A Three-Dimensional Monitoring System and Method for Water Level of Hydraulic Facilities Based on Big Data
By integrating multi-source topographic and water level data, a continuous three-dimensional water level field across the entire region is reconstructed, solving the problem of the single spatial dimension of water level monitoring for hydraulic facilities. This achieves high-precision, three-dimensional water level monitoring, improves the accuracy of reservoir capacity calculation and risk assessment, adapts to complex scouring and silting scenarios, and ensures the long-term accuracy of water level monitoring.
Patent Information
- Application Number
- CN202611388932.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-08
- Publication Date
- 2026-11-17
AI Technical Summary
Water level monitoring of hydraulic facilities suffers from a single spatial dimension, failing to depict the continuous spatial distribution of water levels along the river, throughout the entire reservoir area, and at the boundary between the riverbank and the slope. This leads to systematic biases in the assessment of embankment overflow risks, reservoir capacity calculations, and analysis of river flow capacity. Furthermore, the dynamic changes in topography are completely decoupled from water level monitoring, and the accuracy of water level monitoring continues to decline with the increase in the years of project operation, making it unsuitable for the needs of high-precision flood control scheduling and refined water resource management.
By using a big data-based three-dimensional water level monitoring method, multi-source topographic and water level data are integrated to reconstruct a continuous three-dimensional water level field across the entire region. A topographic-water level linkage correction mechanism is established to dynamically correct water level benchmark deviations caused by scouring and silting. Combined with multi-source collaborative reconstruction and closed-loop accuracy verification, the accuracy of water level monitoring is improved.
This represents a leap from discrete point monitoring to comprehensive three-dimensional sensing, significantly improving the accuracy of reservoir capacity calculation and overflow risk assessment, ensuring the long-term accuracy and reliability of water level monitoring results, and providing high-precision, three-dimensional monitoring data support for hydraulic engineering facilities.
Smart Images

Figure CN123111405A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of water level monitoring, specifically to a three-dimensional water level monitoring system and method for hydraulic facilities based on big data. Background Technology
[0002] Reservoirs, rivers, dikes, and other hydraulic engineering facilities are core engineering carriers for flood control, disaster reduction, and water resource allocation. Water level monitoring data is the core foundation for safe operation and maintenance of these projects and for scheduling decisions. Currently, the industry commonly adopts the approach of deploying radar water level gauges, pressure water level gauges, and other sensing devices at key cross-sections to obtain water level data. Some projects supplement this with satellite remote sensing imagery to retrieve water levels over large areas. Monitoring systems are generally based on initial topographic data from the project construction phase to calibrate core relationships such as water level-reservoir capacity and water level-flow rate, supporting daily operation and maintenance and scheduling management. Current water level monitoring for hydraulic engineering facilities suffers from several drawbacks. It relies on a single spatial dimension, acquiring only instantaneous water level values at discrete points. This fails to depict the continuous spatial distribution of water levels along the riverbank, throughout the entire reservoir area, and at the riverbank boundary, making it difficult to construct a comprehensive three-dimensional water level field. This directly leads to systematic biases in assessing embankment overflow risks, calculating reservoir capacity, and analyzing river flow capacity. Furthermore, the dynamic changes in topography are completely decoupled from water level monitoring. Over long-term operation, hydraulic engineering facilities are subject to scouring, sediment deposition, and riverbank collapse, causing continuous dynamic evolution of the underwater and riverbank topography. This gradually distorts the correlation between water level and volume, resulting in a continuous decline in monitoring accuracy as the project ages. Consequently, this approach cannot meet the demands of high-precision flood control scheduling and refined water resource management.
[0003] Chinese invention patent CN116698151B, published on April 21, 2026, discloses a groundwater recharge inversion system for irrigation and drainage lining channels based on water level monitoring. The system includes a lining state detection unit to obtain the lining state of the channel's lining surface; a groundwater monitoring unit to install monitoring wells along the channel's location; obtaining the water level in the monitoring wells; and determining the groundwater level distribution in local areas along the channel based on the well water levels. An inversion calculation unit is used to calculate the relationship between the groundwater level distribution along the channel and the lining state, and to invert the groundwater level distribution along the channel based on this relationship. However, this technical solution cannot dynamically monitor water level characteristics based on changes in channel topography. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] To address the shortcomings of current water level monitoring systems for hydraulic engineering facilities, which rely on a single spatial dimension and only acquire instantaneous water level values at discrete points, failing to depict the continuous spatial distribution of water levels along the riverbank, throughout the entire reservoir area, and at the riverbank boundary, and thus hindering the construction of a comprehensive three-dimensional water level field, this approach directly leads to systematic biases in assessing embankment overflow risks, calculating reservoir capacity, and analyzing river flow capacity. Furthermore, the dynamic changes in topography are completely decoupled from water level monitoring. During long-term operation, hydraulic engineering facilities are subject to continuous dynamic evolution of underwater and riverbank topography due to water flow scouring, sediment deposition, and riverbank collapse, resulting in changes in water level and volume. The correlation between data and water level data gradually becomes distorted, and the monitoring accuracy continuously declines with the age of the project, failing to meet the needs of high-precision flood control scheduling and refined water resource management. To address this, we integrate multi-source topographic and water level data to reconstruct a continuous three-dimensional water level field across the entire region, overcoming the limitations of single-point monitoring and significantly improving the spatial coverage of water level monitoring. We also establish a topographic-water level linkage correction mechanism to dynamically correct water level benchmark deviations caused by erosion and sedimentation, effectively solving the problem of continuous decline in water level monitoring accuracy with topographic evolution. Finally, we combine multi-source collaborative reconstruction and closed-loop accuracy verification to improve the accuracy of water level monitoring.
[0006] (II) Technical Solution
[0007] This invention is achieved through the following technical solution: a three-dimensional monitoring method for water levels in hydraulic structures based on big data, the method comprising the following steps:
[0008] S1. Simultaneously acquire multi-source topographic and water level raw data of the area covered by hydraulic facilities, perform spatiotemporal benchmark alignment and outlier removal preprocessing, and generate a spatiotemporal integrated monitoring dataset containing underwater topographic sequence, bank slope topographic sequence, point water level sequence and remote sensing inversion water level sequence.
[0009] S2. Input the spatiotemporal integrated monitoring dataset into a sliding spatiotemporal window, and perform spatial difference and temporal statistical feature extraction respectively to generate a multi-dimensional topographic-water level coupled feature vector containing topographic erosion and sedimentation rate, water level spatial gradient coefficient and data confidence weight.
[0010] S3. Input the multidimensional topography-water level coupled feature vector into the trained three-dimensional water level monitoring big data analysis model, and combine it with the current benchmark topography and water level data to obtain the three-dimensional water level field prediction results and topography evolution correction amount for the target area in the future period.
[0011] S4. Based on the predicted results of the three-dimensional water level field, the terrain evolution correction amount, and the current monitoring data, generate a set of three-dimensional monitoring correction parameters including spatial interpolation correction coefficients and terrain compensation parameters.
[0012] S5. Based on the three-dimensional monitoring correction parameter set, generate a continuous three-dimensional water level field distribution result across the entire domain through a multi-source data collaborative reconstruction algorithm, output the three-dimensional water level monitoring data of the target area, and continuously acquire the feedback monitoring data for the next cycle.
[0013] S6. Calculate the spatiotemporal deviation between the feedback monitoring data and the three-dimensional water level field prediction result, and output the corresponding monitoring accuracy evaluation result based on the spatiotemporal deviation.
[0014] Preferably, step S1 includes the following steps:
[0015] S11. Collect underwater topographic point cloud data of the water area of the hydraulic engineering facility, remove suspended solids and sounding noise by combined filtering, and obtain a gridded underwater topographic sequence by spatial interpolation.
[0016] S12. Collect topographic elevation point cloud data of the bank slope and surrounding land area, and obtain a gridded bank slope topography sequence with the same reference as the underwater topography after filtering by ground features and fitting and correcting the elevation of the reference station.
[0017] S13. Collect point water level monitoring data at fixed cross-sections, combine with real-time meteorological parameters to perform wave velocity compensation correction, and obtain a standardized point water level sequence after sliding filtering.
[0018] S14. Acquire satellite radar altimeter remote sensing images, extract water body boundaries after radiometric and geometric fine correction, and obtain remote sensing inverted water level sequences by combining them with bank slope topography inversion.
[0019] S15. Using a unified geodetic coordinate system and time reference as the standard, spatial interpolation and temporal interpolation algorithms are used to perform spatiotemporal registration of each sequence to construct the spatiotemporal integrated monitoring dataset.
[0020] Preferably, step S2 includes the following steps:
[0021] S21. Set a sliding time-space window, traverse the underwater topography sequence and the bank slope topography sequence based on the sliding time-space window, and use the least squares method to fit the temporal change trend of the elevation of each grid within the window, and use the slope of the trend line as the topography erosion and deposition change rate.
[0022] S22. Within the sliding time-space window, calculate the spatial gradient change of water level data along the longitudinal and transverse directions of the river channel, and combine it with the topographic slope characteristics of the corresponding location for weighted correction to obtain the spatial gradient coefficient of water level.
[0023] S23. Extract the data integrity and historical average deviation of each data source within the current sliding window, and calculate the data confidence weight of each data source.
[0024] S24. Normalize and concatenate the topographic erosion and sedimentation rate, water level spatial gradient coefficient, data confidence weight, and the current benchmark topographic value and benchmark water level value to obtain a multidimensional topographic-water level coupled feature vector.
[0025] Preferably, step S3 includes the following steps:
[0026] S31. The multidimensional topography-water level coupling feature vector is used as an input sequence and input into the input layer of the water level three-dimensional monitoring big data analysis model.
[0027] S32. Using the temporal memory unit of the water level three-dimensional monitoring big data analysis model, combined with historical topographic evolution and water level change records, short-term occasional disturbance features are filtered out, and a hidden layer feature vector containing spatiotemporal evolution patterns is output.
[0028] S33. Input the hidden layer feature vector into the space to reconstruct the regression layer, and output the prediction result of the continuous three-dimensional water level field in the whole domain within the future target time period through nonlinear mapping.
[0029] S34. Input the hidden layer feature vector into the terrain correction branch, and calculate the terrain evolution correction amount of each grid in the corresponding time period by combining the historical erosion and deposition patterns.
[0030] Preferably, the training process of the water level three-dimensional monitoring big data analysis model in S31 includes:
[0031] S311. Obtain spatiotemporal integrated monitoring datasets of hydraulic facilities under different historical hydrological conditions, as well as corresponding measured three-dimensional water level distribution and topographic mapping records, and construct a historical topographic-water level evolution sample library.
[0032] S312. Divide the historical topography-water level evolution sample database into training set and validation set according to the chronological order, set the model network layer, number of hidden nodes and learning rate parameters, and construct the initial big data analysis model.
[0033] S313. Input the training set into the initial model, and use the weighted sum of the mean square error of the three-dimensional water level field prediction and the mean absolute error of the terrain evolution correction as the comprehensive loss function.
[0034] S314. The gradient of the comprehensive loss function is calculated using an adaptive optimization algorithm, and the model weights are updated in reverse until the value of the comprehensive loss function on the validation set no longer decreases. The weights are then saved to obtain the trained water level three-dimensional monitoring big data analysis model.
[0035] Preferably, step S4 includes the following steps:
[0036] S41. Based on the terrain evolution correction amount, and combined with the water surface area corresponding to the current water level, calculate the water level-reservoir capacity sensitivity coefficient to obtain the terrain compensation parameters for water level monitoring, which are used to correct the water level benchmark deviation caused by scouring and silting changes.
[0037] S42. Based on the spatial deviation between the three-dimensional water level field prediction result and the point water level data, dynamically adjust the variation function parameters of the spatial interpolation to obtain the spatial interpolation correction coefficient, which is used to optimize the spatial interpolation accuracy of the water level across the entire area.
[0038] S43. Combine the terrain compensation parameters with the spatial interpolation correction coefficients to generate a three-dimensional monitoring correction parameter set.
[0039] Preferably, step S5 includes the following steps:
[0040] S51. Input the three-dimensional monitoring correction parameter set into the multi-source data collaborative reconstruction module, and establish a reconstruction equation with the dual objectives of minimizing the measured deviation of the point and optimizing the spatial continuity and smoothness of the water level field, with data confidence weight as a constraint.
[0041] S52. Solve the reconstruction equation. First, combine the terrain compensation parameters to correct the water level benchmark grid by grid. Then, substitute the corrected interpolation parameters and generate the global continuous three-dimensional water level field distribution result through a weighted spatial interpolation algorithm.
[0042] S53. Outputs three-dimensional water level monitoring data including grid-by-grid water level elevation, water level along the cross section, and inundation boundary along the shore, and simultaneously generates derived calculation results for reservoir capacity and flow capacity.
[0043] S54. Collect multi-source topographic and water level data for the next monitoring cycle at a fixed period, and use the data as feedback monitoring data input for the closed-loop verification process.
[0044] Preferably, step S6 includes the following steps:
[0045] S61. Extract the measured three-dimensional water level distribution data for the next cycle, compare it with the three-dimensional water level field prediction results for the corresponding time period grid by grid, and calculate the average relative error of the whole domain as the spatiotemporal deviation.
[0046] S62. Set an accuracy warning threshold. If the spatiotemporal deviation is greater than the accuracy warning threshold, output an abnormal monitoring accuracy warning. If the spatiotemporal deviation is not greater than the accuracy warning threshold, output a qualified monitoring accuracy indicator and archive and store the current three-dimensional water level monitoring data.
[0047] S63. Statistically analyze the deviation distribution by region and time period, generate a monitoring accuracy assessment report, and record the monitoring error characteristics of different regions and time periods.
[0048] A big data-based three-dimensional water level monitoring system for hydraulic engineering facilities is provided to implement the big data-based three-dimensional water level monitoring method for hydraulic engineering facilities. The system includes:
[0049] The multi-source water level and topographic data acquisition and preprocessing module is used to synchronously access the raw data from multiple sources, complete noise filtering, correction and spatiotemporal registration, and output a spatiotemporal integrated monitoring dataset.
[0050] The topographic-water level coupling feature extraction module is used to extract the scouring and sedimentation rate, spatial gradient coefficient and confidence weight through sliding spatiotemporal window and statistical calculation, and generate a multidimensional topographic-water level coupling feature vector.
[0051] The three-dimensional water level prediction and terrain correction module has a built-in three-dimensional water level monitoring big data analysis model, which is used to output the three-dimensional water level field prediction results and terrain evolution correction.
[0052] The water level field reconstruction parameter optimization module is used to generate a three-dimensional monitoring correction parameter set and solve the reconstruction equation to obtain a continuous three-dimensional water level field over the entire domain.
[0053] The water level monitoring closed-loop verification module is used to calculate the spatiotemporal deviation between the predicted results and the measured data, and output the accuracy assessment results and anomaly warnings.
[0054] (III) Beneficial Effects
[0055] This invention provides a three-dimensional water level monitoring system and method for hydraulic structures based on big data. It has the following beneficial effects:
[0056] I. By integrating multi-source monitoring data such as underwater topography, bank slope topography, point water level, and satellite remote sensing, and combining big data analysis models to reconstruct a continuous three-dimensional water level field across the entire area, the spatial limitations of traditional single-point monitoring have been overcome. It can fully depict the spatial distribution characteristics of water level along the river, the entire reservoir area, and the boundary zone between the bank and slope, realizing a leap from discrete point monitoring to three-dimensional perception across the entire area. This significantly improves the accuracy of reservoir capacity calculation and overflow risk assessment, and provides complete spatial data support for the refined management and control of hydraulic engineering facilities.
[0057] Second, by establishing a linkage correction mechanism between dynamic topographic changes and water level monitoring, the topographic scouring and sedimentation change rate is extracted, the topographic evolution correction amount is calculated, and the water level monitoring benchmark is corrected grid by grid. This solves the problem of continuous attenuation of water level monitoring accuracy caused by sediment scouring and sedimentation and bank collapse during the long-term operation of hydraulic facilities. It ensures that the water level-reservoir capacity and water level-flow relationship always match the actual topographic state, effectively guaranteeing the long-term accuracy and reliability of water level monitoring results. It can be adapted to complex scouring and sedimentation scenarios such as rivers with high sediment content and reservoirs with high sediment content.
[0058] Third, by constructing a big data monitoring system with closed-loop accuracy verification, the system can achieve full-process accuracy control of monitoring results based on spatiotemporal deviation calculation. It can output accuracy anomaly warnings and qualified archiving results in a tiered manner, providing quantitative guidance for monitoring and maintenance. By relying on big data technology, the system can fully explore the coupling value of multi-source monitoring data, realize the coordinated monitoring of topography and water level, and improve the high precision, three-dimensionality, and intelligence of water level monitoring. Attached Figure Description
[0059] Figure 1 A flowchart of the three-dimensional water level monitoring method for hydraulic facilities based on big data provided by the present invention;
[0060] Figure 2 A schematic diagram of the modules of the three-dimensional water level monitoring system for hydraulic facilities based on big data provided by the present invention. Detailed Implementation
[0061] 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.
[0062] An embodiment of the big data-based three-dimensional water level monitoring system and method for hydraulic structures is as follows:
[0063] Example 1:
[0064] Please see Figures 1-2 A three-dimensional monitoring method for water levels in hydraulic structures based on big data, comprising the following steps:
[0065] S1. Simultaneously acquire multi-source topographic and water level raw data of the area covered by hydraulic facilities, perform spatiotemporal benchmark alignment and outlier removal preprocessing, and generate a spatiotemporal integrated monitoring dataset containing underwater topographic sequence, bank slope topographic sequence, point water level sequence and remote sensing inversion water level sequence.
[0066] S1 includes the following steps:
[0067] S11. Collect underwater topographic point cloud data of the water area of the hydraulic engineering facility, remove suspended solids and sounding noise by combined filtering, and obtain a gridded underwater topographic sequence by spatial interpolation.
[0068] Specifically, in embodiment S11, a shipborne multibeam echo sounder system is used to periodically conduct full-coverage mapping of the underwater topography of the reservoir area. The echo sounder system operates at a frequency of 200kHz, with a depth range of 0.5m to 120m and a depth resolution of 1cm. The sampling interval along the track is 0.5m, and a full-coverage mapping of the reservoir area is completed once per quarter. The original point cloud data contains pseudo-elevation points caused by scattering from sediment and plankton in the water. The system uses a combination of "radius filtering + statistical filtering" denoising algorithm. The radius filtering is set with a search radius of 2m and a neighboring point threshold of 10 to remove isolated noise points. The statistical filtering is set with a standard deviation multiple of 3 to remove outliers that deviate from the local mean. Finally, the inverse distance weighted interpolation method is used to generate a gridded underwater topography sequence with a resolution of 5m×5m. The elevation datum is unified as the 1985 National Elevation Datum. The relative error of the processed topography data is less than 5cm.
[0069] S12. Collect topographic elevation point cloud data of the bank slope and surrounding land area, and obtain a gridded bank slope topography sequence with the same reference as the underwater topography after filtering by ground features and fitting and correcting the elevation of the reference station.
[0070] Specifically, in Example S12, a hexacopter UAV-borne LiDAR aerial survey system was used to collect shore slope topographic data. The laser emission frequency was 200kHz, the point cloud density was 10 points / m², and the elevation measurement accuracy was ±3cm. The system covered the area below the 1:2000 elevation zone on both sides of the reservoir, and aerial surveys were conducted once per quarter. After filtering the original point cloud for vegetation, buildings, and other features, the elevation was fitted and corrected using three GNSS reference stations in the reservoir area as controls. The corrected system deviation was less than 2cm. Similarly, a gridded shore slope topographic sequence with a resolution of 5m×5m was generated by interpolation, which was completely consistent with the underwater topographic elevation benchmark.
[0071] S13. Collect point water level monitoring data at fixed cross-sections, combine with real-time meteorological parameters to perform wave velocity compensation correction, and obtain a standardized point water level sequence after sliding filtering.
[0072] Specifically, in embodiment S13, high-frequency pulse radar water level gauges are deployed at three key cross-sections: upstream, midstream, and upstream. The measurement range is 0m to 80m, the measurement accuracy is ±1mm, and the sampling frequency is 1 time / minute. Since changes in air temperature and pressure affect the propagation speed of radar waves in the air, the system accesses real-time air temperature and pressure data from the reservoir meteorological station in real time and corrects the radar wave propagation speed according to the atmospheric refractive index correction formula. Then, a moving average filter with a window length of 5 sampling points is used to eliminate high-frequency fluctuation noise and obtain a standardized point water level sequence. The overall measurement error is controlled within ±2mm.
[0073] S14. Acquire satellite radar altimeter remote sensing images, extract water body boundaries after radiometric and geometric fine correction, and obtain remote sensing inverted water level sequences by combining them with bank slope topography inversion.
[0074] Specifically, in embodiment S14, data from the Sentinel-3 satellite radar altimeter with a spatial resolution of 300m and a revisit period of 1 day are accessed; radiometric calibration and atmospheric correction are performed on the satellite imagery, and geometric fine correction is performed using data from fixed water level stations in the reservoir area as control points, with the corrected geometric error being less than 1 pixel; the water boundary is extracted through the normalized water index, and the remote sensing inversion water level sequence along the route is obtained by combining it with the bank slope topography inversion, serving as a spatial supplement to point monitoring.
[0075] S15. Using a unified geodetic coordinate system and time reference as the standard, spatial and temporal interpolation algorithms are used to perform spatiotemporal registration of each sequence to construct a spatiotemporal integrated monitoring dataset.
[0076] Specifically, in embodiment S15, the 2000 National Geodetic Coordinate System is used as the spatial reference and Beijing time is used as the time reference. Kriging spatial interpolation and linear time interpolation are used to unify the four types of data into a spatiotemporal integrated monitoring dataset with a spatial resolution of 5m×5m and a time resolution of 1 hour. The spatial registration error is less than 0.5m and the time alignment error is less than 1 minute, thus eliminating the spatiotemporal misalignment problem between multi-source data.
[0077] By integrating multi-source monitoring data such as underwater topography, bank slope topography, point water level, and satellite remote sensing, and combining big data analysis models to reconstruct a continuous three-dimensional water level field across the entire area, this approach breaks through the spatial limitations of traditional single-point monitoring. It can fully depict the spatial distribution characteristics of water levels along the river, throughout the entire reservoir area, and at the boundary of the bank slope, achieving a leap from discrete point monitoring to three-dimensional perception across the entire area. This significantly improves the accuracy of reservoir capacity calculation and overflow risk assessment, providing complete spatial data support for the refined management and control of hydraulic engineering facilities.
[0078] S2. Input the spatiotemporal integrated monitoring dataset into the sliding spatiotemporal window, and perform spatial difference and temporal statistical feature extraction respectively to generate a multi-dimensional topographic-water level coupled feature vector containing topographic erosion and deposition rate, water level spatial gradient coefficient and data confidence weight.
[0079] S2 includes the following steps:
[0080] S21. Set a sliding spatiotemporal window, traverse the underwater topography sequence and the bank slope topography sequence based on the sliding spatiotemporal window, and use the least squares method to fit the temporal change trend of the elevation of each grid within the window, and use the slope of the trend line as the topography erosion and deposition change rate.
[0081] Specifically, in embodiment S21, the sliding spatiotemporal window length is set to 30 days and the step size is 1 day. The elevation time series data of each grid is traversed, and the least squares method is used to perform linear regression fitting. The covariance between the time variable and the elevation variable within the window is calculated, and divided by the variance of the time variable, the slope of the regression line is obtained, which is the topographic erosion and deposition change rate of each grid, in cm / month; positive values represent deposition, and negative values represent erosion, which quantitatively reflects the dynamic evolution rate of each topographic unit.
[0082] S22. Within the sliding time-space window, calculate the spatial gradient changes of water level data along the longitudinal and transverse directions of the river channel, and combine them with the topographic slope characteristics of the corresponding location for weighted correction to obtain the spatial gradient coefficient of water level.
[0083] Specifically, in embodiment S22, the spatial difference algorithm is used to calculate the gradient values of water level along the longitudinal direction of the river channel and perpendicular to the transverse direction of the river channel, and then the slope of the bank at the corresponding location is superimposed for weighted correction to obtain the spatial gradient coefficient of water level; this coefficient represents the steepness of the spatial distribution of water level and provides trend constraints for subsequent whole-area water level interpolation.
[0084] S23. Extract the data integrity and historical average deviation of each data source within the current sliding window, and calculate the data confidence weight of each data source.
[0085] Specifically, in embodiment S23, the data from point water level stations has the highest completeness and is assigned a baseline weight of 1.0; the weight of underwater topography data is 0.85; the weight of bank slope topography data is 0.8; the weight of remote sensing data is dynamically adjusted based on cloud cover, ranging from 0.3 to 0.7; the data source with higher completeness and smaller historical deviation has a higher confidence weight, providing a weight basis for subsequent multi-source fusion.
[0086] S24. Normalize and concatenate the topographic erosion and sedimentation rate, water level spatial gradient coefficient, data confidence weight, and the current benchmark topographic value and benchmark water level value to obtain a multidimensional topographic-water level coupled feature vector.
[0087] Specifically, in embodiment S24, the Min-Max normalization method is used to uniformly map the seven dimensions of features, namely topographic erosion and sedimentation rate, water level longitudinal gradient, water level lateral gradient, topographic data confidence, remote sensing data confidence, benchmark topographic value, and benchmark water level value, to the [0,1] interval, and splice them to form a 1×7-dimensional multidimensional topographic-water level coupled feature vector, which comprehensively represents the current topographic state, water level state and its dynamic evolution trend in the region.
[0088] S3. Input the multi-dimensional topography-water level coupled feature vector into the trained water level three-dimensional monitoring big data analysis model, and combine it with the current benchmark topography and water level data to obtain the prediction results of the three-dimensional water level field in the target area for future periods and the topography evolution correction amount.
[0089] S3 includes the following steps:
[0090] S31. The multidimensional topographic-water level coupled feature vector is used as the input sequence and input into the input layer of the water level three-dimensional monitoring big data analysis model.
[0091] The training process of the S31 water level three-dimensional monitoring big data analysis model includes:
[0092] S311. Obtain spatiotemporal integrated monitoring datasets of hydraulic facilities under different historical hydrological conditions, as well as corresponding measured three-dimensional water level distribution and topographic mapping records, and construct a historical topographic-water level evolution sample library.
[0093] S312. Divide the historical topography-water level evolution sample database into training set and validation set according to the chronological order, set the model network layer, number of hidden nodes and learning rate parameters, and build an initial big data analysis model.
[0094] S313. Input the training set into the initial model and use the weighted sum of the mean square error of the three-dimensional water level field prediction and the mean absolute error of the terrain evolution correction as the comprehensive loss function.
[0095] S314. The gradient of the comprehensive loss function is calculated using an adaptive optimization algorithm, and the model weights are updated in reverse until the value of the comprehensive loss function on the validation set no longer decreases. The weights are then saved to obtain the trained water level three-dimensional monitoring big data analysis model.
[0096] Specifically, in embodiment S31, the water level three-dimensional monitoring big data analysis model adopts a network structure of "LSTM temporal coding + spatial convolutional decoding", including two LSTM coding layers (64 neurons per layer, Dropout rate 0.2), one two-dimensional convolutional decoding layer (32 convolutional kernels, kernel size 3×3), and two parallel output layers: a spatial reconstruction regression layer and a terrain correction branch. Historical topographic and water level data of the reservoir during the flood and dry seasons of the past 10 years were collected, constructing a total of 120,000 valid samples to form a historical topographic-water level evolution sample library; these samples were arranged in chronological order. The dataset was divided into training and validation sets in a 7:3 ratio, with an initial learning rate of 0.001. The Adam adaptive optimizer was used, and the batch size was 32. The overall loss function was set as follows: Loss = 0.6 × Mean Squared Error (MSE) of the three-dimensional water level field + 0.4 × Mean Absolute Error (MAE) of the terrain correction, balancing water level prediction accuracy and terrain correction accuracy. An early stopping mechanism was triggered when the overall loss of the validation set did not decrease for 20 consecutive rounds, thus saving the optimal model weights. After training, the average relative error of water level prediction on the validation set was 4.2%, and the average error of terrain correction was 3.5 cm.
[0097] The reason for choosing a deep learning model that integrates temporal memory units and spatial convolutional layers in this step is that topographic erosion and water level changes have strong spatiotemporal coupling, depending on historical temporal patterns and constrained by spatial topography.
[0098] S32. Utilize the temporal memory unit of the water level three-dimensional monitoring big data analysis model, combine historical topographic evolution and water level change records, filter short-term occasional disturbance features, and output a hidden layer feature vector containing spatiotemporal evolution patterns.
[0099] S33. Input the hidden layer feature vector into the space to reconstruct the regression layer, and output the prediction result of the continuous three-dimensional water level field in the whole domain within the future target time period through nonlinear mapping.
[0100] S34. Input the hidden layer feature vector into the terrain correction branch, and calculate the terrain evolution correction amount of each grid in the corresponding time period by combining the historical erosion and deposition patterns.
[0101] Specifically, in actual operation, after the model inputs the current coupled feature vector, the LSTM temporal memory unit will retrieve the historical scouring and sedimentation patterns and water level patterns of the same period, filter out short-term occasional disturbances such as single flood discharge and local rainstorms, and output hidden layer feature vectors containing spatiotemporal evolution patterns; finally, two sets of results are output: one is the predicted value of the global continuous three-dimensional water level field with a resolution of 5m×5m for the next 30 days, and the other is the topographic elevation correction amount of each grid for the corresponding time period, in cm, providing dual basis for subsequent three-dimensional monitoring and correction.
[0102] S4. Based on the three-dimensional water level field prediction results, topographic evolution correction amount and current point monitoring data, generate a three-dimensional monitoring correction parameter set including spatial interpolation correction coefficients and topographic compensation parameters.
[0103] S4 includes the following steps:
[0104] S41. Based on the topographic evolution correction amount and the water surface area corresponding to the current water level, calculate the water level-reservoir capacity sensitivity coefficient to obtain the topographic compensation parameters for water level monitoring, which are used to correct the water level benchmark deviation caused by scouring and silting changes.
[0105] Specifically, in embodiment S41, the water level-reservoir capacity sensitivity coefficient is calculated based on the terrain evolution correction amount of each grid and the water surface area corresponding to the current water level, so as to obtain the terrain compensation parameters for each grid; the compensation parameters for siltation areas are positive, which corrects the reservoir capacity calculation deviation caused by siltation and ensures that the water level-water volume relationship matches the actual terrain.
[0106] S42. Based on the spatial deviation between the three-dimensional water level field prediction results and the point water level data, dynamically adjust the variation function parameters of the spatial interpolation to obtain the spatial interpolation correction coefficient, which is used to optimize the spatial interpolation accuracy of the water level across the entire area.
[0107] Specifically, in embodiment S42, based on the spatial deviation between the predicted three-dimensional water level field and the measured value at the point, the parameters of the spherical variogram function (gold nugget value, sill value, and range) of the Kriging interpolation are dynamically adjusted to obtain the spatial interpolation correction coefficient, so that the interpolation result is more in line with the real spatial distribution law of water level.
[0108] S43. Combine terrain compensation parameters and spatial interpolation correction coefficients to generate a set of three-dimensional monitoring correction parameters.
[0109] Specifically, the calibration parameter set includes grid-by-grid terrain compensation values and global spatial interpolation correction coefficients, which are used to correct terrain baseline deviations and optimize spatial interpolation accuracy, respectively.
[0110] By establishing a linkage correction mechanism between dynamic topographic changes and water level monitoring, the topographic scouring and sedimentation change rate is extracted, the topographic evolution correction amount is calculated, and the water level monitoring benchmark is corrected grid by grid. This solves the problem of continuous attenuation of water level monitoring accuracy caused by sediment scouring and sedimentation and bank collapse during the long-term operation of hydraulic facilities. It ensures that the water level-reservoir capacity and water level-flow relationship always match the actual topographic state, effectively guaranteeing the long-term accuracy and reliability of water level monitoring results. It can be adapted to complex scouring and sedimentation scenarios such as rivers with high sediment content and reservoirs with high sediment content.
[0111] S5. Based on the three-dimensional monitoring correction parameter set, the system generates a continuous three-dimensional water level field distribution result across the entire region through a multi-source data collaborative reconstruction algorithm, outputs the three-dimensional water level monitoring data of the target area, and continuously acquires the feedback monitoring data for the next cycle.
[0112] S5 includes the following steps:
[0113] S51. Input the three-dimensional monitoring correction parameter set into the multi-source data collaborative reconstruction module, and establish a reconstruction equation with the dual objectives of minimizing the measured deviation of the point and optimizing the spatial continuity and smoothness of the water level field, with data confidence weight as a constraint.
[0114] S52. Solve the reconstruction equation. First, combine the terrain compensation parameters to correct the water level benchmark grid by grid. Then, substitute the corrected interpolation parameters and generate the global continuous three-dimensional water level field distribution result through a weighted spatial interpolation algorithm.
[0115] Specifically, in embodiments S51 and S52, the reconstruction process uses the measured water level at the point as a hard constraint, the remote sensing inverted water level as a spatial reference, and the topographic compensation parameters as a benchmark for correction. The optimal global water level field is solved by a constrained weighted kriging algorithm. First, the topographic compensation parameters are superimposed grid by grid to correct the water level benchmark deviation. Then, the corrected variogram parameters are substituted into the grid for spatial interpolation to finally obtain the true water level elevation of each grid point, forming a continuous three-dimensional water level distribution.
[0116] S53. Outputs three-dimensional water level monitoring data including grid-by-grid water level elevation, water level along the cross section, and inundation boundary along the shore, and simultaneously generates derived calculation results for reservoir capacity and flow capacity.
[0117] Specifically, in embodiment S53, the system not only outputs grid-by-grid water level elevation data, but also generates upstream-midstream-dam front water level cross-sectional curves and inundation boundary lines along the riverbank; simultaneously, based on the corrected topography and water level field, it calculates the current real-time reservoir capacity, flow capacity of each cross-section, and other derived results, providing direct support for flood control scheduling and water resource management.
[0118] S54. Collect multi-source topographic and water level data for the next monitoring cycle at a fixed period, and use the data as feedback monitoring data input for the closed-loop verification process.
[0119] S6. Calculate the spatiotemporal deviation between the feedback monitoring data and the three-dimensional water level field prediction results, and output the corresponding monitoring accuracy assessment results based on the spatiotemporal deviation.
[0120] S6 includes the following steps:
[0121] S61. Extract the measured three-dimensional water level distribution data for the next cycle, compare it with the three-dimensional water level field prediction results for the corresponding time period grid by grid, and calculate the average relative error of the whole domain as the spatiotemporal deviation.
[0122] S62. Set the accuracy warning threshold. If the spatiotemporal deviation is greater than the accuracy warning threshold, output an abnormal monitoring accuracy warning. If the spatiotemporal deviation is not greater than the accuracy warning threshold, output a qualified monitoring accuracy indicator and archive and store the current three-dimensional water level monitoring data.
[0123] Specifically, in embodiment S62, the accuracy warning threshold is set to 8%. This threshold is set based on the error statistics of historical verification data and is used to distinguish between normal prediction fluctuations and abnormal deviations. If the spatiotemporal deviation is greater than 8%, an abnormal monitoring accuracy warning is pushed to the operation and maintenance platform to prompt verification of terrain changes or equipment failures in the corresponding area. If the spatiotemporal deviation is not greater than 8%, the monitoring accuracy is determined to be qualified, a qualified identifier is output, and the data is archived to the historical database as a valid sample for subsequent analysis.
[0124] S63. Statistically analyze the deviation distribution by region and time period, generate a monitoring accuracy assessment report, and record the monitoring error characteristics of different regions and time periods.
[0125] By constructing a big data monitoring system with closed-loop accuracy verification, the system enables full-process accuracy control of monitoring results based on spatiotemporal deviation calculation. It can output accuracy anomaly warnings and qualified archiving results in a tiered manner, providing quantitative guidance for monitoring and maintenance. By relying on big data technology to fully explore the coupling value of multi-source monitoring data, it can achieve coordinated monitoring of topography and water level, improving the high precision, three-dimensionality, and intelligence of water level monitoring.
[0126] Specifically, the deviation distribution is statistically analyzed in three zones: upstream, midstream, and upstream of the dam. Monitoring accuracy assessment reports are generated daily, monthly, and quarterly, recording the spatiotemporal distribution characteristics of the error to provide data support for optimizing the monitoring scheme.
[0127] For example, consider the scenario of scouring and silting in a reservoir with high sediment content during the flood season. The reservoir has a total capacity of 1.2 billion cubic meters and a main stream length of 68 km. It is affected by sediment scouring and silting from the Yellow River tributaries all year round. Traditional monitoring only sets up three fixed water level stations in front of the dam, in the middle reaches and upstream. The reservoir capacity error calculated based on the initial topography during the construction period is often more than 11%.
[0128] At time T0, the system completes the preliminary data collection and preprocessing, generates a spatiotemporal integrated dataset, and extracts that the scouring and silting change rate of multiple siltation areas in the reservoir is positive and continues to increase, and the spatial gradient coefficient of water level rises along the way.
[0129] At time T1, the coupled feature vector is input into the big data analysis model. The model combines historical flood season scouring and deposition patterns to predict that the topography of the sedimentation zone in the middle reaches of the reservoir will further rise in the next 30 days. Under the same inflow, the water level in this area will be higher than the water level in front of the dam. The three-dimensional water level field shows significant differences along the route. Moreover, if the topography is not corrected, the reservoir capacity calculation will be 12% too large.
[0130] At time T2, the system generates the terrain compensation parameters and interpolation correction coefficients for the corresponding grid, and generates the corrected three-dimensional water level field through a multi-source collaborative reconstruction algorithm;
[0131] At time T30, the actual measurement data showed that the corrected three-dimensional water level field matched the actual measurement by more than 95%, and the reservoir capacity calculation error was reduced from 12% in the traditional method to 3.7%. The system's spatiotemporal deviation was 3.2%, which is no more than the accuracy warning threshold of 8%. The system output a qualified monitoring accuracy mark and archived the monitoring data into the database.
[0132] Example 2:
[0133] Please see Figures 1-2 A big data-based three-dimensional water level monitoring system for hydraulic structures is used to implement a big data-based three-dimensional water level monitoring method for hydraulic structures. The system includes:
[0134] The multi-source water level and topographic data acquisition and preprocessing module is used to synchronously access the raw data from multiple sources, complete noise filtering, correction and spatiotemporal registration, and output a spatiotemporal integrated monitoring dataset.
[0135] The topographic-water level coupling feature extraction module is used to extract the scouring and sedimentation rate, spatial gradient coefficient and confidence weight through sliding spatiotemporal window and statistical calculation, and generate a multidimensional topographic-water level coupling feature vector.
[0136] The three-dimensional water level prediction and terrain correction module has a built-in three-dimensional water level monitoring big data analysis model, which is used to output the three-dimensional water level field prediction results and terrain evolution correction.
[0137] The water level field reconstruction parameter optimization module is used to generate a three-dimensional monitoring correction parameter set and solve the reconstruction equation to obtain a continuous three-dimensional water level field over the entire domain.
[0138] The water level monitoring closed-loop verification module is used to calculate the spatiotemporal deviation between the predicted results and the measured data, and output the accuracy assessment results and anomaly warnings.
[0139] 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 three-dimensional monitoring method for water levels in hydraulic structures based on big data, characterized in that, The method includes the following steps: S1. Simultaneously acquire multi-source topographic and water level raw data of the area covered by hydraulic facilities, perform spatiotemporal benchmark alignment and outlier removal preprocessing, and generate a spatiotemporal integrated monitoring dataset containing underwater topographic sequence, bank slope topographic sequence, point water level sequence and remote sensing inversion water level sequence. S2. Input the spatiotemporal integrated monitoring dataset into a sliding spatiotemporal window, and perform spatial difference and temporal statistical feature extraction respectively to generate a multi-dimensional topographic-water level coupled feature vector containing topographic erosion and sedimentation rate, water level spatial gradient coefficient and data confidence weight. S3. Input the multidimensional topography-water level coupled feature vector into the trained three-dimensional water level monitoring big data analysis model, and combine it with the current benchmark topography and water level data to obtain the three-dimensional water level field prediction results and topography evolution correction amount for the target area in the future period. S4. Based on the predicted results of the three-dimensional water level field, the terrain evolution correction amount, and the current monitoring data, generate a set of three-dimensional monitoring correction parameters including spatial interpolation correction coefficients and terrain compensation parameters. S5. Based on the three-dimensional monitoring correction parameter set, generate a continuous three-dimensional water level field distribution result across the entire domain through a multi-source data collaborative reconstruction algorithm, output the three-dimensional water level monitoring data of the target area, and continuously acquire the feedback monitoring data for the next cycle. S6. Calculate the spatiotemporal deviation between the feedback monitoring data and the three-dimensional water level field prediction result, and output the corresponding monitoring accuracy evaluation result based on the spatiotemporal deviation.
2. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 1, characterized in that, S1 includes the following steps: S11. Collect underwater topographic point cloud data of the water area of the hydraulic engineering facility, remove suspended solids and sounding noise by combined filtering, and obtain a gridded underwater topographic sequence by spatial interpolation. S12. Collect topographic elevation point cloud data of the bank slope and surrounding land area, and obtain a gridded bank slope topography sequence with the same reference as the underwater topography after filtering by ground features and fitting and correcting the elevation of the reference station. S13. Collect point water level monitoring data at fixed cross-sections, combine with real-time meteorological parameters to perform wave velocity compensation correction, and obtain a standardized point water level sequence after sliding filtering. S14. Acquire satellite radar altimeter remote sensing images, extract water body boundaries after radiometric and geometric fine correction, and obtain remote sensing inverted water level sequences by combining them with bank slope topography inversion. S15. Using a unified geodetic coordinate system and time reference as the standard, spatial interpolation and temporal interpolation algorithms are used to perform spatiotemporal registration of each sequence to construct the spatiotemporal integrated monitoring dataset.
3. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 2, characterized in that, S2 includes the following steps: S21. Set a sliding time-space window, traverse the underwater topography sequence and the bank slope topography sequence based on the sliding time-space window, and use the least squares method to fit the temporal change trend of the elevation of each grid within the window, and use the slope of the trend line as the topography erosion and deposition change rate. S22. Within the sliding time-space window, calculate the spatial gradient change of water level data along the longitudinal and transverse directions of the river channel, and combine it with the topographic slope characteristics of the corresponding location for weighted correction to obtain the spatial gradient coefficient of water level. S23. Extract the data integrity and historical average deviation of each data source within the current sliding window, and calculate the data confidence weight of each data source. S24. Normalize and concatenate the topographic erosion and sedimentation rate, water level spatial gradient coefficient, data confidence weight, and the current benchmark topographic value and benchmark water level value to obtain a multidimensional topographic-water level coupled feature vector.
4. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 3, characterized in that, S3 includes the following steps: S31. The multidimensional topography-water level coupling feature vector is used as an input sequence and input into the input layer of the water level three-dimensional monitoring big data analysis model. S32. Using the temporal memory unit of the water level three-dimensional monitoring big data analysis model, combined with historical topographic evolution and water level change records, short-term occasional disturbance features are filtered out, and a hidden layer feature vector containing spatiotemporal evolution patterns is output. S33. Input the hidden layer feature vector into the space to reconstruct the regression layer, and output the prediction result of the continuous three-dimensional water level field in the whole domain within the future target time period through nonlinear mapping. S34. Input the hidden layer feature vector into the terrain correction branch, and calculate the terrain evolution correction amount of each grid in the corresponding time period by combining the historical erosion and deposition patterns.
5. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 4, characterized in that, The training process of the water level three-dimensional monitoring big data analysis model in S31 includes: S311. Obtain spatiotemporal integrated monitoring datasets of hydraulic facilities under different historical hydrological conditions, as well as corresponding measured three-dimensional water level distribution and topographic mapping records, and construct a historical topographic-water level evolution sample library. S312. Divide the historical topography-water level evolution sample database into training set and validation set according to the chronological order, set the model network layer, number of hidden nodes and learning rate parameters, and construct the initial big data analysis model. S313. Input the training set into the initial model, and use the weighted sum of the mean square error of the three-dimensional water level field prediction and the mean absolute error of the terrain evolution correction as the comprehensive loss function. S314. The gradient of the comprehensive loss function is calculated using an adaptive optimization algorithm, and the model weights are updated in reverse until the value of the comprehensive loss function on the validation set no longer decreases. The weights are then saved to obtain the trained water level three-dimensional monitoring big data analysis model.
6. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 4, characterized in that, S4 includes the following steps: S41. Based on the terrain evolution correction amount, and combined with the water surface area corresponding to the current water level, calculate the water level-reservoir capacity sensitivity coefficient to obtain the terrain compensation parameters for water level monitoring, which are used to correct the water level benchmark deviation caused by scouring and silting changes. S42. Based on the spatial deviation between the three-dimensional water level field prediction result and the point water level data, dynamically adjust the variation function parameters of the spatial interpolation to obtain the spatial interpolation correction coefficient, which is used to optimize the spatial interpolation accuracy of the water level across the entire area. S43. Combine the terrain compensation parameters with the spatial interpolation correction coefficients to generate a three-dimensional monitoring correction parameter set.
7. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 6, characterized in that, S5 includes the following steps: S51. Input the three-dimensional monitoring correction parameter set into the multi-source data collaborative reconstruction module, and establish a reconstruction equation with the dual objectives of minimizing the measured deviation of the point and optimizing the spatial continuity and smoothness of the water level field, with data confidence weight as a constraint. S52. Solve the reconstruction equation. First, combine the terrain compensation parameters to correct the water level benchmark grid by grid. Then, substitute the corrected interpolation parameters and generate the global continuous three-dimensional water level field distribution result through a weighted spatial interpolation algorithm. S53. Outputs three-dimensional water level monitoring data including grid-by-grid water level elevation, water level along the cross section, and inundation boundary along the shore, and simultaneously generates derived calculation results for reservoir capacity and flow capacity. S54. Collect multi-source topographic and water level data for the next monitoring cycle at a fixed period, and use the data as feedback monitoring data input for the closed-loop verification process.
8. The method for three-dimensional monitoring of water levels in hydraulic structures based on big data according to claim 7, characterized in that, S6 includes the following steps: S61. Extract the measured three-dimensional water level distribution data for the next cycle, compare it with the three-dimensional water level field prediction results for the corresponding time period grid by grid, and calculate the average relative error of the whole domain as the spatiotemporal deviation. S62. Set an accuracy warning threshold. If the spatiotemporal deviation is greater than the accuracy warning threshold, output an abnormal monitoring accuracy warning. If the spatiotemporal deviation is not greater than the accuracy warning threshold, output a qualified monitoring accuracy indicator and archive and store the current three-dimensional water level monitoring data. S63. Statistically analyze the deviation distribution by region and time period, generate a monitoring accuracy assessment report, and record the monitoring error characteristics of different regions and time periods.
9. A three-dimensional water level monitoring system for hydraulic structures based on big data, used to implement the three-dimensional water level monitoring method for hydraulic structures based on big data as described in any one of claims 1-8, characterized in that, The system includes: The multi-source water level and topographic data acquisition and preprocessing module is used to synchronously access the raw data from multiple sources, complete noise filtering, correction and spatiotemporal registration, and output a spatiotemporal integrated monitoring dataset. The topographic-water level coupling feature extraction module is used to extract the scouring and sedimentation rate, spatial gradient coefficient and confidence weight through sliding spatiotemporal window and statistical calculation, and generate a multidimensional topographic-water level coupling feature vector. The three-dimensional water level prediction and terrain correction module has a built-in three-dimensional water level monitoring big data analysis model, which is used to output the three-dimensional water level field prediction results and terrain evolution correction. The water level field reconstruction parameter optimization module is used to generate a three-dimensional monitoring correction parameter set and solve the reconstruction equation to obtain a continuous three-dimensional water level field over the entire domain. The water level monitoring closed-loop verification module is used to calculate the spatiotemporal deviation between the predicted results and the measured data, and output the accuracy assessment results and anomaly warnings.
Citation Information
Patent Citations
Groundwater recharge inversion system of irrigation and drainage dual-purpose lining channel based on water level monitoring
CN116698151B
