Water conservancy project operation monitoring management method and system based on digital twinning

Abstract

This invention discloses a method and system for monitoring and managing the operation and maintenance of water conservancy projects based on digital twins. The method includes: collecting a real-time monitoring parameter array through a sensor array deployed at monitoring locations; constructing basic parameters within a real-time twin model based on the digital twin; calculating a difference degree array; performing model update requirement analysis to obtain update timeliness parameters; obtaining the expected time for anomaly analysis by an anomaly analysis agent; optimizing anomaly analysis by combining the difference degree array and update timeliness parameters; optimizing anomaly analysis schemes based on the optimized anomaly analysis scheme set and historical monitoring records; selecting the optimal anomaly analysis scheme; calling the corresponding anomaly analysis agent to perform anomaly analysis; obtaining an anomaly analysis result set; and labeling the real-time twin model as the monitoring and management result. This invention solves the technical problems of untimely analysis, inaccurate anomaly identification, and resource waste in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering technology, specifically to a method and system for water conservancy engineering operation and maintenance monitoring and management based on digital twins. Background Technology

[0002] In the operation and maintenance management of water conservancy projects, various monitoring parameters are key bases for assessing the safety status of the project. Currently, the operation and maintenance monitoring of water conservancy projects usually adopts a pattern of periodic data collection and analysis: that is, sensors are deployed at various monitoring locations, and monitoring data is collected once or multiple times at fixed time intervals. Then, management personnel or analysis systems centrally process the data to determine whether any anomalies exist.

[0003] However, this periodic monitoring model has the following shortcomings: Firstly, the data collection cycle is fixed. If the interval between two monitoring sessions is too long, abnormal changes may occur during the monitoring gap and fail to be detected in time, leading to missed anomalies and untimely analysis. Secondly, due to the numerous monitoring locations and complex parameter types in water conservancy projects, periodic data collection often processes all monitoring locations uniformly, making it difficult to conduct refined anomaly analysis based on the dynamic differences between different locations. This can easily result in some rapidly changing monitoring points not receiving sufficient attention, while long-term stable monitoring points consume the same analytical resources, leading to analytical omissions or resource waste. Therefore, how to dynamically adjust the analysis strategy based on the actual changes in monitoring data in the operation and maintenance monitoring management of water conservancy projects to achieve timely and accurate anomaly identification is an urgent technical problem to be solved. Summary of the Invention

[0004] This invention provides a digital twin-based method for the operation and maintenance monitoring and management of water conservancy projects, addressing the technical problems of untimely analysis, inaccurate anomaly identification, and resource waste in existing technologies. In view of the above problems, this invention provides a digital twin-based method and system for the operation and maintenance monitoring and management of water conservancy projects.

[0005] In a first aspect, the present invention provides a water conservancy project operation and maintenance monitoring and management method based on digital twins. The method includes: collecting a real-time monitoring parameter array by deploying a sensor array at multiple monitoring locations within the water conservancy project, and constructing basic parameters within a real-time twin model based on digital twins. Calculate the difference array between the real-time twin model and the historical twin model from the previous monitoring, perform model update requirement analysis, and obtain update timeliness parameters; Multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis are obtained. Combined with the difference array and update timeliness parameters, anomaly analysis optimization is performed for multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality. Based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, the optimal anomaly analysis scheme is selected, the corresponding anomaly analysis agent is invoked to perform anomaly analysis, an anomaly analysis result set is obtained, and the real-time twin model is labeled as the monitoring management result.

[0006] Secondly, the present invention also provides a water conservancy project operation and maintenance monitoring management system based on digital twins, the system comprising: The data acquisition model construction module is used to collect real-time monitoring parameter arrays through sensor arrays deployed at multiple monitoring locations within the water conservancy project, and to construct the basic parameters within the real-time twin model based on digital twins. The model difference analysis module is used to calculate the difference array between the real-time twin model and the historical twin model monitored last time, perform model update requirement analysis, and obtain update timeliness parameters; An anomaly analysis optimization module is used to obtain multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis. Combining the difference array and update timeliness parameters, it optimizes the anomaly analysis for multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality. The anomaly analysis execution module is used to select the optimal anomaly analysis scheme based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, call the corresponding anomaly analysis agent to perform anomaly analysis, obtain an anomaly analysis result set, and label the real-time twin model as the monitoring management result.

[0007] One or more technical solutions provided in this invention have at least the following technical effects or advantages: First, this invention obtains update timeliness parameters by calculating the difference array between the real-time twin model and the historical twin model, thereby dynamically judging the urgency of model updates based on the actual degree of change in the monitored data. Compared with the fixed-period monitoring mode in the prior art, this invention avoids resource waste caused by over-analysis when data changes slowly, and also prevents abnormal missed reports caused by untimely analysis when data changes drastically, achieving adaptive adjustment of monitoring timing.

[0008] Second, this invention obtains the expected time of multiple anomaly analysis agents, combines the difference array and update timeliness parameters, and optimizes the anomaly analysis scheme using the Pareto optimality principle. This enables the rational allocation of analysis resources among different monitoring locations, prioritizing critical locations with high differences, while ensuring that the total analysis time does not exceed the update timeliness parameters, thus solving the problem of unreasonable resource allocation caused by uniform processing in existing technologies.

[0009] Third, this invention further incorporates anomaly information from historical monitoring records to select the optimal anomaly analysis scheme from the Pareto optimal scheme set. This optimal selection strategy based on historical experience further improves the hit rate and timeliness of anomaly identification, enabling limited analytical resources to be concentrated on the locations most likely to experience anomalies, thereby improving the overall accuracy and efficiency of water conservancy project operation and maintenance monitoring.

[0010] In summary, this invention solves the technical problems of untimely analysis, inaccurate anomaly identification, and resource waste in the prior art by using difference-driven dynamic update scheduling, Pareto optimal resource allocation, and historical experience-guided scheme selection. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the water conservancy project operation and maintenance monitoring management method based on digital twins provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the water conservancy project operation and maintenance monitoring management system based on digital twin provided in the embodiments of this application; The components represented by each number in the attached diagram are described as follows: Data acquisition model construction module 11, model difference analysis module 12, anomaly analysis optimization module 13, and anomaly analysis execution module 14. Detailed Implementation

[0013] This application provides a method and system for monitoring and managing the operation and maintenance of water conservancy projects based on digital twins, which addresses the technical problems of untimely analysis, inaccurate anomaly identification, and resource waste in existing technologies.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0016] Example 1 like Figure 1 As shown, this application provides a method for monitoring and maintaining water conservancy projects based on digital twins, the method comprising: S100: Real-time monitoring parameter arrays are collected through sensor arrays deployed at multiple monitoring locations within the water conservancy project. Based on digital twins, basic parameters within the real-time twin model are constructed. In the operation and maintenance monitoring of water conservancy projects, the physical state changes at different locations vary, and the types of monitoring parameters are diverse. Traditional monitoring methods typically collect data from each monitoring point at fixed intervals, and then record it manually or semi-automatically in tables or databases, lacking a unified description of the relationship between monitoring data and the geometric space of the project.

[0017] Step S100 in the method provided in this application embodiment includes: Real-time monitoring parameter arrays are collected by sensor arrays deployed at multiple monitoring locations within the water conservancy project. Based on the location coordinates and monitoring parameter types of multiple monitoring locations, a digital twin model of the water conservancy project is constructed. The real-time monitoring parameter array is recorded into the twin model to construct the basic parameters within the real-time twin model.

[0018] The specific implementation method is as follows: First, sensor arrays are pre-deployed at key monitoring locations in the water conservancy project, such as the dam foundation, different elevations of the dam body, the dam shoulders on both banks, the spillway gate piers, and the base slab. These arrays include, but are not limited to, piezometers, crack gauges, strain gauges, thermometers, water level gauges, and GPS displacement monitoring stations. Each sensor collects data at a preset sampling frequency, for example, every 5 minutes for seepage pressure and deformation, and every hour for water level and temperature. The data collected by each sensor is recorded as a data tuple with the structure (sensor ID, monitoring location ID, parameter type, timestamp, value). All sensor data tuples are collected in real time via wired or wireless network, aligned according to the timestamp, and formed into a two-dimensional table structure, i.e., the real-time monitoring parameter array. The rows of this array correspond to a single acquisition record from each sensor, and the columns are, in order, sensor ID, location ID, parameter type, timestamp, and value. For example, a row record might be (S001, P-12, seepage pressure, 2025-06-15 14:30:00, 25.6).

[0019] Furthermore, a digital twin model of the water conservancy project is constructed based on digital twin technology. The basic principle of digital twins is to use data such as physical models, sensor updates, and operational history to map physical entities in virtual space, thereby reflecting their entire life cycle. In this example, the digital twin modeling tool AutoCADPlant3D can be used. The CAD drawings, BIM model, and GIS geographic information of the water conservancy project are imported into the modeling tool, and a three-dimensional digital base is formed through coordinate system registration and geometric fusion.

[0020] Furthermore, monitoring nodes are defined in the twin model, each corresponding to a physical monitoring location, recording its spatial coordinates (X, Y, Z) and the engineering part it belongs to. Simultaneously, based on the monitoring parameter type, corresponding attribute fields are configured for each node; for example, a "seepage pressure value" attribute is defined for the seepage pressure node, and "horizontal displacement" and "vertical displacement" attributes are defined for the displacement node. Through this method, the location coordinates and monitoring parameter types of multiple monitoring locations are structured and organized according to the data standards of digital twins, forming the basic twin model framework for water conservancy projects. This framework not only includes geometric shapes but also embeds the spatial distribution of monitoring points and parameter type definitions, achieving a precise mapping from physical entities to digital space.

[0021] Finally, the latest unprocessed records are periodically read from the real-time monitoring parameter array. The corresponding monitoring node is retrieved in the twin model based on the sensor ID or monitoring location ID, and the value is written to the attribute field corresponding to that node, along with a timestamp. For example, for the record (S001, P-12, pressure, 2025-06-15 14:30:00, 25.6), the twin node Node_12 corresponding to location P-12 is found, its "pressure value" field is updated to 25.6, and the "update time" field is set to 2025-06-15 14:30:00. Each monitoring node in the twin model can also maintain a historical value queue, such as a circular array of length 10. New values ​​are pushed in each update, and the oldest values ​​are automatically discarded, thus retaining the most recently collected data. Through this mapping, the structured records in the real-time monitoring parameter array are transformed into basic parameters on the twin model nodes, namely the current value + historical sequence, forming a complete digital mirror of "geometry + attribute + time".

[0022] In summary, this step achieved the following technical effects: by collecting real-time monitoring parameter arrays from multiple locations and of multiple types using a sensor array, and constructing a real-time twin model containing basic parameters based on digital twin technology, spatial and structured organization of water conservancy project monitoring data was realized. Compared to traditional decentralized recording methods, this invention deeply integrates monitoring data with the twin model, providing managers with an intuitive and unified view of the project status, and reducing the difficulty of data understanding and analysis.

[0023] S200: Calculate the difference array between the real-time twin model and the historical twin model monitored last time, perform model update requirement analysis, and obtain update timeliness parameters; In the operation and maintenance monitoring of water conservancy projects, the rate of change in the project status is not constant. During flood season, construction period, or geologically active periods, monitoring parameters may change rapidly, and a fixed period that is too long will lead to a lag in anomaly detection. Conversely, during stable periods, parameter changes are slow, and a fixed period that is too short will waste computational resources. Therefore, a mechanism is needed that can dynamically adjust the analysis frequency based on the degree of change in actual monitoring data. This step quantifies the magnitude of parameter changes at each monitoring location by calculating the difference array between the real-time twin model and the historical twin model, and analyzes the urgency of model updates accordingly, generating update timeliness parameters to provide a basis for the dynamic scheduling of subsequent anomaly analysis resources.

[0024] Step S200 in the method provided in this application embodiment includes: Retrieve the historical twin model from the last monitoring, wherein the historical twin model includes an array of historical monitoring parameters; The difference magnitudes between the historical monitoring parameter array and the real-time monitoring parameter array within the historical twin model and the real-time twin model are calculated to obtain the difference degree array; Based on the difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters. The specific implementation method is as follows: First, retrieve the historical twin model from the database or twin model storage, such as the one saved at the last monitoring time, the last cycle, or the last time the analysis was triggered. This historical twin model has the same geometric structure and monitoring node organization as the current real-time twin model, and its core content is the historical monitoring parameter array, that is, the parameter values ​​recorded at the last acquisition time on each monitoring node.

[0025] Furthermore, the difference between the historical twin model and the real-time twin model is calculated. Specifically, for each type of monitoring parameter at each monitoring node, the difference between the real-time value and the historical value is calculated. The difference can be calculated using the relative difference formula: Difference = |Real-time value - Historical value| / Historical baseline fluctuation range. Based on the 3σ principle in statistics, the historical baseline fluctuation range can be taken as three times the standard deviation of the monitoring parameter under historical normal conditions, which can effectively cover the parameter fluctuation range under normal operating conditions. For example, if the historical value of a piezometer is 25.6 kPa and the real-time value is 26.8 kPa, the historical baseline fluctuation range, calculated as three times the standard deviation of the monitoring parameter under historical normal conditions, is 5 kPa. Then the difference is |26.8 - 25.6| / 5 = 0.24. For discrete parameters, such as the opening and closing state of a gate, the difference is defined as 0 (same) or 1 (different). The differences of all monitoring nodes and all parameter types are arranged in node order to form a two-dimensional array, i.e., the difference array. Each element in this array represents the degree of change of the corresponding parameter at the corresponding monitoring location.

[0026] Finally, based on the difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters. These parameters guide the timeframe within which subsequent anomaly analysis must be completed. Specifically, a larger overall difference indicates more drastic changes in the project status, requiring faster analysis completion, and thus a smaller update timeliness parameter. The update timeliness parameter is calculated as follows: Update Timeliness Parameter = Average Time Period × (Historical Baseline Difference / Average Difference).

[0027] The detailed calculation method for updating the timeliness parameters is as follows: Based on the aforementioned difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters, including: Obtain the average time period for monitoring water conservancy projects; Calculate the mean of the difference array to obtain the average difference. Based on sensor monitoring data over a historical period, obtain the historical baseline difference. The ratio of the historical baseline difference to the average difference is calculated, and the average time period is adjusted to obtain the updated timeliness parameters.

[0028] First, obtain the average time period for monitoring the water conservancy project, and record it as the baseline period. This value can be set according to engineering specifications or historical experience, and this parameter serves as the baseline analysis interval.

[0029] Furthermore, the mean of the difference array is calculated to obtain the average difference. Assuming there are multiple monitoring nodes, each with several parameters, the total number of elements in the difference array equals the number of monitoring nodes multiplied by the number of each parameter type. The average difference equals the sum of all difference values ​​divided by the total number of difference values. For example, if the difference array contains 120 elements and the sum of all differences in the array is 18.0, then the average difference is 0.15.

[0030] Furthermore, based on historical sensor monitoring data, historical baseline variability is obtained. Historical baseline variability reflects the average fluctuation of monitored parameters under normal and stable conditions. It can be obtained by analyzing the variability between all two consecutive monitoring measurements over a period of one year or longer when the project is without anomalies, and taking the average or median of these variability values. For example, the historical baseline variability is statistically calculated to be 0.05.

[0031] Finally, the ratio of historical baseline variance to average variance is calculated, and the baseline period is adjusted accordingly to obtain the update lead time parameter. The calculation formula is: Update lead time parameter = Baseline period × (Historical baseline variance / Average variance). Note that when the average variance is very small, the ratio may be greater than 1. In this case, the update lead time parameter may be greater than the baseline period, indicating that the analysis interval can be extended; however, for safety, an upper limit is usually set, such as not exceeding twice the baseline period. When the average variance is large, the update lead time parameter becomes smaller, indicating that the analysis needs to be accelerated. For example, if the baseline period is 120 seconds, the historical baseline variance is 0.05, and the average variance is 0.15, then the update lead time parameter = 120 × (0.05 / 0.15) = 40 seconds. That is, the update lead time parameter is 40 seconds, which indicates that the analysis needs to be accelerated.

[0032] Through the above calculations, the update timeliness parameters can quantitatively reflect the rate of change in the project status, providing a clear time constraint for the scheduling of the anomaly analysis agent in subsequent steps. If the average difference is close to zero, it indicates that the status has hardly changed, and the analysis interval can be extended or even skipped to save computational resources. If the average difference is much greater than the historical baseline difference, it indicates that a significant change has occurred, and the analysis window must be shortened to ensure that anomalies can be detected in a timely manner.

[0033] It is important to note that when the system is running for the first time or when the historical twin model from the previous monitoring is unavailable, a preset initial time-sensitivity parameter is used as a substitute because the difference array cannot be calculated. The initial time-sensitivity parameter is preset according to the type of water conservancy project and monitoring specifications. For example, for large dams, the initial time-sensitivity parameter can be set to 60 seconds, corresponding to the lower limit of a conventional fixed monitoring cycle, ensuring a more conservative analysis frequency during the system's cold start phase, and switching to dynamic adjustment mode after accumulating sufficient historical data. For small sluice gates or dikes, the initial time-sensitivity parameter can be set to 120 seconds to accommodate the relatively slow changes in parameters for these types of projects. During system operation, after completing the first full monitoring cycle, subsequent updated time-sensitivity parameters can be calculated using the difference array method described above, and the initial time-sensitivity parameter automatically becomes invalid.

[0034] The following technical effects were achieved through this step: By calculating the difference array between the real-time twin model and the historical twin model, the degree of change of each part and parameter of the project can be quantified, enabling precise understanding of the spatial distribution and intensity of state fluctuations. Based on the ratio of the average value of the difference array to the historical baseline difference, the timeliness parameters are dynamically adjusted and updated, ensuring that the analysis frequency matches the actual rate of change in the project. This solves the technical problems of untimely analysis when changes are rapid and resource waste when changes are slow in fixed-cycle monitoring.

[0035] S300: Obtain multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis, combine the difference array and update timeliness parameters, optimize the anomaly analysis for multiple monitoring locations, and obtain a set of optimized anomaly analysis schemes that meet Pareto optimality. After obtaining the dissimilarity array and updating the timeliness parameters, anomaly analysis needs to be performed on multiple monitoring locations. Each monitoring location is configured with a corresponding anomaly analysis agent, such as a machine learning-based classification model, to determine whether the monitoring parameters at that location are abnormal. In this step, the optimization problem includes two optimization objectives: the first objective is to maximize the total dissimilarity of the selected monitoring locations, as a larger total dissimilarity indicates a higher potential anomaly risk; the second objective is to minimize the total resource consumption, which is measured by the total expected time or total memory usage of the selected agent. Pareto optimality is the search for a non-dominated solution between the above two objectives: an anomaly analysis scheme is called a Pareto optimal solution if and only if there is no other feasible scheme that results in a smaller total resource consumption without decreasing the total dissimilarity, or a larger total dissimilarity without increasing the total resource consumption. That is, any two schemes in the Pareto optimal solution set are non-dominated: if the total dissimilarity of scheme A is greater than that of scheme B, then the total resource consumption of scheme A must also be greater than that of scheme B. This non-dominated relationship preserves the trade-off choice space between different resource inputs and risk coverage. Multiple solutions that meet the constraints are generated through the above random search, and solutions that do not dominate each other are selected according to the above dominance relationship, thus forming an optimization anomaly analysis solution set.

[0036] Step S300 in the method provided in this application embodiment includes: Obtain multiple anomaly analysis agents corresponding to multiple pre-configured monitoring locations, and test the anomaly analysis time for each agent to obtain multiple expected times; Based on the difference array, update time parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality.

[0037] The specific implementation method is as follows: First, an anomaly analysis agent is pre-configured for each monitoring location. This agent can be a pre-trained classification model for that location, taking as input historical monitoring parameter sequences or real-time values ​​and outputting as anomaly probability or classification label (normal or anomalous). Then, each agent is tested: under standard operating conditions, typical data is input, and the average time from receiving the input to outputting the result is measured. This average time is taken as the expected time for the anomaly analysis agent, denoted as the i-th expected time, in seconds or milliseconds. For example, the expected time for the seepage pressure anomaly analysis agent at location A is 120 milliseconds, the expected time for the displacement agent at location B is 150 milliseconds, and the expected time for the temperature agent at location C is 80 milliseconds.

[0038] Furthermore, the difference array and update timeliness parameters are obtained from the above step S200. The difference degree corresponding to each monitoring location in the difference array can be a combination of the difference degrees of multiple parameters at that location. For example, the weighted sum is recorded as the i-th difference degree, which represents the degree of drastic change in the state at that location.

[0039] Finally, anomaly analysis and optimization are performed at multiple monitoring locations. The optimization objective is to select a set of monitoring locations (i.e., a set of agents) that maximizes the total dissimilarity of the selected locations, provided that the total analysis time does not exceed the update timeliness parameter. Furthermore, memory constraints must also be met. This embodiment uses a random search method to generate multiple Pareto optimal solutions.

[0040] The configuration steps for the aforementioned multiple anomaly analysis agents include: Based on monitoring data from multiple monitoring locations over a historical period, multiple sets of sample monitoring parameters are collected, and each sample monitoring parameter is labeled as abnormal, resulting in multiple sets of sample anomaly analysis results. Based on machine learning, multiple anomaly analysis agents are constructed. The input data for each anomaly analysis agent is the monitoring parameters of the corresponding monitoring location, and the output data is the anomaly analysis result. Multiple sets of sample monitoring parameters and multiple sets of sample anomaly analysis results are used as training data and supervision labels to conduct supervised training and testing on multiple anomaly analysis agents. Configuration is completed after the test is passed.

[0041] The specific implementation method is as follows: First, for each monitoring location, such as piezometer P-12, historical monitoring parameter sequences for that location over a past period are collected, with the sampling frequency consistent with real-time acquisition. The continuous time series is then divided into multiple samples using a sliding window, for example, every 24 consecutive points. Each sample contains a set of monitoring parameter values. Next, based on known anomaly records, each sample is labeled: if an anomaly occurred within the corresponding time period, it is labeled "abnormal"; otherwise, it is labeled "normal." This yields the set of monitoring parameters for that location and the corresponding set of anomaly analysis results.

[0042] Furthermore, an anomaly analysis agent is constructed based on machine learning methods. Taking seepage pressure anomaly analysis as an example, logistic regression is chosen as the classification model. The input features of the classification model can be the original monitoring value sequence, statistical features (mean, variance, slope), and frequency domain features (Fourier transform dominant frequency). The output is binary classification: 0 represents normal, and 1 represents abnormal. Supervised training is performed using collected and labeled training data: the model initializes a set of weight parameters and bias parameters. For the labeled training data, the model calculates an intermediate score by linear weighted summation, and then maps the score to the zero-to-one interval using the Sigmoid nonlinear activation function, outputting a predicted value representing the probability of anomaly occurrence. The cross-entropy loss function is used to quantify the degree of inconsistency between the model's predicted probability and the true label. The larger the loss value, the more the prediction deviates from the reality. Then, the gradient of the loss function with respect to each parameter of the model is calculated using the chain rule. This gradient indicates the steepest direction for adjusting the parameters to reduce the loss. Based on the calculated gradient direction and the step size controlled by the learning rate, the model's weights and bias parameters are fine-tuned. This process is repeated across all training samples until the loss function, which measures the model's prediction error, stabilizes and no longer decreases significantly. After training stops, a set of independent test samples not used in training is used to perform a final evaluation of the model. The core performance indicators are accuracy and recall. Only when both of these indicators meet preset performance thresholds, such as 95%, is the model considered successfully trained and ready to be configured as an anomaly analysis agent for practical application.

[0043] Finally, for different monitoring locations, due to the different parameter types and variation patterns, different agent structures can be constructed. For example, an LSTM time series model can be used for seepage pressure, while a regression model can be used for displacement. After configuration, each agent runs independently.

[0044] Based on the difference array, update timeliness parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that satisfy Pareto optimality, including: Obtain the memory constraints for making calls to the intelligent agent, and use the total anomaly analysis time being less than the update timeliness parameter as the timeliness constraint; If the computing power of a single call to an agent satisfies the memory constraint and the sum of the expected times for synchronous anomaly analysis by the called agents satisfies the timeliness constraint, multiple anomaly analysis agents are randomly selected to obtain a first anomaly analysis scheme. The sum of the differences of all real-time monitoring parameters used for anomaly analysis within the first anomaly analysis scheme is obtained to obtain the total difference of the first analysis. Continue to randomly select an anomaly analysis agent until the convergence optimization count is reached; During the optimization process, the anomaly analysis scheme with the largest total difference in analysis among the anomaly analysis schemes of each anomaly analysis agent is recorded as the set of optimized anomaly analysis schemes that satisfy Pareto optimality.

[0045] The specific implementation method is as follows: First, obtain the constraints. If there are N monitoring locations, and the anomaly analysis agent corresponding to each location i is A_i, its expected time is t_i (in milliseconds), and its difference is d_i, obtained from the difference array. Update the timeliness parameter T_update (in milliseconds), which is the timeliness constraint. It is important to note that the timeliness constraint of the update timeliness parameter T_update generated in step S200 above means the length of the time window from the completion of the current real-time monitoring parameter collection to the start time of the next Siamese model update. The anomaly analysis task scheduled in step S300, as a sub-task within this time window, should have a total anomaly analysis time less than the timeliness constraint T_update to ensure that the anomaly analysis results can be output and labeled to the real-time Siamese model before the next model update. If the total anomaly analysis time exceeds T_update, the anomaly analysis results will not be fed back to the Siamese model update process in a timely manner, thus losing timeliness. Therefore, the logical basis for the time constraint "total anomaly analysis time is less than update time parameters" in this step is that the anomaly analysis task must be completed within the time window defined by T_update to ensure that the anomaly analysis results can take effect before the next model update.

[0046] Simultaneously, a memory constraint is established: during a single agent invocation, the total memory occupied by all concurrently running agents cannot exceed the available memory M_max. Since the models of each agent have different sizes, let the memory usage of agent A_i be m_i; then the sum of the memory usage of the selected agents must be ≤ M_max. These two constraints ensure that the solution is feasible in terms of both time and resources.

[0047] Further, an initial scheme is randomly generated. A Monte Carlo random search method is used: under the premise of satisfying the above two constraints, a set of agents is randomly selected, i.e., a subset of monitoring locations is randomly selected. Specifically, each agent is traversed, and added to the candidate set with a probability of 0.5. Then, it is checked whether the total expected time and total memory exceed the preset constraint limit; if they do, the most recently added agent is removed or a new random selection is performed. This process is repeated until a feasible scheme is generated, called the first anomaly analysis scheme. For example, a randomly generated scheme includes agents A_2, A_5, and A_7. Assuming the three agents call in parallel, the total expected time is the maximum value of the expected time of each agent, i.e., max(120,150,80)=150ms; the total memory is the sum of the memory usage of each agent, i.e., 200+150+100=450MB. Assuming the time constraint T_update=400ms and the memory constraint M_max=500MB, this scheme satisfies both the time and memory constraints and is a feasible scheme.

[0048] Furthermore, the total dissimilarity of the scheme is calculated. For this scheme, the sum of the dissimilarity values ​​corresponding to all selected agents is calculated, which is the total dissimilarity. For example, if d_2=0.24, d_5=0.33, and d_7=0.12, then the total dissimilarity = 0.69. This value represents the magnitude of the anomaly risk that the scheme can analyze; the larger the dissimilarity, the higher the probability of an anomaly occurring at that location.

[0049] Furthermore, repeat the above random generation steps, generating a new feasible solution each time and calculating its corresponding total dissimilarity. Set a convergence condition, such as performing one thousand random search iterations. During the iteration process, continuously track and record the solution with the largest total dissimilarity among all currently generated feasible solutions, and use it as the Pareto optimal solution.

[0050] Finally, an optimized solution set is output. After iteration, all anomaly analysis solutions found during the optimization process that belong to the Pareto optimal solution are recorded, forming an optimization anomaly analysis solution set. Each solution in this set is a non-dominated solution that satisfies the constraints. Subsequent steps will select the optimal solution from this set based on historical anomaly monitoring. For example, the solution set may contain three solutions: Solution 1 (total anomaly score 0.95, time 380ms), Solution 2 (total anomaly score 0.88, time 310ms), and Solution 3 (total anomaly score 0.82, time 260ms). Different solutions correspond to different resource inputs and risk coverage trade-offs.

[0051] The following technical effects were achieved through this step: First, the expected time of each anomaly analysis agent is obtained. By combining the difference array and the update time parameter, the allocation of anomaly analysis resources is transformed into a multi-objective combinatorial optimization problem, so that limited computing resources can be prioritized for monitoring locations with high difference and high anomaly risk.

[0052] Second, a random search method is used to generate the Pareto optimal solution set, which avoids the combinatorial explosion caused by exhaustive search and can obtain multiple non-dominated solutions with acceptable computational overhead, providing flexibility for subsequent selection.

[0053] Third, by using both memory and time constraints, the generated solution is guaranteed to be executable in practice, avoiding crashes or task timeouts caused by resource overruns.

[0054] Fourth, the concept of Pareto optimality ensures that no single solution in the set of solutions is superior to another in all metrics, thus preserving the choice space for different resource and benefit trade-offs and adapting to operation and maintenance strategies under different working conditions.

[0055] S400: Based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, the optimal anomaly analysis scheme is selected, the corresponding anomaly analysis agent is invoked to perform anomaly analysis, an anomaly analysis result set is obtained, and the real-time twin model is labeled as the monitoring management result.

[0056] Step S300 yielded several Pareto-optimal anomaly analysis schemes. Each scheme corresponds to a set of monitoring locations to be analyzed and has different total variances under time and memory constraints. However, in actual execution, only one scheme can be selected for anomaly analysis. Simply selecting based on variance may not be optimal, as the historical anomaly frequency varies across different monitoring locations. Some locations may currently have low variances but have historically been prone to anomalies and should be given higher attention; conversely, some locations may have high variances but may only exhibit transient fluctuations and have never historically shown anomalies. Therefore, it is necessary to combine historical monitoring records to select the optimal scheme from the set of optimized schemes that is most likely to discover the true anomalies, thereby improving the hit rate of anomaly identification and the effectiveness of operation and maintenance management.

[0057] Step S400 in the method provided in this application embodiment includes: Acquire historical monitoring records from multiple monitoring locations within a historical period and process multiple monitoring anomalies at these locations; Based on multiple monitoring anomalies, the optimal anomaly analysis scheme is selected from the optimized anomaly analysis scheme set. Among them, the sum of monitoring anomalies of the anomaly analysis agents called in the optimal anomaly analysis scheme is the largest. The anomaly analysis agent within the optimal anomaly analysis scheme is invoked, the corresponding real-time monitoring parameters are input, and the anomaly analysis result set is output. The real-time twin model is then labeled as the monitoring and management result.

[0058] The specific implementation method is as follows: First, retrieve historical monitoring records for each monitoring location over a past period from a database or historical record storage. These records include the time and value of each acquisition, as well as a label indicating whether an anomaly occurred, which can be obtained through post-event manual verification or historical alarm logs. For each monitoring location, process its historical monitoring records to obtain a quantified monitoring anomaly score. The monitoring anomaly score reflects the frequency or severity of anomalies occurring at that location over a past period, and can be defined as, for example, the cumulative number of anomalies, the duration of anomalies, or the magnitude of anomalies.

[0059] A simple implementation is to directly count the number of times each monitoring location experienced anomalies in the past year. For example, if a piezometer location experienced 5 anomalies in the past year, a displacement gauge location experienced 2 anomalies, and a thermometer location experienced 0 anomalies, then the anomaly rates would be 5, 2, and 0, respectively.

[0060] Furthermore, from the optimized set of anomaly analysis schemes, the sum of anomaly degrees for all monitoring locations included in each scheme is calculated. The scheme with the largest sum of anomaly degrees is selected as the optimal anomaly analysis scheme. For example, the scheme set contains three schemes: Scheme A includes locations {1,3,5} with anomalies of 5, 0, and 2 respectively, totaling 7; Scheme B includes locations {2,4} with anomalies of 2 and 1 respectively, totaling 3; Scheme C includes locations {1,2,6} with anomalies of 5, 2, and 0 respectively, totaling 7, the same as Scheme A. In this case, any one of them can be selected. If multiple schemes have the same sum of anomaly degrees, the total difference degree can be compared, and the scheme with the larger total difference degree can be selected to ensure that the currently drastically changing points are also covered.

[0061] Finally, the anomaly analysis agent corresponding to each monitoring location in the optimal anomaly analysis scheme is invoked. The current real-time monitoring parameters or recent time-series data are input into the agent to obtain the anomaly analysis results for each location, such as a normal or anomaly label. All results are then aggregated to form an anomaly analysis result set. This result set is then annotated onto the real-time twin model: in the twin model's 3D view, monitoring nodes identified as anomaly are highlighted in red, displaying the specific values ​​of the anomaly parameters and alarm information; normal nodes are displayed in green or blue. Administrators can click on any node through the interactive interface to view detailed analysis results. This yields the final monitoring and management results, which can be used to trigger alarms, generate maintenance work orders, or record logs.

[0062] Acquire historical monitoring records from multiple monitoring locations within a historical time period, and process multiple monitoring anomalies at these locations, including: Acquire historical monitoring records from multiple monitoring locations within a historical time period; Extract the number of times anomalies occur at multiple monitoring locations to obtain multiple monitoring anomaly counts, and calculate multiple monitoring anomaly degrees.

[0063] The specific implementation method is as follows: From historical monitoring records, for each monitoring location, the total number of times it was marked as abnormal within a specified time window, such as the past year, the past three years, or since commissioning, is counted. The sources of these abnormal markings can be historical manual review records, past automatic alarm logs, or known accident records. These counts are used as the monitoring abnormality count for that location, and then this count is directly used as the monitoring abnormality degree, or normalized. Since the subsequent comparison is based on the sum of abnormality degrees, the original count can be used directly without additional calculation. For example, a dam has 20 monitoring locations, and the abnormality count list for each location is obtained as follows: [5, 2, 0, 8, 1, ...]. These abnormality degrees are used in step S400 to select the optimal solution.

[0064] The following technical effects were achieved through this step: By combining anomaly information from historical monitoring records, the combination of monitoring locations with the highest historical anomaly frequency is selected from the Pareto optimal solution set as the optimal solution. This prioritizes limited analytical resources for areas most likely to experience anomalies, improving the hit rate of anomaly identification and operational efficiency. The agents within the optimal solution are then invoked for actual anomaly analysis, avoiding the resource waste caused by blindly calling all agents while ensuring no critical locations are overlooked. The anomaly analysis results are annotated onto a real-time twin model, enabling a visual representation of anomaly states. Managers can intuitively view the safety status of various parts in the 3D model, facilitating rapid problem location and maintenance measures.

[0065] Example 2 like Figure 2 As shown, based on the same inventive concept as the water conservancy project operation and maintenance monitoring and management method based on digital twin provided in Embodiment 1, this embodiment of the invention also provides a water conservancy project operation and maintenance monitoring and management system based on digital twin. The system includes a data acquisition model construction module, a model difference analysis module, an anomaly analysis and optimization module, and an anomaly analysis execution module. The system includes: The data acquisition model construction module 11 is used to collect real-time monitoring parameter arrays through sensor arrays deployed at multiple monitoring locations within the water conservancy project, and to construct basic parameters within the real-time twin model based on digital twins. The model difference analysis module 12 is used to calculate the difference array between the real-time twin model and the historical twin model monitored last time, perform model update requirement analysis, and obtain update timeliness parameters; Anomaly analysis optimization module 13 is used to obtain multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis, and combine the difference array and update timeliness parameters to optimize anomaly analysis for multiple monitoring locations, thereby obtaining a set of optimized anomaly analysis schemes that meet Pareto optimality. The anomaly analysis execution module 14 is used to select the optimal anomaly analysis scheme based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, call the corresponding anomaly analysis agent to perform anomaly analysis, obtain an anomaly analysis result set, and label the real-time twin model as the monitoring management result.

[0066] In one embodiment, the data acquisition model construction module 11 is used to acquire a real-time monitoring parameter array through a sensor array deployed at multiple monitoring locations within the water conservancy project, and construct the basic parameters within the real-time twin model based on digital twins, including: Real-time monitoring parameter arrays are collected by sensor arrays deployed at multiple monitoring locations within the water conservancy project. Based on the location coordinates and monitoring parameter types of multiple monitoring locations, a digital twin model of the water conservancy project is constructed. The real-time monitoring parameter array is recorded into the twin model to construct the basic parameters within the real-time twin model.

[0067] In one embodiment, the model difference analysis module 12 is used to calculate the difference array between the real-time twin model and the historical twin model monitored previously, to perform model update requirement analysis, and to obtain update timeliness parameters, including: Retrieve the historical twin model from the last monitoring, wherein the historical twin model includes an array of historical monitoring parameters; The difference magnitudes between the historical monitoring parameter array and the real-time monitoring parameter array within the historical twin model and the real-time twin model are calculated to obtain the difference degree array; Based on the difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters.

[0068] Based on the difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters, including: Obtain the average time period for monitoring water conservancy projects; Calculate the mean of the difference array to obtain the average difference. Based on sensor monitoring data over a historical period, obtain the historical baseline difference. The ratio of the historical baseline difference to the average difference is calculated, and the average time period is adjusted to obtain the updated timeliness parameters.

[0069] In one embodiment, the anomaly analysis optimization module 13 is used to obtain multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis, and combine the difference array and update timeliness parameters to optimize the anomaly analysis for multiple monitoring locations, thereby obtaining a set of optimized anomaly analysis schemes that satisfy Pareto optimality, including: Obtain multiple anomaly analysis agents corresponding to multiple pre-configured monitoring locations, and test the anomaly analysis time for each agent to obtain multiple expected times; Based on the difference array, update time parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality.

[0070] The configuration steps for multiple anomaly analysis agents include: Based on monitoring data from multiple monitoring locations over a historical period, multiple sets of sample monitoring parameters are collected, and each sample monitoring parameter is labeled as abnormal, resulting in multiple sets of sample anomaly analysis results. Based on machine learning, multiple anomaly analysis agents are constructed. The input data for each anomaly analysis agent is the monitoring parameters of the corresponding monitoring location, and the output data is the anomaly analysis result. Multiple sets of sample monitoring parameters and multiple sets of sample anomaly analysis results are used as training data and supervision labels to conduct supervised training and testing on multiple anomaly analysis agents. Configuration is completed after the test is passed.

[0071] Based on the difference array, update timeliness parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that satisfy Pareto optimality, including: Obtain the memory constraints for making calls to the intelligent agent, and use the total anomaly analysis time being less than the update timeliness parameter as the timeliness constraint; If the computing power of a single call to an agent satisfies the memory constraint and the sum of the expected times for synchronous anomaly analysis by the called agents satisfies the timeliness constraint, multiple anomaly analysis agents are randomly selected to obtain a first anomaly analysis scheme. The sum of the differences of all real-time monitoring parameters used for anomaly analysis within the first anomaly analysis scheme is obtained to obtain the total difference of the first analysis. Continue to randomly select an anomaly analysis agent until the convergence optimization count is reached; During the optimization process, the anomaly analysis scheme with the largest total difference in analysis among the anomaly analysis schemes of each anomaly analysis agent is recorded as the set of optimized anomaly analysis schemes that satisfy Pareto optimality.

[0072] In one embodiment, the anomaly analysis execution module 14 is used to select the optimal anomaly analysis scheme based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, call the corresponding anomaly analysis agent to perform anomaly analysis, obtain an anomaly analysis result set, and label the real-time twin model as a monitoring management result, including: Acquire historical monitoring records from multiple monitoring locations within a historical period and process multiple monitoring anomalies at these locations; Based on multiple monitoring anomalies, the optimal anomaly analysis scheme is selected from the optimized anomaly analysis scheme set. Among them, the sum of monitoring anomalies of the anomaly analysis agents called in the optimal anomaly analysis scheme is the largest. The anomaly analysis agent within the optimal anomaly analysis scheme is invoked, the corresponding real-time monitoring parameters are input, and the anomaly analysis result set is output. The real-time twin model is then labeled as the monitoring and management result.

[0073] This involves acquiring historical monitoring records from multiple monitoring locations within a historical period and processing multiple monitoring anomalies at these locations, including: Acquire historical monitoring records from multiple monitoring locations within a historical time period; Extract the number of times anomalies occur at multiple monitoring locations to obtain multiple monitoring anomaly counts, and calculate multiple monitoring anomaly degrees.

[0074] It should be noted that the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0075] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0076] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for operation, maintenance, monitoring, and management of water conservancy projects based on digital twins, characterized in that: The method includes: By deploying sensor arrays at multiple monitoring locations within the water conservancy project, real-time monitoring parameter arrays are collected, and based on digital twins, basic parameters within the real-time twin model are constructed. Calculate the difference array between the real-time twin model and the historical twin model from the previous monitoring, perform model update requirement analysis, and obtain update timeliness parameters; Multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis are obtained. Combined with the difference array and update timeliness parameters, anomaly analysis optimization is performed for multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality. Based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, the optimal anomaly analysis scheme is selected, the corresponding anomaly analysis agent is invoked to perform anomaly analysis, an anomaly analysis result set is obtained, and the real-time twin model is labeled as the monitoring management result.

2. The water conservancy project operation and maintenance monitoring and management method based on digital twin as described in claim 1, characterized in that, By deploying sensor arrays at multiple monitoring locations within the water conservancy project, real-time monitoring parameter arrays are collected. Based on digital twins, the basic parameters within the real-time twin model are constructed, including: Real-time monitoring parameter arrays are collected by sensor arrays deployed at multiple monitoring locations within the water conservancy project. Based on the location coordinates and monitoring parameter types of multiple monitoring locations, a digital twin model of the water conservancy project is constructed. The real-time monitoring parameter array is recorded into the twin model to construct the basic parameters within the real-time twin model.

3. The water conservancy project operation and maintenance monitoring management method based on digital twin as described in claim 1, characterized in that, Calculate the difference array between the real-time twin model and the historical twin model from the previous monitoring, perform model update requirement analysis, and obtain update timeliness parameters, including: Retrieve the historical twin model from the last monitoring, wherein the historical twin model includes an array of historical monitoring parameters; The difference magnitudes between the historical monitoring parameter array and the real-time monitoring parameter array within the historical twin model and the real-time twin model are calculated to obtain the difference degree array; Based on the difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters.

4. The water conservancy project operation monitoring management method based on digital twinning according to claim 3, characterized in that, Based on the aforementioned difference array, a model update timeliness requirement analysis is performed to obtain update timeliness parameters, including: Obtain the average time period for monitoring water conservancy projects; Calculate the mean of the difference array to obtain the average difference. Based on sensor monitoring data over a historical period, obtain the historical baseline difference. The ratio of the historical baseline difference to the average difference is calculated, and the average time period is adjusted to obtain the updated timeliness parameters.

5. The water conservancy project operation monitoring management method based on digital twinning according to claim 1, characterized in that, Multiple expected times for anomaly analysis agents configured for multiple monitoring locations are obtained. Combining the difference array and update timeliness parameters, anomaly analysis optimization is performed for multiple monitoring locations to obtain a Pareto optimal set of optimized anomaly analysis schemes, including: Obtain multiple anomaly analysis agents corresponding to multiple pre-configured monitoring locations, and test the anomaly analysis time for each agent to obtain multiple expected times; Based on the difference array, update time parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality.

6. The water conservancy project operation monitoring management method based on digital twinning according to claim 5, characterized in that, The configuration steps for multiple anomaly analysis agents include: Based on monitoring data from multiple monitoring locations over a historical period, multiple sets of sample monitoring parameters are collected, and each sample monitoring parameter is labeled as abnormal, resulting in multiple sets of sample anomaly analysis results. Based on machine learning, multiple anomaly analysis agents are constructed. The input data for each anomaly analysis agent is the monitoring parameters of the corresponding monitoring location, and the output data is the anomaly analysis result. Multiple sets of sample monitoring parameters and multiple sets of sample anomaly analysis results are used as training data and supervision labels to conduct supervised training and testing on multiple anomaly analysis agents. Configuration is completed after the test is passed.

7. The water conservancy project operation and maintenance monitoring management method based on digital twin as described in claim 5, characterized in that, Based on the difference array, update timeliness parameters, and multiple expected times, anomaly analysis optimization is performed at multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that satisfy Pareto optimality, including: Obtain the memory constraints for making calls to the intelligent agent, and use the total anomaly analysis time being less than the update timeliness parameter as the timeliness constraint; If the computing power of a single call to an agent satisfies the memory constraint and the sum of the expected times for synchronous anomaly analysis by the called agents satisfies the timeliness constraint, multiple anomaly analysis agents are randomly selected to obtain a first anomaly analysis scheme. The sum of the differences of all real-time monitoring parameters used for anomaly analysis within the first anomaly analysis scheme is obtained to obtain the total difference of the first analysis. Continue to randomly select an anomaly analysis agent until the convergence optimization count is reached; During the optimization process, the anomaly analysis scheme with the largest total difference in analysis among the anomaly analysis schemes of each anomaly analysis agent is recorded as the set of optimized anomaly analysis schemes that satisfy Pareto optimality.

8. The water conservancy project operation monitoring management method based on digital twinning according to claim 1, characterized in that, Based on the optimized anomaly analysis scheme set and historical monitoring records from multiple monitoring locations over a historical period, the optimal anomaly analysis scheme is selected, and the corresponding anomaly analysis agent is invoked to perform anomaly analysis, resulting in an anomaly analysis result set. The real-time twin model is then labeled as the monitoring management result, including: Acquire historical monitoring records from multiple monitoring locations within a historical period and process multiple monitoring anomalies at these locations; Based on multiple monitoring anomalies, the optimal anomaly analysis scheme is selected from the optimized anomaly analysis scheme set. Among them, the sum of monitoring anomalies of the anomaly analysis agents called in the optimal anomaly analysis scheme is the largest. The anomaly analysis agent within the optimal anomaly analysis scheme is invoked, the corresponding real-time monitoring parameters are input, and the anomaly analysis result set is output. The real-time twin model is then labeled as the monitoring and management result.

9. The water conservancy project operation monitoring management method based on digital twinning according to claim 8, characterized in that, Acquire historical monitoring records from multiple monitoring locations within a historical time period, and process multiple monitoring anomalies at these locations, including: Acquire historical monitoring records from multiple monitoring locations within a historical time period; Extract the number of times anomalies occur at multiple monitoring locations to obtain multiple monitoring anomaly counts, and calculate multiple monitoring anomaly degrees.

10. A water conservancy project operation monitoring and management system based on digital twinning, characterized in that, The system is used to implement the water conservancy project operation and maintenance monitoring and management method based on digital twins as described in any one of claims 1-9, the system comprising: The data acquisition model construction module is used to collect real-time monitoring parameter arrays through sensor arrays deployed at multiple monitoring locations within the water conservancy project, and to construct the basic parameters within the real-time twin model based on digital twins. The model difference analysis module is used to calculate the difference array between the real-time twin model and the historical twin model monitored last time, perform model update requirement analysis, and obtain update timeliness parameters; An anomaly analysis optimization module is used to obtain multiple expected times for anomaly analysis agents configured for multiple monitoring locations to perform anomaly analysis. Combining the difference array and update timeliness parameters, it optimizes the anomaly analysis for multiple monitoring locations to obtain a set of optimized anomaly analysis schemes that meet Pareto optimality. The anomaly analysis execution module is used to select the optimal anomaly analysis scheme based on the optimized anomaly analysis scheme set and historical monitoring records of multiple monitoring locations within a historical period, call the corresponding anomaly analysis agent to perform anomaly analysis, obtain an anomaly analysis result set, and label the real-time twin model as the monitoring management result.

Images