical testing at the edge end.

2. The safety early warning method for cold-region dam projects according to claim 1, characterized in that, In step 2, the ice thickness is calculated using the following formula: In the formula, The ice is thick; Living on frozen ice; These are empirical values ​​under different conditions; for lake ice caps without snow cover and with wind blowing. cm / ℃·d, lake ice cap without snow cover cm / ℃·d, when the lake ice cap is covered with snow cm / ℃·d; For dams in areas with frequent temperature changes, the following formula is used: In the formula It is a variable related to temperature.

3. The safety early warning method for cold-region dam projects according to claim 2, characterized in that, When there is no actual measured local temperature data, the temperature is obtained by interpolation from the meteorological department's temperature data.

4. The safety early warning method for cold-region dam projects according to claim 1, characterized in that, The generative networks in the TimeGAN and MAD-GAN models employ extended recurrent neural networks (adRNNs) with attention mechanisms.

5. The safety early warning method for cold-region dam projects according to claim 1, characterized in that, Steps 1 to 4 are completed on-site via edge computing, while the generative network model training is completed in a cloud computing center.

6. The safety early warning method for cold-region dam projects according to claim 1, characterized in that, Steps 5 to 7 are completed through a cloud computing center.

7. A smart monitoring system for safety of dam and gate projects in cold regions, characterized in that: The method for safety early warning of cold-region dam projects as described in any one of claims 1 to 6 comprises: sensors, field computing devices, edge computing gateways, and a back-end server. Sensors collect data; multiple field computing devices form a field edge computing network, each connected to a sensor and storing relevant data on-site; the field edge computing network performs data error identification, processing, and analysis, compares and issues early warnings based on the latest early warning indicators issued by the cloud computing center, and uploads the warning results to the cloud computing center or relevant authorized nodes; the edge computing gateway collects data from each edge-end field computing device, processes it, and uploads it to the cloud computing center; the cloud computing center processes the collected information from multiple edge gateways, employs multi-field coupled numerical analysis finite element method, model correction and data assimilation, and parameter inversion, and calculates the deformation, seepage, and stress-strain at the structural failure critical point as single-point, local, or overall early warning indicators, serving as the basis for data rationality analysis of the edge portion.

8. The intelligent safety monitoring system for cold-region dam projects according to claim 7, characterized in that, The sensors include: several air temperature and wind speed sensors, rain and snow volume sensors, water level sensors, water temperature sensors, ice thickness sensors, strain sensors, crack measuring sensors, inclinometers, dam body temperature sensors, and seepage pressure sensors installed at the site of the water-retaining structure; the air temperature and wind speed sensors collect wind speed, direction, and temperature data at the site of the water-retaining structure and store the data in an embedded field computing and storage device via wireless or wired networking; the rain and snow volume sensors collect rain and snow load data and store it in an embedded field computing and storage device via wireless or wired networking; the water level and water temperature sensors obtain the vertical water level and temperature distribution of the ice layer and water body; the ice thickness gauge measures the real-time thickness of the ice; the strain sensor, crack measuring sensor, and inclinometer monitor deformation and seepage at the site of the water-retaining structure and measure the tilt angle and displacement of the dam structure in the horizontal X and Y directions, and the collected data is stored in an embedded field computing and storage device via Zigbee or wired networking; the dam body temperature sensor obtains the internal temperature distribution of the dam, and the seepage pressure sensor obtains the internal seepage pressure when it has not yet frozen.

Images