Digital twin system and construction method for long-distance water diversion project across regions

Abstract

The present application relates to the technical field of digital twinning of water diversion projects, and discloses a digital twinning system and construction method for cross-regional long-distance water diversion projects. The system comprises a field perception and communication networking layer, an edge aggregation and autonomous processing layer, a cloud digital twinning platform layer, and a dispatching control execution layer. The edge aggregation and autonomous processing layer extracts shallow time sequence features and deep working condition features, calculates the model value weight in combination with the communication link state and the model prediction error, thereby adaptively reconstructing the data packet, and screening high-trust model input data through communication integrity, numerical rationality, and multi-level physical constraint verification. The dispatching control execution layer executes phased regulation and control after safety constraint checking, and feeds back the execution result. The present application can improve the data model input credibility, model prediction accuracy, and dispatching control safety under weak network and complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology for water transfer projects, specifically to a digital twin system and construction method for cross-regional long-distance water transfer projects. Background Technology

[0002] Long-distance inter-regional water transfer projects typically include multiple hydraulic engineering units such as canals, tunnels, sluice gates, pumping stations, reservoirs, water diversion nodes, flood discharge facilities, and pressure relief and diversion facilities. They are characterized by long water conveyance lines, numerous engineering nodes, complex operating conditions, and a wide range of impacts on scheduling. With the increasing informatization and intelligentization of water conservancy projects, digital twin technology is gradually being applied to the operation and management of long-distance water transfer projects. By deploying monitoring equipment at the project site for water level, flow rate, pressure, sluice gate position, pumping station status, water quality, sediment, and structural safety, and combining this with hydrodynamic models, scheduling models, and visualization platforms, real-time mapping, simulation prediction, and auxiliary scheduling of the project's operating status can be achieved. Existing digital twin systems for inter-regional water transfer projects typically focus on constructing a three-dimensional scene of the project, accessing hydrological monitoring data, establishing hydrodynamic simulation models, conducting water demand prediction, inflow prediction, and optimizing scheduling schemes. Some systems also combine weather forecasts, hydrological forecasts, spatial topology models, and intelligent optimization algorithms to calculate water allocation, pump station start-up and shutdown, gate opening, and emergency dispatch plans, in order to improve the operational efficiency and dispatch automation level of water transfer projects.

[0003] In the actual operation of long-distance water transfer projects across regions, especially in mountainous areas, tunnels, remote canal sections, and complex terrain areas, the acquisition and transmission conditions of on-site monitoring data are often poor. Some monitoring nodes suffer from unstable communication links, transmission delays, packet loss, out-of-order delivery, duplicate uploads, or network outages causing buffering issues. Existing digital twin systems typically upload on-site collected data directly to cloud platforms or dispatch centers for modeling and calculation. However, they lack sufficient spatiotemporal calibration, operational condition identification, data value assessment, and credibility processing before the data is incorporated into the model. This results in data entering the model that may have issues such as time asynchrony, unclear spatial correlations, inconsistent physical meanings, or insufficient credibility. Under complex operating conditions such as floods, droughts, high-sediment-laden turbid water flows, rapid scouring and silting evolution, abnormal leakage, and low-temperature freezing in winter, the operational status changes of water transfer projects exhibit strong time-varying and coupled characteristics. For example, flooding can cause rapid changes in water level, flow rate, and pressure; high sediment loads can affect sediment transport, cross-sectional siltation, channel roughness, and flow capacity; changes in cross-sectional scouring and silting can further alter hydrodynamic propagation characteristics; abnormal leakage can lead to water loss and structural safety risks; and in low-temperature environments, tunnel icing, ice blockage, freeze-thaw damage, and sudden changes in local hydraulic conditions may also occur. Summary of the Invention

[0004] The purpose of this invention is to provide a digital twin system and construction method for cross-regional long-distance water transfer projects to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The digital twin system for inter-regional long-distance water transfer projects includes: The field sensing and communication networking layer is used to collect data on water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status, and temperature from channels, tunnels, sluice gates, pumping stations, reservoirs, and water distribution nodes, and upload them to the edge aggregation and autonomous processing layer. An edge aggregation and autonomous processing layer is used to perform preprocessing on the data before it is input into the model, including: Extract shallow temporal features and deep working condition features; Based on the shallow time-series characteristics, deep working condition characteristics, communication link status, and model prediction errors fed back from the cloud digital twin platform layer, the input value weight of each frame of monitoring data is calculated. Based on the input model value weight, each frame of monitoring data is adaptively reconstructed into a corresponding type of data packet, and the data packet is subjected to communication integrity verification, numerical rationality verification and multi-level physical constraint verification. Data packets that pass the verification are marked as high-confidence input model data. The cloud-based digital twin platform layer is used to perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model based on the high-reliability input model data, and generate simulation results of engineering operation status or scheduling and control strategies. The scheduling and control execution layer is used to perform control actions on gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, or pressure reduction and diversion facilities according to the scheduling and control strategy.

[0006] Preferably, the field sensing and communication networking layer: Acquire water level, flow rate, pressure, sediment content, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature-related data collected at each monitoring node; Based on the spatial location of the measuring points, unified timestamps, data types, and engineering topology, the data is calibrated spatiotemporally at the source end, and the sampling frequency is dynamically adjusted according to changes in water level, pressure fluctuations, sediment content, cross-sectional siltation, leakage, temperature, and link quality. Output multi-source raw monitoring data frames with spatial coding of measurement points, unified timestamp, engineering topology relationship, sampling frequency, link status and quality status markers.

[0007] Preferably, the edge convergence and autonomous processing layer: Obtain the original monitoring data frames from the multiple sources; Based on the unified timestamp, spatial coding of measurement points and engineering topology, time sequence alignment and spatial correlation are performed to extract shallow time sequence features and deep working condition features. The input value weight of each frame of monitoring data is calculated by combining the communication link status, source end quality status marking and cloud model prediction error. The data representation format, transmission priority, and input path are determined based on the input model value weight, and at least one of the following is output: differential data packet, simplified hazard feature packet, high-precision correction packet, water-sand coupling data packet, parameter correction data packet, ice condition control data packet, or offline continuation data packet.

[0008] Preferably, the edge convergence and autonomous processing layer: Obtain the adaptively reconstructed data packet; Based on at least one of the following: timestamp continuity, data sequence number, missing packets, duplicate packets, integrity check code, range, rate of change, abrupt change amplitude, water conservation, water and sediment movement, scouring and silting geometry, leakage response, low temperature ice conditions, gate position-flow response, and upstream and downstream propagation causality, the data packet is subjected to communication integrity verification, numerical rationality verification, and multi-level physical constraint verification. Output high-confidence input data, suspected abnormal data, low-confidence data, or data to be supplemented, and form a correction data chain based on the high-confidence input data, which includes data source, spatiotemporal identifier, confidence level, and model parameter mapping relationship.

[0009] Preferably, the cloud-based digital twin platform layer: Obtain the high-confidence input model data and the correction data chain; Based on data packet type, spatial coding of measurement points, unified timestamp, physical constraint verification results and data credibility, establish parameter mapping relationships between highly reliable input data and hydrodynamic models, water and sediment transport models, dynamic scouring and silting models, channel leakage models and tunnel structural safety assessment models, and perform hierarchical online parameter inversion updates and coupling consistency verification. Output simulation results of engineering operation status, flood risk early warning results, high sediment erosion and siltation evolution prediction results, seepage loss prediction results, low temperature ice risk assessment results, tunnel structure safety assessment results, or scheduling and control strategies.

[0010] Preferably, the scheduling control execution layer: Obtain the scheduling and control strategy output by the cloud-based digital twin platform layer; Based on strategy type, strategy priority, model credibility, working condition risk level and on-site execution constraints, hierarchical execution instructions are generated, and safety constraints are checked before execution based on gate opening change rate, pump station start and stop frequency, flood discharge flow rate, upstream and downstream water level difference, tunnel pressure change rate, seepage pressure, vibration and lining strain. Output phased control actions for gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, or pressure reduction and diversion facilities, and feed back the action feedback and engineering response to the edge aggregation and autonomous processing layer or the cloud digital twin platform layer.

[0011] This invention also provides a method for constructing a digital twin system for a cross-regional long-distance water transfer project, comprising: S1. Collect water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status and temperature-related data from channels, tunnels, sluice gates, pumping stations, reservoirs and water distribution nodes, and perform source-end spatiotemporal calibration and quality status marking on the data to form multi-source raw monitoring data frames. S2. Based on the multi-source original monitoring data frames, extract shallow time-series features and deep operating condition features. According to the shallow time-series features, deep operating condition features, communication link status, and model prediction error fed back by the cloud digital twin platform, calculate the input value weight of each frame of monitoring data. Based on the input value weight, adaptively reconstruct each frame of monitoring data into a corresponding type of data packet, and perform communication integrity verification, numerical rationality verification, and multi-level physical constraint verification on the data packets. Mark the data packets that pass the verification as high-confidence input data. S3. Based on the high-reliability input data, establish a mapping relationship between the high-reliability input data and the model parameters to be updated, and perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model. S4. Generate simulation results of engineering operation status or scheduling control strategy based on the updated model, and perform safety constraint verification before execution. Execute control actions or limit, delay, downgrade, execute in stages or roll back the scheduling control strategy according to the verification results. Use action feedback and engineering response feedback for model parameter correction, updating of input value weights and subsequent rolling generation of scheduling control strategies.

[0012] Preferably, the input value weight is determined by hydraulic mutation factor, sand erosion and deposition factor, working condition risk factor, link urgency factor, model error factor, data reliability factor and link occupancy factor; The hydraulic mutation factor is used to characterize the rate of change and mutation intensity of water level, flow rate or pressure; the sand erosion and sedimentation factor is used to characterize the degree of change of sediment content, turbidity or cross-sectional sedimentation; the link urgency factor is used to characterize the communication link delay, packet loss rate or buffer congestion; and the model error factor is used to characterize the deviation between the digital twin model prediction value and the field measured value. The edge aggregation and autonomous processing layer determines the data packet type, transmission priority, and input path of each frame of monitoring data based on the comparison results of the input value weight and the dynamic threshold; the dynamic threshold is adaptively adjusted based on historical high-reliability input data, current operating condition risk level, communication link quality, and model prediction error.

[0013] Preferably, when the edge convergence and autonomous processing layer identifies data disorder, delay, missing or duplicate data caused by multi-hop transmission in a weak network, it performs hydraulic causal reconstruction on the data based on the unified timestamp, spatial coding of measurement points, engineering topology relationship, upstream and downstream hydraulic propagation direction, gate position-flow response relationship and water level propagation time delay. The hydraulic causal reconstruction includes determining the temporal position of the data in the actual hydraulic process based on the sequential relationship of water level, flow rate or pressure changes between upstream measuring points, tunnel measuring points and downstream measuring points, and marking missing data or estimating and filling in missing data based on adjacent high-confidence input data. The edge convergence and autonomous processing layer outputs a time-series data chain and its reconstruction credibility after hydraulic causality reconstruction, which is used by the cloud-based digital twin platform layer to reconstruct the flood propagation process, correct hydrodynamic propagation parameters, or update the simulation results of engineering operation status.

[0014] Preferably, under low-temperature conditions in winter, the edge convergence and autonomous treatment layer identifies low-temperature ice condition characteristics based on water temperature, air temperature, tunnel wall temperature, soil temperature, water level, flow velocity, pressure, upstream and downstream gate opening degree, and low-temperature duration; the low-temperature ice condition characteristics include at least one of the following: freezing risk level, ice blockage risk level, freeze-thaw damage risk level, low-temperature stagnant water zone, and freeze-thaw sensitive water level zone. When the icing risk level reaches the preset antifreeze threshold and no ice blockage is detected, the cloud digital twin platform layer or the edge convergence and autonomous processing layer generates an anti-icing control strategy. This strategy maintains the minimum antifreeze flow rate in the tunnel by adjusting the upstream and downstream gates in a coordinated manner, shortens the water's residence time in the low-temperature tunnel section, and avoids the freeze-thaw sensitive water level range. When icing or ice blockage risk is detected in the tunnel, an icing impact mitigation strategy is generated. This strategy reduces the impact of ice blockage, water hammer, frost heave, or ice impact on the tunnel structure by adjusting upstream and downstream gates in stages, limiting the flow rate, limiting the water level change rate, reducing tunnel pressure fluctuations, or triggering pressure reduction and diversion operations.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention performs source-end spatiotemporal calibration and quality status marking on data related to water level, flow rate, pressure, sediment content, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature at the field sensing and communication network layer. This ensures that each frame of monitoring data is bound to a spatial code of the measuring point, a unified timestamp, data type, engineering topology, and link status. This clarifies the spatial location, temporal sequence, and hydraulic correlation of monitoring data in long-distance, cross-regional water transfer projects, reducing problems such as data disorder, delay, duplication, or unclear spatial attribution caused by weak network transmission, network outage caching, and multi-hop forwarding. It provides a structured and traceable raw data foundation for subsequent digital twin models.

[0016] This invention incorporates a pre-processing mechanism at the edge convergence and autonomous processing layer. By extracting shallow temporal features and deep operational features, it distinguishes between ordinary numerical fluctuations, real hydraulic abrupt changes, sensor anomalies, communication disturbances, and complex operational evolution. Shallow temporal features reflect instantaneous changes in water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, seepage, and temperature data. Deep operational features reflect operational states such as floods, droughts, high sediment concentration erosion and sedimentation, abnormal seepage, low-temperature ice conditions, upstream and downstream propagation delays, and communication link degradation. This improves the system's ability to identify complex hydrological conditions, sediment, structural, and communication states, avoiding misjudgments or omissions caused by relying solely on a single threshold. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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.

[0018] Figure 1 This is a flowchart of the digital twin system construction method of the present invention.

[0019] Figure 2 This is a module connection diagram of the digital twin system of the present invention.

[0020] Figure 3 This is a timing diagram of data processing, model updating, and scheduling execution of the digital twin system of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1-2 As shown, the present invention provides a technical solution: a digital twin system for a cross-regional long-distance water transfer project, comprising: The field sensing and communication networking layer is used to collect data on water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status, and temperature from channels, tunnels, sluice gates, pumping stations, reservoirs, and water distribution nodes, and upload them to the edge aggregation and autonomous processing layer. An edge aggregation and autonomous processing layer is used to perform preprocessing on the data before it is input into the model, including: Extract shallow temporal features and deep working condition features; Based on the shallow time-series characteristics, deep working condition characteristics, communication link status, and model prediction errors fed back from the cloud digital twin platform layer, the input value weight of each frame of monitoring data is calculated. Based on the input model value weight, each frame of monitoring data is adaptively reconstructed into a corresponding type of data packet, and the data packet is subjected to communication integrity verification, numerical rationality verification and multi-level physical constraint verification. Data packets that pass the verification are marked as high-confidence input model data. The cloud-based digital twin platform layer is used to perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model based on the high-reliability input model data, and generate simulation results of engineering operation status or scheduling and control strategies. The scheduling and control execution layer is used to perform control actions on gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, or pressure reduction and diversion facilities according to the scheduling and control strategy.

[0023] In this embodiment, the field sensing and communication network layer is deployed in zones according to the physical composition of the cross-regional long-distance water transfer project. Water level monitoring equipment, flow monitoring equipment, sediment concentration monitoring equipment, turbidity monitoring equipment, cross-sectional sedimentation monitoring equipment, leakage monitoring equipment, and temperature monitoring equipment are deployed in the channel section; pressure monitoring equipment, leakage monitoring equipment, temperature monitoring equipment, and structural response monitoring equipment are deployed in the tunnel section; and water level, flow, pressure, equipment status, and communication status acquisition equipment corresponding to their operating status are deployed at sluice gates, pumping stations, reservoirs, and water distribution nodes. Each monitoring device generates a frame of monitoring data during sampling. The monitoring data includes at least the measuring point number, project object number, sampling time, data type, measured value, equipment status, sensor range, calibration parameters, and communication link identifier. The field sensing and communication network layer aggregates the monitoring data generated within the same sampling period according to the project object number and uploads it to the edge aggregation and autonomous processing layer via fiber optic private network, industrial Ethernet, cellular communication, microwave communication, satellite communication, or multi-link redundant communication.

[0024] After receiving the monitoring data, the edge aggregation and autonomous processing layer first performs time alignment processing on the monitoring data. For monitoring data of different data types within the same project object, a time series is formed based on a unified sampling period. When multiple data of the same type exist at the same sampling time, the data with normal equipment status, smallest timestamp deviation, and passed the initial communication integrity check is selected as the valid candidate data for that sampling time. When a certain data type is missing at the same sampling time, interpolated data is not used to replace the measured data; instead, a missing measurement flag is written into the data field so that the missing data can be identified in the subsequent verification process. After completing the time alignment, the edge aggregation and autonomous processing layer calibrates the measured values ​​according to the sensor calibration records. The calibration process is as follows: multiply the measured value by the calibration slope corresponding to the sensor, and add the calibration offset corresponding to the sensor to obtain the calibrated monitoring value. The calibration slope and calibration offset are derived from the calibration records before equipment commissioning, periodic verification records, and on-site verification records. When the sensor is replaced or recalibrated, the edge aggregation and autonomous processing layer uses the latest version of the calibration parameters to process subsequent monitoring data.

[0025] The edge convergence and autonomous processing layer extracts shallow temporal features from the calibrated monitoring data. For each measuring point and each data type, a sliding window is formed by taking several consecutive sampled values ​​prior to the current sampling time. The length of the sliding window is determined according to the physical response time of the monitored object. Specifically, the sliding window length for channel water level, channel flow, and reservoir water level is determined by dividing the minimum hydraulic propagation period specified in the operation and scheduling procedures by the sampling period and rounding up; the sliding window length for pressure, leakage, and communication link status is determined by dividing the anomaly identification period specified in the equipment technical procedures by the sampling period and rounding up; and the sliding window length for sediment concentration, turbidity, and cross-sectional siltation is determined by dividing the shortest effective observation period required in the sediment monitoring procedures by the sampling period and rounding up.

[0026] Shallow time-series features include window mean, window dispersion, current difference, rate of change, window range, and trend slope. The window mean is obtained by averaging all calibration values ​​within the sliding window and is used to characterize the stability level of the monitored object over a recent period. Window dispersion is obtained by calculating the deviation of each calibration value within the sliding window from the window mean and is used to characterize the fluctuation intensity of water level, flow rate, pressure, sediment concentration, turbidity, leakage, or communication link status. The current difference is obtained by subtracting the calibration value at the previous sampling time from the calibration value at the current sampling time. The rate of change is obtained by dividing the current difference by the sampling period. The window range is obtained by subtracting the minimum calibration value from the maximum calibration value within the sliding window. The trend slope is obtained by fitting a linear trend between the sampling time and the corresponding calibration value within the sliding window; the fitting result reflects the direction of change of this data type within the window, whether it is continuously rising, continuously falling, or basically stable. The above shallow time-series features are calculated at the edge side and enter the subsequent processing flow along with the current monitoring frame.

[0027] The edge convergence and autonomous processing layer further extracts in-depth operating condition characteristics. For the same engineering object, the edge convergence and autonomous processing layer arranges the water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature-related data within the same time range according to the sampling time sequence, forming an operating condition input data group. Before entering the feature extraction model, the operating condition input data group undergoes dimensionless processing. The dimensionless processing process is as follows: subtract the median of qualified data of this data type at the current calibration value from the median of qualified data at the measurement point within the last 30 natural days, and then divide the difference by the interquartile range of the same qualified data set. The qualified data refers to data that has passed communication integrity verification, numerical rationality verification, and physical constraint verification. If there are maintenance, shutdown, trial operation, or sensor replacement situations within the last 30 natural days, the corresponding time period is removed and the median and interquartile range are recalculated; when the number of effective samples is insufficient, qualified operating data formed during the initial commissioning of the project is used as the statistical basis.

[0028] It should be noted that the dimensionless input data sets of the operating conditions are fed into the temporal feature extraction model deployed at the edge. This temporal feature extraction model consists of a temporal convolution processing unit, a gated loop processing unit, and an operating condition output unit. The temporal convolution processing unit extracts local change features between adjacent sampling points to identify short-term water level jumps, pressure fluctuations, flow rate abrupt changes, turbidity changes, and communication quality degradation. The gated loop processing unit reads the output results of the temporal convolution processing unit in chronological order to identify the temporal continuity features of gate regulation, pump station start-up and shutdown, flood discharge, water diversion adjustment, low-temperature operation, high-sediment inflow, suspected leakage, suspected siltation, and suspected tunnel structural anomalies. The operating condition output unit outputs the identification results and corresponding confidence levels for various operating conditions. The training samples for the temporal feature extraction model are derived from historical operating data, scheduling records, equipment action records, inspection records, and abnormal event records confirmed by operation management personnel. After the model is verified to meet the engineering operation identification requirements, it is distributed from the cloud-based digital twin platform layer to the edge aggregation and autonomous processing layer.

[0029] The edge aggregation and autonomous processing layer calculates the input value weight of each monitoring data frame based on shallow temporal characteristics, deep operational characteristics, communication link status, and model prediction errors fed back from the cloud-based digital twin platform layer. The input value weight is used to determine the contribution of the current monitoring frame to model correction, anomaly identification, and scheduling control. Its calculation process consists of shallow temporal scoring, deep operational condition scoring, communication link degradation scoring, and model prediction error scoring. These four scores are then weighted and synthesized according to their respective weights to obtain the input value weight of the current monitoring frame.

[0030] The shallow time series score is determined by the rate of change, window dispersion, and window range. The edge aggregation and autonomous processing layer compares the rate of change of the current monitoring frame with the rate of change threshold, the window dispersion with the fluctuation threshold, and the window range with the range threshold, and obtains the corresponding anomaly degree for each. The above three anomaly degrees are weighted according to data type to form the shallow time series score. The rate of change threshold, fluctuation threshold, and range threshold are determined according to the measurement point and data type. The specific determination process is as follows: take qualified data that has passed all verifications within the last 30 natural days, and statistically analyze the historical distribution of the rate of change, window dispersion, and window range. The rate of change threshold is taken as the 95th quantile of the absolute value of the rate of change, the fluctuation threshold is taken as the 95th quantile of the window dispersion, and the range threshold is taken as the 95th quantile of the window range. If there is maintenance, shutdown, trial operation, or sensor replacement within the last 30 natural days, the corresponding time period is excluded and the statistics are recalculated. When there are insufficient effective samples, the rate of change threshold is the smaller of the maximum rate of change allowed by the engineering scheduling procedure and the rate of change allowed by the sensor range; the fluctuation threshold is the larger of the equipment measurement uncertainty and the historical joint debugging fluctuation range; and the range threshold is the smaller of the allowable fluctuation range of the design operating boundary and the allowable fluctuation range of the equipment range.

[0031] Specifically, the deep working condition score is determined by the identification results of the working condition output unit. The edge convergence and autonomous processing layer sets working condition weights for each of the following: suspected leakage, suspected siltation, suspected tunnel structural anomalies, high sediment load water, gate regulation, pump station start-up and shutdown, and flood discharge. The weight of working conditions involving engineering safety boundaries is greater than that of conventional steady-state water conveyance working conditions; the weight of working conditions involving changes in water supply distribution is greater than that of ordinary small water level fluctuations. The edge convergence and autonomous processing layer multiplies the confidence level of each working condition by its corresponding working condition weight and then sums them to obtain the deep working condition score. The working condition weights are determined by the cloud-based digital twin platform layer based on the results of historical event playback. When determining the weights, confirmed leakage events, siltation events, structural anomaly events, flood discharge events, water distribution adjustment events, and steady-state operation segments are used as verification samples. Weight combinations that can improve the retention rate of input data when anomalies occur without significantly increasing the amount of steady-state redundant data are selected.

[0032] The communication link degradation score is determined by link latency, packet loss rate, bit error rate, jitter, and bandwidth utilization. The edge aggregation and autonomous processing layer compares these communication indicators with the upper limits of the engineering communication design indicators or operation and maintenance management indicators to obtain the degree of degradation for each indicator. These indicators are then weighted and summed according to the communication evaluation weights to obtain the link degradation degree. The communication link score is obtained by subtracting the link degradation degree from the full score, and the score increases as the link degradation degree increases. The upper limits for link latency, packet loss rate, bit error rate, jitter, and bandwidth utilization are taken from the limits specified in the communication design documents, operation and maintenance procedures, and link service quality records. When multiple communication modes exist for the same link, the corresponding limit is selected according to the communication mode currently carrying the monitored frame. The communication availability threshold is taken as the quintile of the communication link score under normal communication conditions over the past 30 calendar days, and is not lower than the minimum availability score specified in the communication operation and maintenance procedures. When the current communication link score is lower than the communication availability threshold, the edge aggregation and autonomous processing layer adds a link degradation flag to the monitoring frame; monitoring data involving water level exceeding limits, pressure exceeding limits, flood discharge, leakage, structural safety, and water distribution control are not directly removed due to link degradation.

[0033] In this embodiment, the model prediction error score is determined by the difference between the predicted value fed back by the cloud-based digital twin platform layer and the field calibration value. The cloud-based digital twin platform layer outputs the predicted values ​​for each data type at each measurement point in the previous calculation cycle and feeds these predicted values ​​back to the edge convergence and autonomous processing layer. The edge convergence and autonomous processing layer subtracts the current field calibration value from the corresponding predicted value and takes the absolute difference. Then, it normalizes the result using the larger of the absolute value of the current field calibration value and 10 times the sensor resolution as a benchmark, obtaining the relative prediction error for that data type. The relative prediction error of each data type is compared with its corresponding prediction error threshold, and then weighted according to the data type weights to form the model prediction error score. The prediction error threshold is determined by the cloud-based digital twin platform layer after the model's phased calibration. Specifically, it statistically analyzes the distribution of the relative prediction error for each data type at each measurement point in the model validation dataset and takes the 95th quantile as the prediction error threshold for that data type at that measurement point. After the project undergoes changes in water diversion flow level, pump station operation mode, seasonal temperature boundary changes, or significant changes in sediment conditions, the cloud-based digital twin platform layer recalculates and predicts the error threshold using all verified data within 7 consecutive natural days after the switch, and sends the updated results to the edge aggregation and autonomous processing layer.

[0034] In this embodiment, the edge aggregation and autonomous processing layer weights and synthesizes the shallow time-series score, deep operating condition score, communication link degradation score, and model prediction error score according to preset calculation coefficients to obtain the input value weight of each frame of monitoring data. The preset calculation coefficients are determined by the cloud digital twin platform layer through historical operation playback. The determination process is as follows: confirmed abnormal events, scheduling events, and steady-state operation segments are selected as verification samples, and the high-value data recall rate, redundant data compression rate, and model error reduction rate under different combinations of calculation coefficients are tested respectively; under the premise of meeting the minimum recall requirements of abnormal events stipulated by the operation management unit, the calculation coefficient combination with a higher redundant data compression rate and a larger model error reduction rate is selected. The determined calculation coefficients are stored according to the engineering object, measurement point category, and data type. Different channel sections, tunnel sections, gate stations, pumping stations, reservoirs, and water distribution nodes use calculation coefficients corresponding to their operating characteristics.

[0035] Specifically, the edge aggregation and autonomous processing layer adaptively reconstructs each frame of monitoring data into a corresponding data packet type based on the input model value weight. Data packet reconstruction uses high-value thresholds and low-value thresholds. The high-value threshold is the 85th quantile of the input model value weights of qualified monitoring frames within the most recent 30 calendar days, and the low-value threshold is the 45th quantile of the input model value weights of the same dataset. If there are maintenance, shutdown, trial operation, or sensor replacement events within the most recent 30 calendar days, the corresponding time period is removed and the calculation is recalculated. When the project is initially put into operation and there are insufficient valid samples, the high-value threshold is set to 0.75 and the low-value threshold to 0.35; after continuous operation for 30 calendar days and the formation of sufficient qualified data, the quantile threshold is switched to.

[0036] Furthermore, when the input value weight of the current monitoring frame reaches or exceeds the high-value threshold, the edge aggregation and autonomous processing layer reconstructs the monitoring frame into a high-precision data packet. The high-precision data packet retains the original sampled value, calibration value, sampling time, measurement point number, spatial location code, equipment status, communication link indicators, shallow temporal features, deep operating condition features, model prediction error, and quality label. When the input value weight of the current monitoring frame is lower than the high-value threshold but reaches or exceeds the low-value threshold, the edge aggregation and autonomous processing layer reconstructs the monitoring frame into a feature-enhanced data packet. The feature-enhanced data packet retains the calibration value, sampling time, measurement point number, shallow temporal feature summary, operating condition identification result, model prediction error, and quality label. When the input value weight of the current monitoring frame is lower than the low-value threshold, the edge aggregation and autonomous processing layer reconstructs the monitoring frame into a compressed summary data packet. The compressed summary data packet retains the measurement point number, sampling period, window mean, maximum value, minimum value, rate of change, and quality label. The above reconstruction results are directly determined by the comparison results of input value weights and thresholds, ensuring that mutation data, abnormal risk data and data with high model correction value are retained as more complete data packages, and stable running data are compressed and expressed.

[0037] Specifically, the edge aggregation and autonomous processing layer performs communication integrity checks on the reconstructed data packets. Communication integrity checks include sequence number continuity checks, timestamp monotonicity checks, field integrity checks, and content checks. Sequence number continuity checks determine if there are missing, duplicate, or skipped packets of the same data type at the same measurement point. Timestamp monotonicity checks determine if data packets are out of order, have abnormal return transmissions, or have sampling time conflicts. Field integrity checks determine if a data packet contains all the required fields for its data packet type. Content checks compare the checksum generated at the edge with the checksum regenerated at the receiver to determine if the content has changed during data transmission. If any checksum fails, the data packet is marked as incomplete communication data, and the defect type is fed back to the corresponding communication link detection unit, preventing it from entering the high-confidence input data set.

[0038] After the communication integrity check passes, the edge aggregation and autonomous processing layer performs numerical rationality checks on the data packets. Numerical rationality checks include absolute boundary checks, rate of change boundary checks, and consistency checks between adjacent measuring points. Absolute boundary checks determine whether data related to water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, and temperature are within permissible ranges. Permissible ranges are jointly determined by engineering design documents, operation and scheduling procedures, sensor ranges, and equipment protection settings; when the design operating boundary is stricter than the sensor range, the design operating boundary is used; when the equipment protection setting is stricter than the design operating boundary, the equipment protection setting is used. Rate of change boundary checks determine whether the rate of change of the current sampled value relative to the previous sampled value exceeds the permissible rate of change. The permissible rate of change is the smaller of the 99th percentile of the absolute values ​​of the rates of change of qualified data over the most recent 30 calendar days and the maximum permissible rate of change in the engineering scheduling procedures. Adjacent measuring point consistency verification is used to determine whether adjacent water levels, flow rates, and pressures in the same water conveyance direction conform to the hydraulic relationship between the engineering objects. The allowable difference between adjacent measuring points is determined based on the distance between the two measuring points, channel slope, gate status, pumping station status, water distribution status, and current scheduling boundary. Data packets that fail the numerical rationality verification are marked as suspected abnormal data and are not included in the high-confidence model input data set.

[0039] Furthermore, after the numerical rationality verification is passed, the edge aggregation and autonomous processing layer performs multi-level physical constraint verification on the data packets. Level 1 physical constraint verification is performed on a single engineering object. For channel sections, the verification checks whether the water level exceeds the design control water level, whether the flow rate exceeds the cross-sectional water conveyance capacity, whether the trends of sediment content and turbidity are consistent, whether the cross-sectional siltation is non-negative and the change process is continuous, and whether the leakage is within the allowable range for each section. For tunnel sections, the verification checks whether the pressure exceeds the allowable pressure, whether the pressure change rate exceeds the limit given by the structural safety assessment model, whether the leakage is non-negative and does not exceed the allowable leakage limit for each section, and whether the temperature change exceeds the range specified in the structural safety evaluation document. For pumping stations, the verification checks whether the inlet and outlet pressures, flow rates, head, and pump unit operating status match. For sluice gates, the verification checks whether the gate opening, upstream and downstream water level difference, and gate flow rate match. For reservoirs, the verification checks whether the inflow, outflow, and reservoir water level changes meet the reservoir capacity requirements. The water distribution node verification checks whether the water distribution flow, node water level, and water distribution facility status meet the water distribution plan; the second-level physical constraint verification is performed on adjacent engineering objects. The edge convergence and autonomous treatment layer are judged according to the continuity of water volume. The specific process is as follows: first, obtain the flow rate entering the control section from upstream, then obtain the flow rate leaving the control section from downstream; subsequently, calculate the change in water storage capacity based on the water level change and cross-sectional geometric parameters within the control section, then calculate the lost water volume based on the channel leakage model or segmented leakage monitoring results; finally, subtract the downstream outflow, the flow rate corresponding to the change in water storage capacity, and the lost flow rate from the upstream inflow to obtain the water balance residual. The allowable threshold for the water balance residual is jointly determined by the uncertainty of upstream flowmeter measurement, downstream flowmeter measurement, water storage calculation uncertainty, and leakage estimation uncertainty. The determination method is to square each uncertainty, add them together, and then take the square root of the sum. When the water balance residual does not exceed the allowable threshold, it is determined to pass the water continuity constraint; when the water balance residual exceeds the allowable threshold, it is determined to be physically inconsistent data; the third-level physical constraint verification is performed on the entire line's operational boundary. The edge aggregation and autonomous processing layer reads the current scheduling boundary issued by the cloud digital twin platform layer and determines whether the water level, water distribution, flow limit, pressure release action, flood discharge facility status, and pump station operation mode along the line meet the approved water transfer plan, operation scheduling procedures, equipment protection settings, and safety evaluation report. After the current data packet passes the communication integrity verification, numerical rationality verification, and the first, second, and third-level physical constraint verifications, the edge aggregation and autonomous processing layer marks it as high-confidence input data and uploads it to the cloud digital twin platform layer.

[0040] In this embodiment, after receiving the highly reliable input model data, the cloud-based digital twin platform layer maps the data to the corresponding digital twin objects according to the engineering topology. The channel object receives data related to water level, flow rate, sediment concentration, turbidity, cross-sectional siltation, leakage, and temperature; the tunnel object receives data related to pressure, leakage, temperature, and structural safety; the sluice gate object receives upstream and downstream water levels, gate status, and flow rate; the pumping station object receives inlet and outlet pressure, flow rate, and pump status; the reservoir object receives inlet flow rate, outlet flow rate, and reservoir water level; and the water distribution node object receives water distribution flow rate, node water level, and control facility status. After mapping, the cloud-based digital twin platform layer performs online parameter inversion updates on at least one of the following models based on the highly reliable input model data: hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model, and tunnel structural safety assessment model.

[0041] The online parameter inversion of the hydrodynamic model uses high-confidence water level and high-confidence flow rate as observational basis. The cloud-based digital twin platform first uses the hydrodynamic parameters from the previous calculation cycle to perform simulations, obtaining the simulated water level and flow rate at each control section. Then, it compares the simulated water level with the measured water level point by point, and the simulated flow rate with the measured flow rate point by point, obtaining the water level deviation and flow rate deviation. Subsequently, the deviations are weighted according to the sensor measurement uncertainty, giving higher weight to data with higher measurement accuracy in the parameter inversion. Finally, within the allowable range of channel roughness, local head loss coefficient, and boundary flow correction, parameter combinations are searched to reduce the overall weighted water level deviation and flow rate deviation, and to ensure that the updated parameters remain continuous with the parameters from the previous calculation cycle. The allowable range is determined by engineering design documents, channel lining material parameters, historical calibration results, and operation management documents.

[0042] The online parameter inversion of the water and sediment transport model uses high-confidence sediment concentration and high-confidence turbidity as observation basis. The cloud-based digital twin platform compares the sediment concentration change trend and turbidity change trend calculated by the model with the measured results, and adjusts the sediment settling parameters, start-up parameters and sediment transport capacity parameters according to the deviation direction.

[0043] The online parameter inversion of the dynamic scour and sedimentation model uses high-confidence cross-sectional sedimentation volume as the observation basis. By comparing the cross-sectional changes predicted by the model with the measured cross-sectional sedimentation volume, the scour coefficient and sedimentation coefficient are corrected. The online parameter inversion of the channel leakage model uses segmented leakage volume and water balance residual as the observation basis. By comparing the leakage volume calculated by the model with the leakage volume monitored in the field, the segmented permeability coefficient and lining leakage parameters are corrected. The online parameter inversion of the tunnel structural safety assessment model uses pressure, temperature-related data, leakage volume, and structural response data as the observation basis. By comparing the structural response calculated by the model with the field monitoring results, the pressure load parameters, temperature influence parameters, and safety evaluation parameters are corrected. Boundary checks are performed after updating the parameters of all models. The updated parameters must not exceed the ranges defined in the design documents, material performance documents, equipment protection documents, and structural safety evaluation documents.

[0044] After the model parameters are updated, the cloud-based digital twin platform generates simulation results of the engineering operation status. These simulation results include water level distribution along the route, flow distribution, pressure distribution, water and sediment transport trends, cross-sectional scouring and silting changes, channel leakage risk values, and tunnel structural safety evaluation values. The channel leakage risk value is determined by the current leakage volume, leakage growth rate, and water balance residual. Specifically, the closer the current leakage volume is to the segmental allowable leakage limit, the greater the leakage growth rate, and the larger the water balance residual, the higher the channel leakage risk value. The tunnel structural safety evaluation value is obtained by inputting pressure load, temperature change, leakage volume, and structural response data into the tunnel structural safety assessment model. When the pressure approaches the allowable pressure, the temperature change exceeds the range specified in the safety evaluation document, the leakage volume continues to increase, or the structural response deviates from the historical stable range, the tunnel structural safety evaluation value is adjusted towards a higher risk level.

[0045] The cloud-based digital twin platform layer compares simulation results with scheduling targets and operational boundaries. Scheduling targets are derived from approved water transfer plans, water supply allocation plans, and engineering operation management requirements; operational boundaries are derived from engineering operation scheduling procedures, equipment protection settings, safety evaluation reports, and hydraulic structure design documents. When the water level, flow rate, pressure, water distribution, leakage risk value, or tunnel structural safety evaluation value at a certain control section exceeds the corresponding boundary, or when prediction results indicate that it will approach the corresponding boundary in subsequent scheduling cycles, the cloud-based digital twin platform layer generates a scheduling control strategy. This scheduling control strategy uses reducing water supply target deviations, reducing water level exceedances along the route, reducing pressure exceedances, reducing leakage risk, reducing structural safety risk, and limiting frequent equipment operations as calculation objectives. Control results are determined from gate opening, pump station start / stop, flood discharge facility opening / closing, water distribution facility allocation, flow restriction facility flow limit, and pressure reducing / diversion facility operation. After the scheduling control strategy is generated, the cloud-based digital twin platform layer first performs a safety check. The gate control strategy verifies whether the single gate opening change exceeds the equipment's allowable value, and whether the upstream and downstream water level difference meets safety requirements after the gate action; the pump station control strategy verifies whether the pump start-stop interval, target speed, inlet and outlet water pressure, and operating flow rate meet equipment protection requirements; the flood discharge facility control strategy verifies whether the downstream water level exceeds the control water level after flood discharge; the water diversion facility control strategy verifies whether the adjusted water diversion volume at each water diversion node meets the allocation constraints; and the flow restriction facility and pressure reducing diversion facility control strategies verify whether the pressure exceeds the allowable pressure after action. After the safety verification is passed, the cloud-based digital twin platform layer sends the scheduling control strategy to the scheduling control execution layer.

[0046] After receiving the dispatch control strategy, the dispatch control execution layer decomposes the strategy into control instructions that the corresponding facilities can execute. Gate control instructions include target opening degree, direction of action, speed of action, and confirmation conditions for arrival; pump station control instructions include start / stop status, pump group number, target speed, and operating sequence; flood discharge facility control instructions include opening / closing status, opening amplitude, and duration; water diversion facility control instructions include target water diversion flow rate and adjustment period; flow restriction facility control instructions include flow restriction target value and closing amplitude; pressure reduction and diversion facility control instructions include diversion ratio and target pressure. Before issuing control instructions, the dispatch control execution layer reads the status of the field equipment and issues control instructions only after confirming that the equipment is in a permissible operating state. After the equipment completes its action, the dispatch control execution layer transmits back the action completion status, actual opening degree, actual speed, actual flow rate, action time, and any abnormal status. The field sensing and communication network layer collects data on water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature after the action, and then enters the next round of edge processing, cloud inversion, and dispatch execution process.

[0047] The field sensing and communication networking layer: Acquire water level, flow rate, pressure, sediment content, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature-related data collected at each monitoring node; Based on the spatial location of the measuring points, unified timestamps, data types, and engineering topology, the data is calibrated spatiotemporally at the source end, and the sampling frequency is dynamically adjusted according to changes in water level, pressure fluctuations, sediment content, cross-sectional siltation, leakage, temperature, and link quality. Output multi-source raw monitoring data frames with spatial coding of measurement points, unified timestamp, engineering topology relationship, sampling frequency, link status and quality status markers.

[0048] In this embodiment, the field sensing and communication network layer configures monitoring nodes at channels, tunnels, sluice gates, pumping stations, reservoirs, and water distribution nodes. Each monitoring node is programmed with a fixed spatial code before commissioning. The spatial code is generated hierarchically according to the water diversion project route, management section, project object, control section, monitoring point location, and equipment serial number, ensuring that each water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status, and temperature-related data has a unique spatial assignment at the source. The spatial code for channel sections includes the channel station number, left or right bank location, cross-section number, and sensor installation elevation; the spatial code for tunnel sections includes the tunnel number, tunnel mileage, lining zone, and pressure or leakage monitoring point location; the spatial code for sluice gates, pumping stations, reservoirs, and water distribution nodes includes the building number, equipment unit number, and monitoring location. Using the above encoding method, the subsequent edge processing layer can directly identify whether the data comes from the upstream section, downstream section, in front of the gate, behind the gate, in front of the pump, behind the pump, water diversion outlet, flood discharge outlet, or tunnel risk section.

[0049] Furthermore, the unified timestamp is written when the monitoring node completes sampling, rather than being added when data is received at the edge. The monitoring node prioritizes using the engineering master clock for time synchronization, which is maintained hierarchically by the dispatch center clock, the regional management station clock, and the node's local clock. When regional communication is normal, the monitoring node corrects its local sampling time according to the engineering master clock; when regional communication is interrupted, the monitoring node continues sampling using its local timekeeping clock and records the timekeeping status flag. After communication is restored, the edge aggregation and autonomous processing layer performs time drift correction on the data during the interruption period based on the clock deviation before and after the communication interruption, the node's timekeeping accuracy, and the continuity of the sampling sequence. The specific process of time drift correction is as follows: first, obtain the last valid time synchronization before the communication interruption and the first valid time synchronization after communication restoration; then calculate the cumulative deviation of the local clock relative to the engineering master clock between the two; subsequently, distribute this cumulative deviation according to the order of the sampling times during the interruption, so that the correction amount is smaller for data closer to the time before the interruption and larger for data closer to the time of communication restoration. The corrected data retains the original sampling time and the corrected unified timestamp for subsequent verification and traceability.

[0050] The engineering topology is pre-defined in the field sensing and communication network layer as a topology table. This table includes at least upstream objects, downstream objects, adjacent measuring points, water conveyance direction, control facilities, water distribution relationships, flood discharge relationships, tunnel connection relationships, and hydraulic propagation distance. When a monitoring node generates a data frame, it binds its own measuring point spatial code to the engineering objects in the topology table, ensuring that the same data frame simultaneously reflects both spatial location and hydraulic connectivity. For long-distance water conveyance projects with alternating canals and tunnels, the topology table also records the connection sequence between open channel sections, culvert sections, inverted siphon sections, tunnel sections, pumping station sections, and regulating reservoirs, avoiding the problem of continuous measuring point spatial codes but discontinuous hydraulic connectivity at the boundaries of adjacent administrative sections.

[0051] In one example implementation, the field sensing and communication networking layer completes quality status marking at the source end. The quality status marking is jointly determined by the device self-test status, calibration validity status, power supply status, installation status, communication status, and sampling status. The device self-test status characterizes whether the sensor has a disconnection, saturation, drift alarm, or internal fault; the calibration validity status characterizes whether the current sensor is within its calibration validity period; the power supply status characterizes whether the battery power, AC power, or solar power meets the sampling requirements; the installation status characterizes whether the device attitude, probe submersion conditions, probe contamination, or the stability of the mounting bracket are normal; the communication status characterizes whether the current link is in a normal, weak connection, interrupted recovery, or offline buffering state; and the sampling status characterizes whether the data frame is sampled at a regular frequency, encrypted sampling, down-frequency sampling, or event-triggered sampling. These markings are output along with the original monitoring values ​​to prevent subsequent models from misinterpreting data significantly affected by device status as actual changes in hydraulic engineering conditions.

[0052] Optionally, the sampling frequency is determined according to the basic sampling frequency and dynamic adjustment rules. The basic sampling frequency is jointly determined by the engineering operation and scheduling procedures, monitoring equipment technical documents, and data types. The basic sampling frequency for water level, flow rate, and pressure data is higher than that for cross-sectional sedimentation data, and the basic sampling frequency near sluice gates, pumping stations, and flood discharge facilities is higher than that for ordinary channel sections. During dynamic adjustment, the field sensing and communication network layer calculates the degree of change for water level changes, pressure fluctuations, sediment concentration changes, cross-sectional sedimentation changes, leakage changes, temperature changes, and link quality. The process for determining the degree of change is as follows: the current sampled value is compared with the previous sampled value to obtain the change amount of adjacent samples; then the current sampled value is compared with the stable benchmark value over the most recent period to obtain the deviation from the stable state; subsequently, the current degree of change is determined by combining the rate of change and the deviation of the data type. The stable benchmark value is the median of the data that passed the initial quality status inspection within the most recent 30 calendar days; when there is maintenance, shutdown, or equipment replacement, the data for the corresponding period is removed and the value is re-determined.

[0053] In addition, the thresholds for water level change, pressure fluctuation, sediment concentration change, cross-sectional siltation change, leakage change, and temperature change are stored separately for each measuring point. The water level change threshold is determined jointly based on the 95th percentile of the normal operating water level change rate over the most recent 30 calendar days at that measuring point, the allowable water level adjustment rate according to the dispatching regulations, and the allowable water level change rate in the hydraulic structure design; the most stringent value is taken as the execution threshold. The pressure fluctuation threshold is determined jointly based on the allowable pressure change value given in the tunnel or culvert structure safety evaluation document, the measurement uncertainty of the pressure sensor, and the 95th percentile of historical normal pressure fluctuations. The sediment concentration change threshold and turbidity change threshold are determined based on historical sediment concentration change records of inflow, upstream water source dispatching records, and sediment monitoring regulations. The cross-sectional siltation change threshold is determined based on the cross-sectional measurement accuracy, historical scouring and silting rates, and the safety margin of the flow capacity. The leakage change threshold is determined based on the allowable leakage rate per segment, the measurement uncertainty of the leakage monitoring device, and the safety evaluation requirements of the lining structure. The temperature change threshold is determined based on the low-temperature operation procedures, historical ice conditions, and the results of the temperature stress evaluation of the tunnel structure.

[0054] In some embodiments, when the degree of change in water level, pressure fluctuation, sediment content, cross-sectional siltation, leakage, or temperature reaches the corresponding encrypted sampling threshold, the field sensing and communication network layer increases the sampling frequency of the corresponding data type by one level; when the degree of change is below the frequency reduction threshold for multiple consecutive sampling periods, the sampling frequency of the corresponding data type decreases by one level. The encrypted sampling threshold is the more stringent of the 95th percentile of the historical distribution of normal operation or the engineering safety control value, and the frequency reduction threshold is the 50th percentile of the historical distribution of normal operation, combined with the minimum monitoring requirements of the equipment. The sampling frequency adjustment setting maintains a duration of at least 3 sampling periods after the sampling frequency is increased, and at least 5 consecutive sampling periods of stable conditions must be met before the sampling frequency is reduced, to prevent frequent frequency increases and decreases from causing data sequence instability.

[0055] In another specific implementation, link quality participates in sampling frequency adjustment but is not simply equivalent to frequency reduction. When link quality deteriorates and the monitored object is in a steady state, the field sensing and communication network layer reduces the sampling frequency of low-value data types while retaining window statistics results. When link quality deteriorates and there are simultaneously trends of water level exceeding limits, abnormal pressure fluctuations, increased leakage, low-temperature ice conditions, or flood discharge conditions, the field sensing and communication network layer maintains or increases the sampling frequency of safety-related data, while switching the data frame structure to short frame mode. Short frame mode retains the spatial coding of measurement points, unified timestamps, key measured values, sampling frequency, link status, and quality status markers, while removing unnecessary extended fields to prioritize the transmission of safety-related information under weak communication conditions.

[0056] As shown above, the multi-source raw monitoring data frames output by the field sensing and communication networking layer are not single sensor values, but source data units containing spatial, temporal, topological, frequency, link, and quality information. This data frame includes at least the measurement point spatial code, unified timestamp, engineering topology relationship identifier, data type, original measured value, calibration parameter version, current sampling frequency, link status, quality status marker, and sampling trigger reason. After receiving this data frame, the edge aggregation and autonomous processing layer can directly determine where the data comes from, which engineering object it corresponds to, what sampling mode it is in, whether there is link degradation, and whether it meets the basic reliable conditions for entering subsequent pre-processing.

[0057] As can be seen from the embodiments of this application, the field sensing and communication networking layer completes the spatial encoding of measurement points, unified timestamps, and binding of engineering topology relationships at the source end. This enables the monitoring data to have traceable spatiotemporal attributes as soon as it leaves the sensor, reducing data mismatch caused by manual processing or batch relabeling after long-distance transmission. For long-distance water transfer projects that cross administrative regions, hydraulic structures, and communication networks, this processing method can improve the accuracy of temporal alignment and spatial correlation at the edge.

[0058] Therefore, in this embodiment, the dynamic sampling frequency is determined based on the actual changes in water level, pressure, sediment content, cross-sectional siltation, leakage, temperature, and link quality. This reduces redundant sampling during stable water transport and retains critical data during sudden hydraulic disturbances, sediment changes, increased leakage, low-temperature ice conditions, or communication degradation. This approach reduces communication and storage loads while avoiding the loss of critical processes due to fixed low-frequency sampling during anomalies.

[0059] In this embodiment, the edge aggregation and autonomous processing layer acquires multi-source raw monitoring data frames; performs temporal alignment and spatial correlation based on unified timestamps, spatial coding of measuring points, and engineering topology relationships; extracts shallow temporal features and deep working condition features; and calculates the input value weight of each frame of monitoring data by combining communication link status, source end quality status marking, and cloud model prediction error; determines the data expression form, transmission priority, and input path based on the input value weight, and outputs at least one of differential data packets, simplified hazard feature packets, high-precision correction packets, water and sediment coupling data packets, parameter correction data packets, ice condition control data packets, or offline continuation data packets.

[0060] In this embodiment, after receiving the multi-source raw monitoring data frames output by the field sensing and communication networking layer, the edge aggregation and autonomous processing layer first establishes a cache queue according to the engineering objects. The cache queue uses the spatial code of the measurement point, the unified timestamp, and the engineering topology as indexes to group data from the same channel section, the same tunnel section, the same gate station, the same pumping station, the same reservoir, or the same water distribution node into the corresponding queue. For out-of-order data caused by communication delays, the edge aggregation and autonomous processing layer restores the sampling order based on the unified timestamp; for late data caused by offline resume transmission, the edge aggregation and autonomous processing layer does not overwrite the data already included in the model, but instead enters it as historical supplementary data into the offline resume transmission processing channel and retains the late mark.

[0061] In this process, temporal alignment primarily uses a unified timestamp, supplemented by sampling frequency and sampling trigger reasons. The edge convergence and autonomous processing layer first aligns the sampling times of different data types within the same engineering object to a unified time axis, and then generates aligned segments based on the differences in sampling frequencies of each data type. High-frequency data is not simply thinned to low-frequency data; instead, the maximum, minimum, average, trend, and abrupt change points are extracted between adjacent sampling times of low-frequency data as supplementary features for that time segment. Low-frequency data is not supplemented to the high-frequency time axis with fictitious values; instead, the most recent valid observation, observation age, and missing measurement status are retained, enabling subsequent model input value weights to distinguish between truly stable and low-frequency unobserved data. Furthermore, spatial association is completed based on the spatial coding of measuring points and the engineering topology. The edge convergence and autonomous processing layer establishes association windows for upstream measuring points, the current measuring point, and downstream measuring points in the same water conveyance direction. The time width of the association window is determined based on the hydraulic propagation time between adjacent measuring points. Hydraulic propagation time is estimated by combining the distance between adjacent measuring points, current flow rate, channel cross-sectional parameters, and pump / gate operating status. The estimation uses the most recently verified flow rate and cross-sectional parameters. For gate stations, pump stations, and water distribution nodes, spatial correlation also includes the relationship between equipment operation status and changes in water level, flow rate, and pressure. For example, after the gate opening increases, the downstream flow rate change should occur within the corresponding propagation time range; after the pump station starts, the outlet pressure and flow rate should enter a reasonable range of change according to the equipment characteristic curve.

[0062] Shallow temporal features are extracted directly from the aligned data sequence at the edge. For water level, flow rate, and pressure, the current value, changes in adjacent samples, rate of change, short-term average, short-term fluctuation, maximum short-term change amplitude, and trend direction are extracted. For sediment concentration and turbidity, the short-term increase amplitude, duration of continuous increase, peak location, and synchronization with flow rate changes are extracted. For cross-sectional sedimentation, the sedimentation increment in adjacent measurements, number of consecutive increments, location of local abrupt changes in the cross-section, and effective profile ratio are extracted. For seepage, the growth rate, duration of continuous growth, and lag relationship with upstream water level changes are extracted. For temperature-related data, the cooling rate, duration of low temperature, and degree of approach to the temperature of interest for ice conditions are extracted. For communication link status, delay changes, packet loss changes, number of retransmissions, and offline duration are extracted. These features do not change the original monitoring values ​​but serve as supplementary information to determine whether the current data has value for modeling.

[0063] In some embodiments, deep-level working condition features are extracted jointly by an edge-side working condition identification model and engineering rules. The edge-side working condition identification model reads data on water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, temperature, communication link status, and equipment action records within continuous time segments, outputting the confidence levels of working conditions such as steady-state water conveyance, gate regulation, pump station start-up and shutdown, water distribution adjustment, flood discharge, high-sediment-concentration inflow, suspected sedimentation, suspected leakage, low-temperature ice conditions, and tunnel structural anomalies. Engineering rules are used to correct the working condition identification results that are significantly affected by controlled actions. For example, short-term water level fluctuations after gate operation are not directly identified as abnormal water level changes, short-term pressure changes after pump station start-up and shutdown are not directly identified as tunnel structural anomalies, and when turbidity increases and sediment concentration increases in the same direction during high-sediment-concentration inflows, they are preferentially identified as water and sediment condition changes. Deep-level working condition features are formed jointly by the model identification results and the engineering rule correction results.

[0064] Optionally, the input value weight is formed by weighting shallow time-series change scores, deep operating condition scores, communication link scores, source-end quality scores, and cloud-based model prediction error scores. The shallow time-series change score is determined based on the rate of change, degree of fluctuation, trend direction, and magnitude of abrupt changes; the deep operating condition score is determined based on the confidence level of the operating condition identification results and the operating condition safety level; the communication link score is determined based on link latency, packet loss rate, retransmission count, and offline duration; the source-end quality score is determined based on the equipment self-test status, calibration validity status, power supply status, installation status, and sampling status; and the cloud-based model prediction error score is determined based on the deviation between the predicted value fed back from the cloud and the field calibration value. During calculation, the above scores are first converted to a unified score range, and then synthesized according to the weights corresponding to the engineering object and data type to obtain the input value weight for each frame of monitoring data.

[0065] In addition, each scoring threshold is determined according to data type and measurement point category. The rate of change threshold is the stricter of the 95th percentile of the rate of change of qualified data over the most recent 30 calendar days and the rate of change allowed by the scheduling regulations; the magnitude of abrupt change threshold is the stricter of the 95th percentile of the short-term magnitude of historical qualified data and the magnitude of change corresponding to the safety margin of hydraulic structures; the communication link degradation threshold is the 5th percentile of the normal link score over the most recent 30 calendar days, and shall not be lower than the minimum usable score specified in the communication operation and maintenance regulations; the source-end quality degradation threshold is determined by the equipment self-inspection specifications and calibration validity requirements; the cloud model prediction error threshold is obtained by the cloud digital twin platform layer from the model verification data, taking the 95th percentile of the relative prediction error of each measurement point. When the project is in the process of water diversion level switching, pump station operation mode switching, or low-temperature operation season, the cloud digital twin platform layer reissues the prediction error threshold, and the edge aggregation and autonomous processing layer calculates the input value weight according to the latest threshold.

[0066] Therefore, in this embodiment, the data representation is jointly determined by the input value weight, operating condition type, and communication status. When the monitoring data changes smoothly, the source quality is normal, the cloud model prediction error is small, and the link status is good, the edge aggregation and autonomous processing layer outputs differential data packets. The differential data packets retain the change amount, change direction, time stamp, and quality marker of the current value relative to the previous valid value, which is used to reduce the transmission volume during steady-state water conveyance. When safety-related data shows a tendency to exceed limits, the communication link is in a weak connection, or the reporting delay is large, the edge aggregation and autonomous processing layer outputs simplified hazard feature packets. The simplified hazard feature packets only retain the spatial code of the measurement point, unified timestamp, risk type, key measured values, change rate, degree of exceeding limits, and source quality status, so as to prioritize reporting under restricted link conditions.

[0067] In another specific implementation, when the prediction error of the cloud model continuously exceeds the prediction error threshold, and the source-end quality status is normal and the communication link is available, the edge aggregation and autonomous processing layer outputs a high-precision correction package. This high-precision correction package retains the original values, calibration values, shallow time-series characteristics, deep-seated operating condition characteristics, correlation results of adjacent measuring points, prediction errors, and equipment calibration parameter versions, used for parameter correction in the cloud model. When there are linked changes in sediment concentration, turbidity, flow rate, and cross-sectional sedimentation, the edge aggregation and autonomous processing layer outputs a water-sediment coupling data package. This water-sediment coupling data package retains sediment concentration, turbidity, flow rate, cross-sectional sedimentation, trend direction, effective profile ratio, and upstream-downstream correlations, used for joint updates of the water-sediment transport model and the dynamic scouring and sedimentation model.

[0068] It should be noted that in this application, when monitoring data does not exhibit sudden anomalies but continuously causes an increase in model prediction bias, the edge convergence and autonomous processing layer outputs a parameter correction data package. The parameter correction data package is used to represent slowly changing operating conditions, such as changes in channel roughness, segmented leakage parameters, cross-sectional geometry, or tunnel structural response parameters. The parameter correction data package retains statistical characteristics within continuous time segments, model bias trends, validated data ranges, and parameter mapping suggestions. For low-temperature operating conditions, when temperature-related data approaches the ice condition concern threshold and is accompanied by decreased flow, water level fluctuations, or pressure changes, the edge convergence and autonomous processing layer outputs an ice condition control data package. The ice condition control data package retains the rate of temperature decrease, duration of low temperature, water level fluctuations, flow rate changes, pressure changes, and ice condition risk markers.

[0069] In one optional implementation, when the communication link is interrupted or the edge node is in an offline buffering state, the edge aggregation and autonomous processing layer generates offline resume data packets. These offline resume data packets are numbered sequentially according to the sampling time, retaining the original data summary, key features, missing packet location, buffering start and end times, link recovery time, and quality status. After the link is restored, the offline resume data packets first enter the retransmission queue, avoiding contention for the highest transmission priority with real-time emergency data. The retransmission queue is sorted according to risk level, data time sensitivity, and model correction value; security-related retransmission data is prioritized over ordinary steady-state retransmission data, and data with high model correction value is prioritized over data used only for archiving.

[0070] As can be seen from the foregoing embodiments, the method of this application can determine the data upload order according to transmission priority. The simplified hazard characteristic packet and ice condition control data packet are given the highest priority; the high-precision correction packet, water-sediment coupling data packet, and parameter correction data packet are given the next highest priority; the differential data packet is given the normal priority; and offline resume data packets are uploaded in batches according to the retransmission strategy. The input path includes the real-time safety assessment path, the model parameter correction path, the operational status simulation path, the historical retransmission and archiving path, and the manual review path. Data that fails the source-end quality conditions but has anomaly indication significance is not directly discarded, but instead enters the manual review path or the suspected anomaly path, preventing underlying equipment failures and real engineering anomalies from being mixed up in the early stages.

[0071] The method described in this application enables temporal alignment, spatial correlation, feature extraction, input value weight calculation, data representation selection, and transmission priority allocation to be completed at the edge. This ensures that the data received by the cloud-based digital twin platform layer is no longer an indiscriminate raw data stream, but rather data packets matched to operating conditions, risks, model errors, and communication conditions. This processing method can reduce the pressure of full data transmission in long-distance water diversion projects and improve the timeliness of critical operating condition data entering the model. In the embodiments of this application, by outputting differential data packets, simplified hazard feature packets, high-precision correction packets, water-sediment coupling data packets, parameter correction data packets, ice condition control data packets, and offline continuation data packets, it can respectively serve steady-state compression, hazard reporting, model correction, water-sediment coupling analysis, slowly varying parameter correction, low-temperature ice condition control, and link breakage retransmission scenarios, enabling the same edge aggregation and autonomous processing layer to adapt to multiple operating states.

[0072] In this embodiment, the edge aggregation and autonomous processing layer acquires the adaptively reconstructed data packets; based on at least one of the following: timestamp continuity, data sequence number, missing packets, duplicate packets, integrity check code, range, rate of change, abrupt change amplitude, water conservation, water and sediment movement, scouring and silting geometry, leakage response, low temperature ice conditions, gate position-flow response, and upstream and downstream propagation causality, the data packets are subjected to communication integrity verification, numerical rationality verification, and multi-level physical constraint verification; high-confidence input data, suspected abnormal data, low-confidence data, or data to be retransmitted are output, and a correction data chain containing data source, spatiotemporal identifier, credibility, and model parameter mapping relationship is formed based on the high-confidence input data.

[0073] In this embodiment, the edge aggregation and autonomous processing layer first performs communication integrity verification on the adaptively reconstructed data packets. Communication integrity verification includes timestamp continuity verification, data sequence number verification, missing packet verification, re-packet verification, and integrity check code verification. Timestamp continuity verification determines whether the sampling times of data packets of the same data type at the same measurement point form a continuous sequence according to the sampling frequency; if the time interval between adjacent data packets exceeds the corresponding period of the current sampling frequency, and the data packets do not carry down-frequency sampling, offline buffering, or event-triggered sampling markers, then a time gap is determined. Data sequence number verification determines whether the data packet numbers are continuously increasing; a jump in number is marked as a missing packet, and repeated arrivals of the same number are marked as a re-packet. Integrity check code verification determines whether the content has been altered during transmission by regenerating a check code from the received data and comparing it with the check code carried in the data packet.

[0074] Furthermore, the thresholds and permissible conditions for communication integrity verification are determined based on the data packet type. Simplified hazard feature packets and ice condition control packets require complete key fields, including spatial coding of measurement points, unified timestamp, hazard type, key measured values, and rate of change; high-precision calibration packets require complete original values, calibration values, feature fields, quality status, and model prediction error fields; differential data packets require complete previous valid value index, current differential value, and quality status; offline resume data packets require complete cache start and end times, resume sequence number, and missing packet location. If a regular differential data packet has missing non-critical extended fields but complete core fields, it enters a pending supplementation state; if a hazard-type data packet has missing key fields, it enters a pending data transmission state and does not proceed to subsequent physical constraint verification.

[0075] Numerical rationality verification is performed after communication integrity verification passes. Numerical rationality verification includes range verification, rate of change verification, and abrupt change amplitude verification. Range verification compares data related to water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, and temperature with the allowable ranges for the corresponding measuring points. The allowable range is jointly determined by the sensor range, engineering design boundaries, operational scheduling boundaries, and equipment protection settings; when multiple boundaries exist simultaneously, the more stringent boundary related to operational safety is adopted. Rate of change verification determines whether the rate of change of the current data relative to the previous valid data exceeds the allowable rate of change. The allowable rate of change is jointly determined by the 99th percentile of the rate of change of qualified data over the most recent 30 calendar days and the maximum allowable rate of change in the scheduling regulations, with the more stringent value being adopted. Abrupt change amplitude verification determines whether short-term changes exceed the physical response capability of the engineering object; the abrupt change amplitude threshold is jointly determined by the hydraulic propagation time between measuring points, gate or pump station operation records, historical abrupt change samples, and equipment measurement uncertainty.

[0076] In one example implementation, the first level of multi-level physical constraint verification is intra-object constraint verification. The channel section data package needs to meet the following requirements: water level does not exceed the design control water level, flow rate does not exceed the cross-sectional water conveyance capacity, leakage does not exceed the allowable leakage of the section, and the cross-sectional siltation does not exhibit abnormal jumps that do not conform to the measurement cycle. The tunnel section data package needs to meet the following requirements: pressure does not exceed the allowable pressure, pressure change rate does not exceed the limits specified in the structural safety evaluation document, and there is a reasonable response relationship between leakage and pressure changes. The gate station data package needs to meet the response relationship between gate opening, upstream and downstream water level difference, and gate flow. The pump station data package needs to meet the operational relationship between pump start / stop status, inlet and outlet pressure, and flow rate. The water distribution node data package needs to meet the correspondence between the water distribution facility status, water distribution flow rate, and node water level.

[0077] In addition, the second level of the multi-level physical constraint verification is the adjacent object constraint verification. The edge convergence and autonomous processing layer checks the upstream and downstream propagation causality based on the engineering topology relationship between upstream measuring points, the current measuring point, and downstream measuring points. In specific processing, the timing of the upstream control action or hydraulic disturbance is first determined. Then, the hydraulic propagation time range is determined based on the distance between adjacent measuring points, the current flow rate, cross-sectional parameters, and operating status. Subsequently, it is determined whether the water level, flow rate, or pressure response of the downstream measuring point appears within this propagation time range. If the downstream response occurs earlier than the upstream disturbance, or if no response consistent with the engineering status appears after the propagation time range, the upstream and downstream propagation causality is deemed abnormal. The propagation time range is determined by the statistical results of hydraulic response times at the same flow rate level in historical qualified data, and is corrected in conjunction with the current flow rate level.

[0078] In another specific implementation, the water balance verification is used to determine whether there is a balance between upstream inflow, downstream outflow, water diversion, flood discharge, water storage changes, and seepage losses within the control section. The edge convergence and autonomous processing layer first reads the flow rate entering the control section from the upstream, and then reads the flow rate leaving the control section from the downstream. Subsequently, it determines the water storage change based on water level changes, cross-sectional geometric parameters, and reservoir or channel water storage curves. Then, it estimates the segmented seepage losses based on seepage monitoring values ​​or channel seepage models. Finally, it compares the above water items to obtain the water balance deviation. The allowable threshold for water balance deviation is jointly determined by the uncertainty of upstream flow measurement, downstream flow measurement, water storage calculation, water diversion measurement, and seepage estimation. The determination method is to first obtain the square value of each uncertainty, add them together, and then take the square root. When the water balance deviation exceeds the allowable threshold, the data packet is not used as high-confidence input data.

[0079] Optionally, water and sediment transport constraints are used to determine whether changes in sediment concentration, turbidity, flow rate, and cross-sectional sedimentation are consistent. Under high sediment load conditions, the marginal confluence and autonomous treatment layer are checked to see if an increase in sediment concentration is accompanied by an increase in turbidity, whether changes in flow rate correspond to changes in sediment transport capacity, and whether changes in downstream cross-sectional sedimentation lag behind changes in upstream sediment concentration. If sediment concentration increases significantly but turbidity remains unresponsive for a long period, or if cross-sectional sedimentation suddenly increases but upstream sediment concentration, flow rate, and measurement attitude do not support this change, then an abnormal water and sediment constraint is identified. The water and sediment response threshold is jointly determined by historical high sediment load events, cross-sectional measurement accuracy, the correlation between turbidity and sediment concentration, and the water transport flow rate level.

[0080] In some embodiments, scour-deposition geometric constraints are used to determine whether changes in cross-sectional sedimentation conform to the evolutionary pattern of cross-sectional morphology. The edge convergence and autonomous processing layer reads the number of effective profiles, sounding point locations, and local sedimentation heights from the cross-sectional measurement data, checking whether changes in cross-sectional sedimentation are concentrated in low-velocity regions and whether they match changes in upstream sediment concentration and velocity. For data showing large-scale cross-sectional uplift within a short period, if the number of effective profiles is insufficient, the probe orientation is abnormal, or upstream and downstream water and sediment data do not support this change, the data is marked as low-confidence data. The scour-deposition geometric threshold is determined by cross-sectional measurement accuracy, historical scour-deposition rates, the safety margin of cross-sectional water conveyance capacity, and inspection and re-measurement records.

[0081] It should be noted that in this application, leakage response constraints are used to determine the relationship between changes in leakage volume and water level, pressure, temperature, and structural condition. Increases in leakage volume in channel sections should correspond to increases in water level, changes in lining condition, or changes in segmental seepage conditions; increases in leakage volume in tunnel sections should correspond to changes in internal water pressure, temperature stress, or structural response. The edge convergence and autonomous treatment layer performs a lag comparison between changes in leakage volume and changes in upstream water level or tunnel pressure. If leakage volume suddenly increases but water level, pressure, and structural condition do not change accordingly, and the leakage monitoring equipment is in abnormal condition, low-confidence data is output; if leakage volume continues to increase and there is a reasonable lag relationship with changes in water level or pressure, suspected abnormal data or high-confidence input data is output. The specific category is determined jointly by the numerical boundary and physical constraint results.

[0082] In one optional implementation, cryogenic ice condition constraints are used to determine whether changes in temperature, water level, flow rate, and pressure conform to the cryogenic operation mechanism. The edge convergence and autonomous processing layer first determines whether the temperature-related data has reached the ice condition concern temperature, then determines whether the duration of the cryogenic temperature has reached the ice condition warning duration, and subsequently checks whether the flow rate has decreased, whether the water level has risen abnormally, and whether the pressure has fluctuated irregularly. The ice condition concern temperature is determined by the cryogenic operation procedures and historical ice condition records, and the ice condition warning duration is determined by the statistical results of the continuous cryogenic duration before the formation of historical ice conditions. If the temperature has reached the concern condition and the water level, flow rate, or pressure has changed accordingly, the ice condition control data packet passes the cryogenic ice condition constraint; if only a single temperature sensor abnormally decreases and there is no similar trend among adjacent temperature measuring points, it is marked as a suspected sensor anomaly.

[0083] Furthermore, the gate position-flow response constraint is used to determine whether changes in gate opening and flow rate match. The edge aggregation and autonomous processing layer reads the current gate opening, the previous opening, the upstream and downstream water level difference, and the flow rate. Based on the historical calibration curve of the gate station or the gate position-flow relationship confirmed by the operation and management unit, it determines whether the flow rate after the opening change is within the allowable response range. The allowable response range is jointly determined by the gate calibration error, the flowmeter measurement uncertainty, and the water level measurement uncertainty. If the gate opening increases but the flow rate does not increase reasonably, or if the gate does not move but the flow rate changes abruptly, the data packet enters the suspected abnormal data set and triggers a check of the gate status and flowmeter status.

[0084] Therefore, in this embodiment, the edge aggregation and autonomous processing layer outputs four types of data based on the verification results. Data with complete communication, reasonable values, and passing multi-level physical constraint verification is output as high-confidence input data. Data with complete communication and values ​​showing a tendency to exceed limits, but whose physical relationships can support this change, is output as suspected abnormal data, used by the cloud-based early warning model and for manual review. Data with complete communication but poor source quality and mismatched values ​​and physical relationships is output as low-confidence data and is not used for model parameter inversion. Data with missing packets, duplicate packets, inconsistent checksums, missing key fields, or incomplete offline caching is output as data to be retransmitted, awaiting link recovery or retransmission from the source.

[0085] As a preferred implementation, to address the issue of conflicting but within-limit data processing from multiple sensors, a conflict evidence separation mechanism is implemented at the edge aggregation and autonomous processing layers. For example, if upstream flow data, downstream flow data, and water level changes are all within acceptable limits, but the water conservation verification fails for an extended period, instead of directly classifying all data as low-confidence, separate flow rate deviation chains, water storage deviation chains, and leakage deviation chains are generated. The cloud-based digital twin platform layer continuously determines the source of the deviation based on subsequent data. This implementation can resolve the problem that slight instrument drift, while not exceeding limits in the early stages, can still affect parameter inversion.

[0086] In another specific implementation, to address the issue of insignificant hydraulic response in the early stages of low-temperature ice conditions, the edge convergence and autonomous processing layer combine temperature drop, flow rate reduction, and minor pressure fluctuations into a unified constraint. Even if the water level has not yet exceeded the limit, when the duration of low temperature reaches the ice condition warning duration and the frequency of pressure fluctuations exceeds the 95th percentile of the normal fluctuations before low temperature in the last 30 calendar days, the ice condition control data packet is output as suspected abnormal data and entered into the cloud-based low-temperature ice condition risk assessment model. This embodiment can identify weak signals in the underlying ice condition data in advance, avoiding reliance solely on water level exceeding the limit to trigger warnings.

[0087] Furthermore, for offline data retransmission caused by intermittent cross-regional link disruptions, the edge aggregation and autonomous processing layer performs a retransmission consistency check after retransmission. This consistency check does not require the retransmitted data to alter the completed real-time scheduling conclusions; instead, it checks whether the retransmitted data supports the model prediction errors and control actions at the time of retransmission. If there are significant differences between the retransmitted data and the real-time data, a historical model verification marker is generated for subsequent model parameter replay training. This embodiment can solve the problem that offline retransmitted data arrives after real-time control but still has model correction value.

[0088] In this embodiment, the cloud-based digital twin platform layer acquires highly reliable input model data and calibration data chains; based on data packet type, spatial coding of measurement points, unified timestamp, physical constraint verification results, and data reliability, it establishes parameter mapping relationships between the highly reliable input model data and hydrodynamic models, water and sediment transport models, dynamic scouring and silting models, channel leakage models, and tunnel structure safety assessment models, and performs layered online parameter inversion updates and coupling consistency verification; it outputs simulation results of engineering operation status, flood risk warning results, high sediment scouring and silting evolution prediction results, leakage loss prediction results, low temperature ice risk assessment results, tunnel structure safety assessment results, or scheduling and control strategies.

[0089] In this embodiment, after receiving the high-reliability input model data and the calibration data chain, the cloud-based digital twin platform layer first identifies the data purpose based on the data packet type. Differential data packets are mainly used for continuous updates of operational status; simplified hazard feature packets are mainly used for flood risk early warning, low-temperature ice condition risk assessment, and rapid assessment of tunnel structural safety; high-precision calibration packets are mainly used for hydrodynamic model parameter correction; water-sediment coupling data packets are mainly used for updating water-sediment transport models and dynamic scour and deposition models; parameter calibration data packets are mainly used for slow-varying parameter inversion; ice condition control data packets are mainly used for low-temperature ice condition risk assessment and scheduling control strategy generation; and offline continuation data packets are mainly used for historical playback, model verification, and parameter supplementation and calibration. The cloud-based digital twin platform layer does not indiscriminately mix different data packets, but instead establishes corresponding processing channels based on the data packet source and verification results. Measurement point spatial encoding is used to map the high-reliability input model data to specific engineering objects and model calculation units. Channel water level, channel flow, cross-sectional siltation, sediment content, turbidity, and seepage are mapped to the corresponding channel calculation cross-sections; tunnel pressure, temperature-related data, seepage, vibration, and lining strain are mapped to the corresponding tunnel structural segments; upstream and downstream water levels, gate opening, and flow through gates at sluice gates are mapped to the sluice gate control unit; inlet and outlet pressure, flow, and pump status at pumping stations are mapped to the pumping station control unit; inlet flow, outlet flow, and reservoir water level at reservoirs are mapped to the reservoir regulation unit; and water level, flow, and facility status at water distribution nodes are mapped to the water distribution control unit. A unified timestamp is used to align different models within the same calculation cycle, and physical constraint verification results and data reliability are used to determine the degree of data participation in parameter inversion.

[0090] Furthermore, the parameter mapping relationships are generated by the calibration data chain. For the hydrodynamic model, high-confidence water level, flow rate, gate opening, pump station status, and upstream / downstream water level difference are mapped to channel roughness, local head loss coefficient, boundary flow correction, and gate station flow parameters. For the water and sediment transport model, high-confidence sediment concentration, turbidity, flow rate, and upstream inflow conditions are mapped to sediment deposition parameters, sediment transport capacity parameters, and sediment-laden boundary conditions. For the dynamic scour and deposition model, high-confidence cross-sectional sedimentation, effective cross-sectional profile ratio, sediment concentration, and flow velocity variation are mapped to scour coefficient, sedimentation coefficient, and cross-sectional geometric correction parameters. For the channel leakage model, high-confidence leakage, water level, segmented lining status, and water balance deviation are mapped to segmented permeability coefficient and lining leakage parameters. For the tunnel structure safety assessment model, high-confidence pressure, temperature-related data, leakage, vibration, and lining strain are mapped to pressure load parameters, temperature influence parameters, structural response parameters, and safety evaluation parameters.

[0091] In one example implementation, the hierarchical online parameter inversion update is performed in the order of single-point correction, section correction, and full-line correction. Single-point correction addresses measurement deviations and local responses at individual measuring points; section correction addresses continuous deviations in water level, flow rate, pressure, sediment concentration, siltation, or leakage between adjacent measuring points; and full-line correction addresses changes in water conveyance capacity, inflow conditions, scheduling boundaries, and seasonal operational variations across multiple engineering objects. The cloud-based digital twin platform first uses single-point correction to eliminate local deviations, then uses section correction to adjust segmented parameters, and finally uses full-line correction to check whether the parameter updates conform to the overall operational patterns of the water transfer project. This sequence avoids undue influence of a single abnormal measuring point on the overall parameters.

[0092] Furthermore, the calculation process for online parameter inversion and updating aims to reduce the deviation between measured data and model calculation results. The cloud-based digital twin platform first runs simulations using the model parameters from the previous calculation cycle to obtain predicted water level, predicted flow rate, predicted sediment concentration, predicted siltation, predicted leakage, or predicted structural response for each measuring point. Then, the predicted results are compared item by item with the high-confidence input model data to obtain the deviation for each data type. Next, the weight of each deviation in the parameter update is determined based on data confidence, sensor measurement uncertainty, and physical constraint verification results. Finally, new parameter combinations are searched within the parameter range allowed by the design documents, operating procedures, material performance documents, and safety evaluation reports to reduce the overall weighted deviation while ensuring that parameter changes between adjacent calculation cycles do not exceed the parameter continuity threshold. The parameter continuity threshold is determined by historical calibration results, seasonal variation amplitude, and engineering operation management requirements.

[0093] Optionally, after the hydrodynamic model parameters are updated, the cloud-based digital twin platform immediately performs a coupling consistency check on the water and sediment transport model, the dynamic scouring and silting model, the channel seepage model, and the tunnel structural safety assessment model. The flow rate, water level, and velocity output by the hydrodynamic model serve as inputs to the water and sediment transport model; the sediment transport trend output by the water and sediment transport model serves as input to the dynamic scouring and silting model; the cross-sectional changes output by the dynamic scouring and silting model, in turn, affect the flow cross-section of the hydrodynamic model; the seepage loss output by the channel seepage model affects the water balance of the hydrodynamic model; and the tunnel structural safety assessment model determines the structural safety level based on pressure, temperature, and seepage status. If an update to one model causes the calculation results of other models to deviate from the data range that has passed the physical constraint check, the cloud-based digital twin platform backtracks and corrects the data chain, identifying the measurement points, data packet types, and parameter items that caused the inconsistency.

[0094] In some embodiments, the coupling consistency verification threshold is determined according to the model type. The consistency threshold between hydrodynamics and water balance is jointly determined by the uncertainty of flow measurement, the uncertainty of water storage calculation, and the uncertainty of leakage estimation; the consistency threshold between water and sediment transport and dynamic scouring and deposition is determined by the uncertainty of sediment concentration measurement, turbidity conversion error, cross-sectional measurement error, and statistical results of historical high sediment concentration processes; the consistency threshold between the leakage model and the hydrodynamic model is determined by the segmented leakage monitoring error and the normal range of water balance residuals; the consistency threshold between the tunnel structural safety assessment model and pressure, temperature, leakage, vibration, and lining strain is determined by the structural safety assessment document, sensor measurement uncertainty, and historical stable operating range. When the deviation between models exceeds the corresponding consistency threshold, the cloud-based digital twin platform layer does not output the final scheduling control strategy, but instead first performs parameter rollback, reduces the weight of suspicious data, or requests the edge side to supplement a high-precision correction package.

[0095] Therefore, in this embodiment, the simulation results of the engineering operation status are generated by the model after parameter updates and coupling consistency verification. The simulation results include water level distribution along the route, flow distribution, pressure distribution, pumping station operation status, sluice gate flow status, reservoir regulation status, water distribution node allocation status, cross-sectional scouring and silting changes, and seepage losses. Flood risk warning results are determined based on the degree to which the water level at the control section approaches the control level, the upstream inflow growth rate, the status of flood discharge facilities, the downstream water level tolerance, and the water level rise trend during the forecast period. The risk threshold is determined by the approved flood control scheduling plan, the engineering control water level, and the safety requirements of downstream protected objects.

[0096] In another specific implementation, the prediction results of high sediment load erosion and deposition evolution are jointly generated by the water and sediment transport model and the dynamic erosion and deposition model. The cloud-based digital twin platform first determines the current sediment transport capacity based on high-confidence sediment load, turbidity, and flow rate, then determines the location of deposition development based on the cross-sectional deposition volume and historical erosion and deposition processes, and subsequently predicts the cross-sections that may experience increased deposition during subsequent scheduling cycles. The high sediment load risk threshold is jointly determined by historical high sediment load inflow processes, the safety margin of cross-sectional water transport capacity, sediment monitoring procedures, and dredging control requirements determined by the operation and management unit. The leakage loss prediction results are generated by the channel leakage model in combination with water level, segmented lining status, leakage monitoring values, and water balance deviation. The leakage risk threshold is determined by the segmented allowable leakage volume, lining safety evaluation results, and historical leakage change records.

[0097] It should be noted that in this application, the low-temperature ice condition risk assessment results are determined jointly by temperature-related data, water level, flow rate, pressure, and historical ice condition samples. The cloud-based digital twin platform first determines whether the temperature meets the ice condition concern conditions, and then determines whether the duration of low temperature, the degree of flow rate reduction, abnormal water level rise, and pressure fluctuations occur simultaneously. The low-temperature ice condition risk threshold is determined by the low-temperature operation procedures, historical ice condition formation time, and flow conditions of the engineering object. The tunnel structure safety assessment results are output by the tunnel structure safety assessment model, with input data including pressure, pressure change rate, temperature change, leakage, vibration, and lining strain. The safety level threshold is determined by the tunnel structure safety evaluation report, design allowable pressure, material performance parameters, and the stable operating range of monitoring sensors.

[0098] As can be seen from the foregoing embodiments, the method of this application can generate a scheduling control strategy when the simulation results of the engineering operation status, the flood risk warning results, the high sediment load erosion and siltation evolution prediction results, the leakage loss prediction results, the low temperature ice risk assessment results, or the tunnel structure safety assessment results indicate that there is a scheduling need. The scheduling control strategy includes adjusting the gate opening, starting and stopping or adjusting the speed of pumping stations, opening and closing flood discharge facilities, adjusting the water distribution of water diversion facilities, adjusting the flow limit of flow restriction facilities, and the action of pressure reducing and diversion facilities. When the strategy is generated, the cloud-based digital twin platform layer simultaneously outputs the strategy type, strategy priority, model credibility, involved engineering objects, expected impact range, and safety verification conditions for subsequent hierarchical execution by the scheduling control execution layer.

[0099] As can be seen, the embodiments of this application, through hierarchical online parameter inversion update and coupled consistency verification, can reduce the impact of single measurement point anomalies, single model deviations or single physical process distortions on scheduling strategies, and improve the credibility of cloud simulation results, risk assessment results and scheduling control strategies.

[0100] In this embodiment, the scheduling control execution layer obtains the scheduling control strategy output by the cloud digital twin platform layer; generates hierarchical execution instructions based on strategy type, strategy priority, model credibility, working condition risk level and on-site execution constraints, and performs safety constraint verification based on gate opening change rate, pump station start-up and shutdown frequency, flood discharge flow rate increase, upstream and downstream water level difference, tunnel pressure change rate, seepage pressure, vibration and lining strain before execution; outputs phased control actions for gates, pump stations, flood discharge facilities, water diversion facilities, flow restriction facilities or pressure reduction and diversion facilities, and feeds back the action feedback and engineering response to the edge aggregation and autonomous processing layer or the cloud digital twin platform layer.

[0101] In this embodiment, after receiving the scheduling and control strategy output by the cloud-based digital twin platform layer, the scheduling and control execution layer first reads the strategy type, strategy priority, model credibility, operational risk level, and on-site execution constraints. Strategy types include routine water supply scheduling, flood risk management, high sediment load erosion control, seepage loss control, low-temperature ice control, and tunnel structure safety control. Strategy priority is determined by the cloud-based digital twin platform layer based on the risk object, risk level, impact range, and expected occurrence time. Model credibility is determined by the number of models involved in generating the strategy, the model coupling consistency verification results, the proportion of high-credibility input data, and the integrity of the correction data chain. On-site execution constraints include the current gate opening degree, pump group availability status, flood discharge facility opening and closing conditions, current water distribution facility allocation, flow restriction facility location, pressure reducing and diversion facility status, power supply conditions, communication status, and on-site maintenance interlock status.

[0102] The tiered execution instructions are determined based on the strategy risk level and model credibility. When the operational risk level is high and the model credibility reaches the execution threshold, the scheduling control execution layer generates a direct execution instruction and initiates a safety constraint verification. When the operational risk level is medium or the model credibility is within the review range, the scheduling control execution layer generates a phased execution instruction, first executing small-amplitude control actions, and then deciding whether to proceed to the next stage based on the on-site response. When the operational risk level is low and the model credibility is insufficient, the scheduling control execution layer generates a suggested execution instruction, which is submitted to the operators for confirmation before execution. The execution threshold is determined by the operation management unit based on historical scheduling playback, model validation results, and safety management requirements; the review range is the range below the execution threshold but above the minimum available credibility, which is jointly determined by the upper limit of model validation error and the engineering risk tolerance.

[0103] Furthermore, before generating gate control commands, the scheduling control execution layer performs safety constraint checks on the gate opening change rate. During the check, the current gate opening, target opening, permissible operating speed, upstream and downstream water level difference, and the status of the gate opening / closing equipment are first read. Then, the required opening change and operating time from the current opening to the target opening are calculated. Subsequently, it is determined whether the change exceeds the permissible opening change value of the equipment in a single operation, whether the operating speed exceeds the permissible speed of the gate hoist, and whether the upstream and downstream water level difference after the operation exceeds the allowable range of the gate structure and the downstream channel. The gate opening change rate threshold is determined by the gate hoist technical documents, gate structural safety evaluation, downstream water level control requirements, and historical safety operation records, with the most stringent limit being selected.

[0104] In one example implementation, before generating pump station control commands, the scheduling control execution layer verifies the pump station start-up and shutdown frequency, target pump speed, inlet and outlet water pressure, and operating flow rate. The pump station start-up and shutdown frequency threshold is jointly determined by the pump equipment manual, motor protection procedures, inverter operating requirements, and operation management procedures. The scheduling control execution layer first checks whether the interval between the target pump and the last start-up and shutdown meets the equipment protection requirements, then checks whether the target speed is within the allowable speed range, and subsequently checks whether the changes in inlet and outlet water pressure after starting or stopping the pump exceed the allowable pressure change range. If the pump station start-up and shutdown frequency does not meet the threshold requirements, the scheduling control execution layer does not directly execute the start-up and shutdown actions, but instead generates alternative execution schemes, such as adjusting the speed of the already running pump, delaying the start-up and shutdown, or coordinating with minor adjustments to the gate.

[0105] In addition, before generating control commands for the flood discharge facilities, the dispatch control execution layer verifies the flood discharge flow rate growth rate and the downstream water level tolerance. The flood discharge flow rate growth rate threshold is determined by the flood control dispatch plan, the allowable water level rise rate of the downstream river or channel, the opening and closing capacity of the flood discharge facilities, and the safety requirements of the downstream protected objects. The dispatch control execution layer first determines the flood discharge flow rate growth process based on the target opening range, and then, in conjunction with the downstream water level, the flow capacity of the downstream control section, and the dispatch actions already executed, judges whether the flood discharge may cause the downstream water level to exceed the limit. If the flood discharge flow rate growth rate exceeds the threshold, a single flood discharge action is divided into multiple stages. After each stage is executed, feedback on the on-site water level and flow rate is awaited, and the next stage is executed only after confirming that the downstream response meets the safety requirements.

[0106] Optionally, before implementing tunnel-related control strategies, the scheduling control execution layer performs safety constraint checks on the upstream and downstream water level difference, tunnel pressure change rate, seepage pressure, vibration, and lining strain. The upstream and downstream water level difference threshold is determined by the tunnel's hydraulic design conditions and operating procedures; the tunnel pressure change rate threshold is determined by the tunnel structural safety evaluation report, lining material performance, and historical pressure response records; the seepage pressure threshold is determined by the lining anti-seepage design, drainage system capacity, and the measurement range of the monitoring device; the vibration threshold is determined by the structural safety monitoring procedures and historical stable operating vibration range; and the lining strain threshold is determined by the structural safety evaluation report and the material's allowable strain. The scheduling control execution layer compares the current monitored values ​​with the predicted changes after the action. If any key indicator exceeds its threshold, the control action must not be executed directly; the action amplitude must be reduced or a pressure reduction and diversion strategy must be switched.

[0107] In some embodiments, the phased control actions are executed in three stages: pre-adjustment, master adjustment, and steady-state confirmation. The pre-adjustment stage is used to observe the field response with small-scale actions, such as slight opening of gates, slight adjustment of pump station speed, or small-scale opening of pressure-reducing and diversion facilities. The master adjustment stage is used to execute the main control variables after the pre-adjustment response meets safety constraints, causing water level, flow rate, pressure, water diversion, or risk indicators to converge towards the target range. The steady-state confirmation stage is used to maintain the execution results and continue to collect field feedback, confirming that water level fluctuations, pressure changes, leakage response, vibration, and lining strain do not exceed limits. After each stage, the dispatch control execution layer reads the data collected by the field perception and communication network layer and compares the action results with the cloud-based prediction results.

[0108] It should be noted that, in this application, the interval between phased control actions is determined by the response time of the engineering object. The interval after gate operation is determined by the upstream and downstream water level propagation time and the gate's flow response time; the interval after pump station operation is determined by the pressure stabilization time and flow stabilization time; the interval after flood discharge facility operation is determined by the water level response time at the downstream control section; the interval after water diversion facility operation is determined by the flow stabilization time at the water diversion node; and the interval after flow restriction and pressure-reducing diversion facilities operation is determined by the pressure release time and the tunnel structure response time. The above response times are jointly determined by historical dispatch records, engineering hydraulic calculation results, and on-site commissioning test records.

[0109] In another specific implementation, the action feedback includes command reception status, action start time, action completion time, actual opening degree, actual rotation speed, actual flow rate, actual water diversion volume, actual flood discharge volume, flow restriction position, diversion ratio, equipment alarm, and lockout status. Engineering response feedback includes post-action data such as water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status, temperature-related data, seepage pressure, vibration, and lining strain. The scheduling and control execution layer sends the action feedback to the cloud-based digital twin platform layer to verify whether the scheduling and control strategy is executed as expected; it also sends the engineering response feedback to the edge aggregation and autonomous processing layer for the next round of input value weight calculation and data packet reconstruction.

[0110] As can be seen from the foregoing embodiments, the method of this application can transform the scheduling control strategy into executable, verifiable, and feedback-enabled hierarchical execution instructions. Gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, and pressure-reducing diversion facilities no longer receive only a single target value, but instead receive control instructions that include the action amplitude, action sequence, action speed, safety threshold, feedback requirements, and interlocking conditions. This processing method can reduce the semantic differences between the model strategy and the field execution equipment, avoiding instability in field execution due to overly coarse control instructions.

[0111] As can be seen, in this embodiment, by executing instructions in a hierarchical manner and controlling actions in stages, the strategy output by the cloud-based digital twin platform layer can be adapted to the capabilities of on-site equipment, the level of operational risk, and the reliability of the model. After action feedback and engineering response feedback enter the edge and cloud, they can form the basis for the next round of model correction and scheduling optimization, improving the reliability of the closed-loop operation of long-distance water transfer projects.

[0112] Table 1. Performance Comparison Results of the Invention and Existing Technologies As can be seen from the table data, the system of the present invention has significant improvements over the existing technology system.

[0113] First, the daily average data upload volume of the system of this invention decreased from 100.0 GB to 42.6 GB, indicating that by calculating the input model value weight and adaptively reconstructing data packets, low-value redundant data transmission can be reduced, thus lowering communication and storage pressure. The proportion of high-confidence input model data increased from 78.9% to 94.3%, and the recall rate of high-value data increased from 81.5% to 96.8%, demonstrating that this invention can effectively screen out low-confidence data before data input and prioritize the retention of data valuable for model correction and risk identification. The average prediction error of the model decreased from 14.5% to 6.3%, indicating that this invention can improve the fitting ability of the digital twin model to the actual engineering operation status through high-confidence input model data and online parameter inversion updates. The availability rate of weak network data increased from 76.4% to 93.7%, indicating that this invention is applicable to common scenarios in cross-regional long-distance water transfer projects, such as weak networks, multi-hop transmission, and link breakage retransmission. The average early warning capability for risks increased from 1.0 times to 2.8 times, and the number of scheduling execution anomalies decreased by 68.2%. This indicates that the present invention can identify flood, siltation, leakage, ice conditions and structural safety risks earlier, and improve the safety of scheduling control through pre-execution safety constraint verification.

[0114] This invention also provides a method for constructing a digital twin system for cross-regional long-distance water transfer projects, including: S1. Collect water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status and temperature-related data from channels, tunnels, sluice gates, pumping stations, reservoirs and water distribution nodes, and perform source-end spatiotemporal calibration and quality status marking on the data to form multi-source raw monitoring data frames. S2. Based on the multi-source original monitoring data frames, extract shallow time-series features and deep operating condition features. According to the shallow time-series features, deep operating condition features, communication link status, and model prediction error fed back by the cloud digital twin platform, calculate the input value weight of each frame of monitoring data. Based on the input value weight, adaptively reconstruct each frame of monitoring data into a corresponding type of data packet, and perform communication integrity verification, numerical rationality verification, and multi-level physical constraint verification on the data packets. Mark the data packets that pass the verification as high-confidence input data. S3. Based on the high-reliability input data, establish a mapping relationship between the high-reliability input data and the model parameters to be updated, and perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model. S4. Generate simulation results of engineering operation status or scheduling control strategy based on the updated model, and perform safety constraint verification before execution. Execute control actions or limit, delay, downgrade, execute in stages or roll back the scheduling control strategy according to the verification results. Use action feedback and engineering response feedback for model parameter correction, updating of input value weights and subsequent rolling generation of scheduling control strategies.

[0115] In this embodiment, step 1 is executed by the field sensing and communication networking layer. Water level monitoring equipment, flow rate monitoring equipment, sediment concentration monitoring equipment, turbidity monitoring equipment, cross-sectional siltation monitoring equipment, leakage monitoring equipment, and temperature monitoring equipment are deployed in the channel section; pressure monitoring equipment, leakage monitoring equipment, temperature monitoring equipment, vibration monitoring equipment, lining strain monitoring equipment, and structural temperature monitoring equipment are deployed in the tunnel section; water level, flow rate, pressure, opening degree, rotational speed, start / stop status, flood discharge status, water diversion status, and communication link detection equipment related to their operation and control are deployed at sluice gates, pumping stations, reservoirs, and water diversion nodes. Each monitoring device generates an original monitoring record during sampling. The original monitoring record includes the measuring point number, engineering object number, sampling time, data type, original sampled value, equipment status, sensor range, calibration parameter version, power supply status, and communication link status.

[0116] In this embodiment, source-end spatiotemporal calibration is completed before the data leaves the monitoring node. Each monitoring node is pre-written with a spatial code for the measuring point. This spatial code is generated hierarchically according to the water diversion project line number, management section number, project object number, control section number, installation location, and equipment serial number. The spatial code for channel measuring points includes the channel station number, bank type, section location, and installation elevation; the spatial code for tunnel measuring points includes the tunnel number, tunnel mileage, lining zone, and monitoring section; the spatial code for sluice gates, pumping stations, reservoirs, and water diversion nodes includes the structure number, equipment unit number, and upstream / downstream location. Through this spatial code, subsequent edge processing and cloud modeling can directly identify the entity objects, control sections, and hydraulic connection locations corresponding to the data.

[0117] The unified timestamp is jointly guaranteed by the project master clock, the regional management station clock, and the local clock of the monitoring nodes. When communication is normal, the monitoring nodes write the sampling time according to the project master clock; when communication is interrupted, the monitoring nodes continue to write the sampling time using their local timekeeping clock and add a timekeeping status marker to the data frame. After communication is restored, the edge aggregation and autonomous processing layer corrects the sampling time during the interruption period based on the last valid time synchronization before the communication interruption, the first valid time synchronization after communication restoration, the local timekeeping clock drift record, and the data sequence number during the interruption. During correction, the total offset accumulated by the local clock during the interruption is first determined, and then the offset is distributed according to the position of the sampling time during the interruption, so that the data correction amplitude is smaller near the start of the interruption and larger near the time of communication restoration. The corrected data retains both the original sampling time and the unified timestamp to meet subsequent traceability needs.

[0118] Furthermore, the quality status markers are jointly determined by the equipment self-test status, calibration validity status, power supply status, installation status, sampling status, and communication status. The equipment self-test status indicates whether the sensor has any open circuits, saturation, drift alarms, or internal faults; the calibration validity status indicates whether the sensor is within its calibration validity period; the power supply status indicates whether the mains power, battery, or solar power supply meets the sampling requirements; the installation status indicates whether the equipment attitude, probe submersion conditions, support stability, probe contamination, or echo quality meet the monitoring requirements; the sampling status indicates whether the data is from regular sampling, encrypted sampling, down-frequency sampling, event-triggered sampling, or offline buffered sampling; and the communication status indicates whether the link is normal, degraded, recovered from an interruption, or resumed offline transmission. These markers, along with the original monitoring values, form a multi-source original monitoring data frame, preventing subsequent models from misinterpreting changes in equipment status as changes in the actual engineering status.

[0119] In one example implementation, the multi-source raw monitoring data frames also carry engineering topology relationship identifiers. These identifiers are generated from a pre-set topology table, which records upstream objects, downstream objects, adjacent measuring points, water conveyance direction, gate control relationships, pumping station pressurization relationships, reservoir regulation relationships, water distribution relationships, flood discharge relationships, and tunnel connection relationships. When generating data frames, the field sensing and communication networking layer binds the spatial codes of the measuring points to the engineering objects in the topology table, ensuring that data related to water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional siltation, leakage, communication link status, and temperature have spatial, temporal, and topological attribution at the source end.

[0120] In this embodiment, step 2 is performed by the edge aggregation and autonomous processing layer. After receiving the multi-source raw monitoring data frames, the edge aggregation and autonomous processing layer first establishes a buffer queue according to the engineering object, and classifies the data of the same channel section, the same tunnel section, the same gate station, the same pumping station, the same reservoir, or the same water distribution node into the corresponding queue. For out-of-order data caused by link delay, the edge aggregation and autonomous processing layer restores the sampling order according to the unified timestamp; for data retransmitted after communication is restored, the edge aggregation and autonomous processing layer classifies it into the historical retransmission queue according to the original sampling time and the continuation sequence number, without overwriting the data already used for real-time scheduling.

[0121] Furthermore, temporal alignment is completed using a unified timestamp, sampling frequency, and sampling triggering reason. For data with different sampling frequencies within the same engineering object, the edge aggregation and autonomous processing layer does not fabricate low-frequency data as high-frequency data, but instead retains the most recent valid observation, observation duration, and missing measurement status of the low-frequency data. For high-frequency data, the edge aggregation and autonomous processing layer extracts the maximum, minimum, average, trend, and abrupt change points between adjacent low-frequency data as supplementary features for the same time segment. Spatial association is completed according to the spatial coding of measurement points and the engineering topology. Upstream measurement points, the current measurement point, and downstream measurement points are placed in the same association window. The time range of the association window is jointly determined by the distance between adjacent measurement points, current flow rate, cross-sectional parameters, gate status, pump station status, and historical propagation time.

[0122] The shallow temporal features are extracted from the aligned data sequence. For water level, flow rate, and pressure, the current value, changes in adjacent samples, rate of change per unit time, short-term average level, short-term fluctuation degree, short-term maximum change amplitude, and trend direction are extracted. For sediment concentration and turbidity, the short-term increase amplitude, duration of continuous increase, peak position, and synchronization with flow rate changes are extracted. For cross-sectional sedimentation, the sedimentation increment in adjacent measurements, number of consecutive increments, location of local abrupt changes, and effective profile ratio are extracted. For seepage, the growth rate, duration of continuous growth, and lag relationship relative to upstream water level or tunnel pressure are extracted. For temperature-related data, the cooling rate, duration of low temperature, and degree of approach to the temperature of concern for ice conditions are extracted. For communication link status, delay changes, packet loss changes, number of retransmissions, and offline duration are extracted. All of the above features are calculated by the edge aggregation and autonomous processing layer based on measured data and time series.

[0123] Furthermore, the deep-level operating condition characteristics are jointly generated by the edge-side operating condition identification model and engineering rules. The edge-side operating condition identification model reads water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional siltation, leakage, communication link status, temperature-related data, and equipment action records within continuous time segments to identify operating conditions such as steady-state water conveyance, gate regulation, pump station start-up and shutdown, water distribution adjustment, flood discharge, high-sediment-concentration inflow, suspected siltation, suspected leakage, low-temperature ice conditions, and tunnel structural anomalies, and outputs the corresponding confidence levels. Engineering rules are used to correct identification results affected by known control actions. For example, short-term water level fluctuations after gate opening changes are not directly treated as abnormal water level changes, short-term pressure changes after pump station start-up and shutdown are not directly treated as tunnel structural anomalies, and when turbidity increases and sediment concentration increases in the same direction during high-sediment-concentration inflows, they are preferentially treated as water and sediment condition changes.

[0124] Therefore, in this embodiment, the input value weight is synthesized from the shallow time-series change score, the deep working condition score, the communication link score, and the cloud model prediction error score. The shallow time-series change score is determined based on the rate of change, the degree of fluctuation, the trend direction, and the magnitude of abrupt changes. Specifically, it compares the rate of change of the current monitored data with a rate of change threshold, the degree of fluctuation with a fluctuation threshold, and the magnitude of abrupt changes with a abrupt change threshold, then determines the shallow time-series change score based on the degree of exceedance. The deep working condition score is determined based on the confidence level of the working condition identification and the working condition safety level. Suspected leakage, suspected structural anomalies, flood discharge, low-temperature ice conditions, and high-sediment erosion have higher weights than ordinary steady-state water conveyance conditions. The communication link score is determined based on link latency, packet loss rate, bit error rate, retransmission count, and offline duration. The cloud model prediction error score is determined based on the difference between the predicted value fed back from the previous calculation cycle in the cloud and the field calibration value; the larger the difference, the higher the contribution of the data to model correction.

[0125] As can be seen, in this embodiment, various thresholds are determined according to the measuring point, data type, and engineering object. The rate of change threshold is taken as the 95th percentile of the rate of change of qualified data that has passed all verifications within the most recent 30 natural days, and compared with the maximum rate of change allowed by the operation and scheduling procedures, adopting the more stringent value. The fluctuation threshold is taken as the 95th percentile of the short-term fluctuation of qualified data within the most recent 30 natural days; the mutation threshold is taken as the 95th percentile of the short-term change amplitude of historical qualified data, and corrected in conjunction with the design safety margin of hydraulic structures. The communication link degradation threshold is taken as the 5th percentile of the link score under normal communication conditions within the most recent 30 natural days, and shall not be lower than the minimum available value specified in the communication operation and maintenance procedures. The cloud model prediction error threshold is obtained by the cloud digital twin platform from the model verification data, taking the 95th percentile of the relative prediction error of each data type at each measuring point; when the project undergoes a change in water diversion level, a change in pump station operation mode, a change in low temperature season, or a significant change in sediment conditions, it is re-determined using data that has passed all verifications within 7 consecutive natural days after the change.

[0126] In some embodiments, the composite coefficients of the input value weights are determined through historical operation playback. The cloud-based digital twin platform selects confirmed abnormal events, scheduling events, and steady-state operation segments as verification samples to test the high-value data recall, redundant data compression, and model error reduction under different combinations of composite coefficients. Under the premise of meeting the minimum recall requirements for abnormal events stipulated by the operation management unit, the composite coefficient combination with better redundant data compression and a larger reduction in model error is selected. The determined composite coefficients are stored separately for channel sections, tunnel sections, sluice gates, pumping stations, reservoirs, and water distribution nodes to avoid misjudgment caused by using the same set of parameters for different engineering objects.

[0127] In another specific implementation, the edge aggregation and autonomous processing layer determines the data packet type based on the input model value weight. Monitoring data with an input model value weight reaching a high value threshold are reconstructed into high-precision correction packets, simplified hazard feature packets, water-sediment coupling data packets, parameter correction data packets, or ice condition control data packets; monitoring data with an input model value weight in the middle range are reconstructed into feature enhancement data packets; monitoring data with an input model value weight below the low value threshold and in a stable operating state are reconstructed into differential data packets. The high value threshold is taken as the 85th percentile of the input model value weight of qualified monitoring frames in the most recent 30 calendar days, and the low value threshold is taken as the 45th percentile of the input model value weight of the same dataset. When the project is initially put into operation and there are insufficient valid samples, the high value threshold is taken as 0.75 and the low value threshold is taken as 0.35; after continuous operation for 30 calendar days and the formation of sufficient qualified data, the threshold is switched to the quantile threshold.

[0128] Optionally, the data packet type is also constrained by the operating condition and communication status. Data outputs with steady-state water conveyance and small model prediction errors are differential data packets; data outputs with significant safety risks and poor link quality are simplified hazard feature packets; data outputs with model prediction errors consistently exceeding the threshold and normal source-end quality are high-precision correction packets; data outputs with interconnected changes in sediment concentration, turbidity, flow rate, and cross-sectional sedimentation are water-sediment coupling data packets; data outputs with insignificant short-term changes but a long-term increase in model bias are parameter correction data packets; data outputs with temperatures approaching ice condition concern conditions and accompanied by abnormal water level, flow rate, or pressure are ice condition control data packets; data outputs buffered during link interruption and sent after recovery are offline resume data packets.

[0129] It should be noted that, in this application, after the data packet reconstruction is completed, the edge aggregation and autonomous processing layer sequentially perform communication integrity verification, numerical rationality verification, and multi-level physical constraint verification. Communication integrity verification includes timestamp continuity, data sequence number, missing packet, duplicate packet, and integrity check code verification. Timestamp continuity is used to determine whether data packets of the same data type at the same measurement point form a continuous sequence according to the sampling frequency; data sequence number is used to determine whether there are skipped or duplicate numbers; integrity check code is used to determine whether the transmitted content has been altered. If key fields are missing, check codes are inconsistent, or core fields of security-type data packets are incomplete, the data packet will not be included in the high-confidence input data set.

[0130] Furthermore, numerical rationality verification includes range verification, rate of change verification, and abrupt change amplitude verification. Range verification compares data related to water level, flow rate, pressure, sediment concentration, turbidity, cross-sectional sedimentation, leakage, and temperature with allowable ranges. Allowable ranges are jointly determined by sensor range, engineering design boundaries, operational scheduling boundaries, and equipment protection settings, using the most stringent boundary that is relevant to operational safety. Rate of change verification compares the rate of change of the current data relative to the previous valid data with the allowable rate of change. The allowable rate of change is the more stringent of the 99th percentile of the absolute value of the rate of change of valid data over the most recent 30 calendar days and the maximum allowable rate of change in the operational scheduling regulations. Abrupt change amplitude verification determines whether short-term changes exceed the physical response capability of the engineering object. The threshold is jointly determined by the hydraulic propagation time between measuring points, gate or pump station operation records, historical abrupt change samples, and equipment measurement uncertainty.

[0131] In addition, multi-level physical constraint verification includes intra-object constraints, adjacent object constraints, and full-line operational boundary constraints. Intra-object constraints are used to determine whether the water level in the channel section exceeds the design control water level, whether the flow rate exceeds the cross-sectional water conveyance capacity, whether the leakage exceeds the segmental allowable value, and whether the cross-sectional siltation shows a jump inconsistent with the measurement period; they are used to determine whether the pressure in the tunnel section exceeds the allowable pressure, whether the pressure change rate exceeds the limit specified in the structural safety evaluation document, and whether there is a reasonable response relationship between leakage and pressure change; they are used to determine whether the gate opening, upstream and downstream water level difference, and gate flow rate match; and they are used to determine whether the pump set start-up and shutdown status, inlet and outlet pressure, and flow rate match. Adjacent object constraints are used to determine whether the upstream disturbance and downstream response conform to the propagation time relationship. Full-line operational boundary constraints are used to determine whether the water level, flow rate, pressure, water distribution, flood discharge status, flow restriction status, and pressure reduction and diversion status along the line meet the approved water transfer plan, operation and scheduling procedures, equipment protection settings, and safety evaluation report.

[0132] As shown above, the process for determining the water conservation constraint is as follows: First, obtain the upstream inflow and downstream outflow of the control section. Then, determine the water storage change based on water level changes, cross-sectional geometric parameters, and reservoir or channel storage curves. Subsequently, determine the segmented seepage loss based on seepage monitoring values ​​or channel seepage models, and compare the water balance with the distribution volume and flood discharge volume. The deviation generated by the water balance comparison must not exceed the allowable deviation. The allowable deviation is jointly determined by the uncertainty of upstream flow measurement, downstream flow measurement, water storage calculation, distribution volume measurement, and seepage estimation. To determine this, the squares of each uncertainty are summed, and the square root of the sum is taken. If the deviation does not exceed the allowable deviation, the water conservation constraint is deemed passed; if it exceeds the allowable deviation, the data is deemed physically inconsistent.

[0133] In this embodiment, data packets that pass communication integrity verification, numerical reasonableness verification, and multi-level physical constraint verification are marked as high-confidence input data. Data that fails communication integrity verification but has the conditions for retransmission is marked as data to be retransmitted; data with complete communication and a tendency for values ​​to exceed limits, while the physical relationship can support this change, is marked as suspected abnormal data; data with complete communication but poor source quality and a mismatch between values ​​and physical relationships is marked as low-confidence data. High-confidence input data, carrying data source, spatiotemporal identifier, data packet type, confidence level, verification results, and model parameter mapping suggestions, enters the cloud-based digital twin platform.

[0134] Furthermore, step 3 is executed by the cloud-based digital twin platform. After receiving the highly reliable input data, the cloud-based digital twin platform first establishes a parameter mapping relationship according to the data packet type, measurement point spatial encoding, unified timestamp, physical constraint verification results, and data reliability. Water level, flow rate, upstream and downstream water level difference, gate opening, and pump station status are mapped to channel roughness, local head loss parameters, boundary flow correction parameters, and flow parameters in the hydrodynamic model; sediment concentration, turbidity, flow rate, and inflow conditions are mapped to sediment deposition parameters, initiation parameters, sediment transport capacity parameters, and sediment-laden boundary conditions in the water-sediment transport model; cross-sectional sedimentation, effective profile ratio, velocity variation, and sediment concentration variation are mapped to scour parameters, sedimentation parameters, and cross-sectional geometric correction parameters in the dynamic scour and sedimentation model; leakage, water level, pressure, lining status, and water balance deviation are mapped to segmented seepage parameters and lining leakage parameters in the channel seepage model; pressure, temperature-related data, leakage, vibration, and lining strain are mapped to pressure load parameters, temperature influence parameters, structural response parameters, and safety evaluation parameters in the tunnel structure safety assessment model.

[0135] It should be noted that in this application, step 4 is completed in the scheduling control execution layer. After receiving the scheduling control strategy, the scheduling control execution layer first performs safety constraint verification based on the strategy type, strategy priority, model credibility, operating condition risk level, and on-site execution constraints. The gate control strategy verifies the target opening degree, current opening degree, single opening degree change, opening degree change rate, hoist status, and upstream and downstream water level difference; the pump station control strategy verifies the pump set start-stop interval, start-stop frequency, target speed, inlet and outlet water pressure, and operating flow rate; the flood discharge facility control strategy verifies the flood discharge flow growth rate, downstream control water level, and downstream bearing capacity; the water diversion facility control strategy verifies the target water diversion volume, water diversion node water level, and allocation constraints; and the flow restriction facility and pressure reducing diversion facility control strategies verify the pressure change rate, seepage pressure, vibration, lining strain, and target pressure.

[0136] Furthermore, various safety thresholds are jointly determined based on equipment documents, engineering design documents, operating procedures, and safety evaluation results. The threshold for the gate opening change rate is determined by the hoist technical documents, gate structural safety evaluation, downstream water level control requirements, and historical safety action records; the threshold for the pump station start-stop frequency is determined by the pump unit equipment manual, motor protection procedures, frequency converter operating requirements, and operation management procedures; the threshold for the flood discharge flow rate increase is determined by the flood control scheduling plan, the allowable water level rise rate in downstream channels, the opening and closing capacity of flood discharge facilities, and the safety requirements of downstream protected objects; the threshold for the tunnel pressure change rate is determined by the tunnel structural safety evaluation report, lining material performance, and historical pressure response records; and the thresholds for seepage pressure, vibration, and lining strain are determined by structural safety monitoring procedures, material allowable values, and historical stable operating ranges.

[0137] This construction method starts with spatiotemporal calibration at the source end, and writes the spatial coding of the measurement points, unified timestamps, engineering topology relationships and quality status markers into the multi-source original monitoring data frames. This gives the subsequent edge processing and cloud modeling clear data sources and traceability basis, and can reduce model errors caused by data mismatch, time drift and unclear measurement point attribution after long-distance transmission across regions.

[0138] As can be seen, in this embodiment, the edge aggregation and autonomous processing layer calculates the input model value weights based on shallow temporal characteristics, deep working condition characteristics, communication link status, and cloud model prediction errors. It then adaptively generates different types of data packets according to these input model value weights. This enables the compression of low-value data during steady-state water conveyance and prioritizes the retention of critical data during periods of sudden risks, slow-changing deviations, low-temperature ice conditions, high-sediment-content water inflows, and communication degradation. This increases the effective proportion of high-reliability input model data. Through communication integrity verification, numerical rationality verification, and multi-level physical constraint verification, it can distinguish between transmission errors, equipment anomalies, real-world engineering anomalies, and data awaiting retransmission before the data enters the model, reducing the interference of low-reliability data on model parameter inversion. After establishing a mapping relationship between high-reliability input model data and the model parameters to be updated, the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model, and tunnel structural safety assessment model can be corrected online based on the verified data, improving the consistency between simulation results and scheduling control strategies and actual on-site conditions.

[0139] Optionally, in sections where local measuring points are sparse or sensors are temporarily unavailable, the cloud-based digital twin platform employs adjacent similar sections for assisted inversion processing. The platform first selects adjacent sections with similar materials, cross-sections, slopes, operating flow rates, maintenance status, and historical operating characteristics as reference sections. Then, it obtains parameter change trends based on high-reliability input data from the reference sections. Subsequently, it performs restrictive corrections by combining the remaining valid data for this section, engineering design boundaries, and historical operating records. This embodiment does not directly copy parameters from adjacent sections to this section; instead, it only uses their change trends, with the design boundaries and measured data of this section as constraints. This approach is suitable for scenarios in long-distance engineering projects where some monitoring nodes are temporarily unavailable but model updates still need to be maintained.

[0140] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin system for a cross-regional, long-distance water transfer project, characterized in that, include: The field sensing and communication networking layer is used to collect data on water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status, and temperature from channels, tunnels, sluice gates, pumping stations, reservoirs, and water distribution nodes, and upload them to the edge aggregation and autonomous processing layer. An edge aggregation and autonomous processing layer is used to perform preprocessing on the data before it is input into the model, including: Extract shallow temporal features and deep working condition features; Based on the shallow time-series characteristics, deep working condition characteristics, communication link status, and model prediction errors fed back from the cloud digital twin platform layer, the input value weight of each frame of monitoring data is calculated. Based on the input model value weight, each frame of monitoring data is adaptively reconstructed into a corresponding type of data packet, and the data packet is subjected to communication integrity verification, numerical rationality verification and multi-level physical constraint verification. Data packets that pass the verification are marked as high-confidence input model data. The cloud-based digital twin platform layer is used to perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model based on the high-reliability input model data, and generate engineering operation status simulation results or scheduling and control strategies. The scheduling and control execution layer is used to perform control actions on gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, or pressure reduction and diversion facilities according to the scheduling and control strategy.

2. The digital twin system for a cross-regional long-distance water transfer project according to claim 1, characterized in that, The field sensing and communication networking layer: Acquire water level, flow rate, pressure, sediment content, turbidity, cross-sectional sedimentation, leakage, communication link status, and temperature-related data collected at each monitoring node; Based on the spatial location of the measuring points, unified timestamps, data types, and engineering topology, the data is calibrated spatiotemporally at the source end, and the sampling frequency is dynamically adjusted according to changes in water level, pressure fluctuations, sediment content, cross-sectional siltation, leakage, temperature, and link quality. Output multi-source raw monitoring data frames with spatial coding of measurement points, unified timestamp, engineering topology relationship, sampling frequency, link status and quality status markers.

3. The digital twin system for a cross-regional long-distance water transfer project according to claim 1, characterized in that: The edge convergence and autonomous processing layer: Obtain the original monitoring data frames from the multiple sources; Based on the unified timestamp, spatial coding of measurement points and engineering topology, time sequence alignment and spatial correlation are performed to extract shallow time sequence features and deep working condition features. The input value weight of each frame of monitoring data is calculated by combining the communication link status, source end quality status marking and cloud model prediction error. The data representation format, transmission priority, and input path are determined based on the input model value weight, and at least one of the following is output: differential data packet, simplified hazard feature packet, high-precision correction packet, water-sand coupling data packet, parameter correction data packet, ice condition control data packet, or offline continuation data packet.

4. The digital twin system for a cross-regional long-distance water transfer project according to claim 3, characterized in that: The edge convergence and autonomous processing layer: Obtain the adaptively reconstructed data packet; Based on at least one of the following: timestamp continuity, data sequence number, missing packets, duplicate packets, integrity check code, range, rate of change, abrupt change amplitude, water conservation, water and sediment movement, scouring and silting geometry, leakage response, low temperature ice conditions, gate position-flow response, and upstream and downstream propagation causality, the data packet is subjected to communication integrity verification, numerical rationality verification, and multi-level physical constraint verification. Output high-confidence input data, suspected abnormal data, low-confidence data, or data to be supplemented, and form a correction data chain based on the high-confidence input data, which includes data source, spatiotemporal identifier, confidence level, and model parameter mapping relationship.

5. The digital twin system for a cross-regional long-distance water transfer project according to claim 4, characterized in that: The cloud-based digital twin platform layer: Obtain the high-confidence input model data and the correction data chain; Based on data packet type, spatial coding of measurement points, unified timestamp, physical constraint verification results and data credibility, establish parameter mapping relationships between highly reliable input data and hydrodynamic models, water and sediment transport models, dynamic scouring and silting models, channel leakage models and tunnel structural safety assessment models, and perform hierarchical online parameter inversion updates and coupling consistency verification. Output simulation results of engineering operation status, flood risk early warning results, high sediment erosion and siltation evolution prediction results, seepage loss prediction results, low temperature ice risk assessment results, tunnel structure safety assessment results, or scheduling and control strategies.

6. The digital twin system for a cross-regional long-distance water transfer project according to claim 1, characterized in that: The scheduling control execution layer: Obtain the scheduling and control strategy output by the cloud-based digital twin platform layer; Based on strategy type, strategy priority, model credibility, working condition risk level and on-site execution constraints, hierarchical execution instructions are generated, and safety constraints are checked before execution based on gate opening change rate, pump station start and stop frequency, flood discharge flow rate, upstream and downstream water level difference, tunnel pressure change rate, seepage pressure, vibration and lining strain. Output phased control actions for gates, pumping stations, flood discharge facilities, water diversion facilities, flow restriction facilities, or pressure reduction and diversion facilities, and feed back the action feedback and engineering response to the edge aggregation and autonomous processing layer or the cloud digital twin platform layer.

7. A method for constructing a digital twin system applicable to any one of claims 1-6 of a trans-regional long-distance water transfer project, characterized in that, include: S1. Collect water level, flow rate, pressure, sediment content, turbidity, cross-sectional siltation, leakage, communication link status and temperature-related data from channels, tunnels, sluice gates, pumping stations, reservoirs and water distribution nodes, and perform source-end spatiotemporal calibration and quality status marking on the data to form multi-source raw monitoring data frames. S2. Based on the multi-source original monitoring data frames, extract shallow time-series features and deep operating condition features. According to the shallow time-series features, deep operating condition features, communication link status, and model prediction error fed back by the cloud digital twin platform, calculate the input value weight of each frame of monitoring data. Based on the input value weight, adaptively reconstruct each frame of monitoring data into a corresponding type of data packet, and perform communication integrity verification, numerical rationality verification, and multi-level physical constraint verification on the data packets. Mark the data packets that pass the verification as high-confidence input data. S3. Based on the high-reliability input data, establish a mapping relationship between the high-reliability input data and the model parameters to be updated, and perform online parameter inversion and update of at least one of the hydrodynamic model, water and sediment transport model, dynamic scouring and silting model, channel leakage model and tunnel structure safety assessment model. S4. Generate simulation results of engineering operation status or scheduling control strategy based on the updated model, and perform safety constraint verification before execution. Execute control actions or limit, delay, downgrade, execute in stages or roll back the scheduling control strategy according to the verification results. Use action feedback and engineering response feedback for model parameter correction, updating of input value weights and subsequent rolling generation of scheduling control strategies.

8. The method for constructing a digital twin system for a cross-regional long-distance water transfer project according to claim 7, characterized in that: The weights of the input model value are determined by the hydraulic mutation factor, sand erosion and deposition factor, working condition risk factor, link urgency factor, model error factor, data reliability factor, and link occupancy factor. The hydraulic mutation factor is used to characterize the rate of change and mutation intensity of water level, flow rate or pressure; the sand erosion and sedimentation factor is used to characterize the degree of change of sediment content, turbidity or cross-sectional sedimentation; the link urgency factor is used to characterize the communication link delay, packet loss rate or buffer congestion; and the model error factor is used to characterize the deviation between the digital twin model prediction value and the field measured value. The edge aggregation and autonomous processing layer determines the data packet type, transmission priority, and input path of each frame of monitoring data based on the comparison results of the input value weight and the dynamic threshold; the dynamic threshold is adaptively adjusted based on historical high-reliability input data, current operating condition risk level, communication link quality, and model prediction error.

9. The method for constructing a digital twin system for a cross-regional long-distance water transfer project according to claim 7, characterized in that: When the edge aggregation and autonomous processing layer identifies data disorder, delay, missing or duplicate data caused by multi-hop transmission in weak network, it performs hydraulic causal reconstruction on the data based on unified timestamp, measurement point spatial coding, engineering topology relationship, upstream and downstream hydraulic propagation direction, gate position-flow response relationship and water level propagation time delay. The hydraulic causal reconstruction includes determining the temporal position of the data in the actual hydraulic process based on the sequential relationship of water level, flow rate or pressure changes between upstream measuring points, tunnel measuring points and downstream measuring points, and marking missing data or estimating and filling in missing data based on adjacent high-confidence input data. The edge convergence and autonomous processing layer outputs a time-series data chain and its reconstruction credibility after hydraulic causality reconstruction, which is used by the cloud-based digital twin platform layer to reconstruct the flood propagation process, correct hydrodynamic propagation parameters, or update the simulation results of engineering operation status.

10. The digital twin system or construction method for a cross-regional long-distance water transfer project according to claim 7, characterized in that: Under low-temperature conditions in winter, the edge convergence and autonomous treatment layer identifies low-temperature ice conditions based on water temperature, air temperature, tunnel wall temperature, soil temperature, water level, flow velocity, pressure, upstream and downstream gate opening degree, and duration of low temperature. The low-temperature ice conditions include at least one of the following: freezing risk level, ice blockage risk level, freeze-thaw damage risk level, low-temperature stagnant water zone, and freeze-thaw sensitive water level zone. When the icing risk level reaches the preset antifreeze threshold and no ice blockage is detected, the cloud digital twin platform layer or the edge convergence and autonomous processing layer generates an anti-icing control strategy. This strategy maintains the minimum antifreeze flow rate in the tunnel by adjusting the upstream and downstream gates in a coordinated manner, shortens the water's residence time in the low-temperature tunnel section, and avoids the freeze-thaw sensitive water level range. When icing or ice blockage risk is detected in the tunnel, an icing impact mitigation strategy is generated. This strategy reduces the impact of ice blockage, water hammer, frost heave, or ice impact on the tunnel structure by adjusting upstream and downstream gates in stages, limiting the flow rate, limiting the water level change rate, reducing tunnel pressure fluctuations, or triggering pressure reduction and diversion operations.

Images