Intelligent water conservancy digital twin simulation system based on multi-source data

Abstract

The invention provides an intelligent water conservancy digital twinborn simulation system based on multi-source data, and belongs to the technical field of digital monitoring. Minute-level data acquisition and transmission are realized by constructing a space-air-ground three-dimensional sensing network and combining edge intelligent preprocessing, data acquisition, noise reduction and abnormity identification are realized by utilizing sensors such as millimeter wave radar and laser radar, and the system is used for realizing data acquisition and transmission. Dynamic data fusion and intelligent calibration are carried out, and second-level alignment and credible verification of data are realized by means of a space-time calibration algorithm, Kalman filtering and a block chain evidence storage technology; a hybrid simulation and intelligent decision model is established, a physical model, a machine learning architecture and a dynamic threshold decision tree are adopted, flood routing minute-level prediction and emergency response are realized, and the system improves water conservancy monitoring prediction precision and emergency response efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital monitoring, and more specifically, it is a smart water conservancy digital twin simulation system based on multi-source data. Background Art

[0002] In the field of smart water conservancy, there are many technical bottlenecks in the existing digital twin simulation systems during practical applications.

[0003] On the one hand, the traditional data acquisition network relies on a single sensor. In the face of complex water conservancy scenarios, the data acquisition efficiency is low and the noise interference is serious. The burst pulse noise and Gaussian white noise lead to a decline in data quality, and the ability to identify invalid data and abnormal data is insufficient, making it difficult to achieve minute-level data acquisition and transmission.

[0004] On the other hand, there are problems of time asynchrony and spatial heterogeneity in the fusion of multi-source heterogeneous data. The clock deviation and coordinate system differences of lidar, millimeter-wave radar, and camera devices result in inaccurate data alignment, affecting subsequent analysis.

[0005] In addition, the physical model calculation efficiency and robustness defects are significant. Traditional models such as the Saint-Venant equation have a slow calculation process, unable to meet the minute-level prediction requirements of flood evolution, and the model depends on parameters such as roughness and bottom slope, and is easily affected by the environment to produce errors. At the decision-making level, the fixed threshold strategy is difficult to adapt to complex flood scenarios, the visualization effect of flood evolution data is poor, and the abstract data affects the decision-making efficiency. At the same time, the system computing power scheduling lacks flexibility, unable to dynamically adjust according to flood evolution, the fault assessment and self-healing mechanism is imperfect, and the data transmission continuity and reliable verification ability are insufficient under extreme weather conditions, seriously restricting the emergency response and precise management of smart water conservancy.

[0006] Therefore, the present invention provides a smart water conservancy digital twin simulation system based on multi-source data. Summary of the Invention

[0007] In order to make up for the deficiencies of the existing technology and solve at least one of the technical problems proposed in the background art.

[0008] The technical solution adopted by the present invention to solve its technical problems is as follows: In the first aspect, the present invention provides a smart water conservancy digital twin simulation system based on multi-source data, including: Edge intelligent preprocessing module: constructing an air-space-ground three-dimensional perception network, and realizing minute-level data acquisition and transmission in combination with edge intelligent preprocessing; Trusted verification module: Align the data collected by the air-space-ground stereo perception network in terms of time consistency, calibrate the spatial coordinates of the data collected by the air-space-ground stereo perception network using a spatio-temporal calibration algorithm, and implement the dynamic fusion of the physical model and the spatially calibrated data using digital twin-driven Kalman filtering to improve the prediction accuracy of the physical model; Intelligent decision-making module: Construct a hybrid model based on long short-term memory network and graph neural network to solve the prediction error of the physical model, and use a lightweight architecture of physical model and machine learning, dynamic threshold decision tree, and digital twin visualization technology to achieve minute-level prediction of flood evolution; System self-healing module: Construct an elastic computing power and self-healing mechanism, and ensure the dynamic scheduling of system computing power and fault self-healing through edge-cloud collaborative computing and Bayesian network fault assessment.

[0009] As a further improvement scheme: The specific process of constructing the air-space-ground stereo perception network is as follows: Synchronously deploy millimeter-wave radars at river cross-sections and water level monitoring points, deploy lidar within the flooded area of the river and the basin topography, and deploy 4K high-definition cameras at the water edge line of the river and the overtopping identification area.

[0010] As a further improvement scheme: The specific process of combining edge intelligence preprocessing to achieve minute-level data collection and transmission is as follows: Deploy an AI edge box at the sensor terminal to optimize the data layer quality; Use Daubechies-4 wavelet to decompose the time-series data collected by the air-space-ground stereo perception network into 5 layers to balance the computational complexity and denoising effect. Use soft thresholding shrinkage to avoid signal mutations of hard thresholding. The soft threshold is adaptively calculated through Stein's unbiased risk estimate to identify and process invalid and abnormal data in the data collected by the air-space-ground stereo perception network; Use the isolation forest anomaly detection method to calculate the anomaly score of the data collected by the air-space-ground stereo perception network to identify sensor drift. Use the PCA feature compression method to perform principal component analysis on the data collected by the air-space-ground stereo perception network, remove mutant abnormal data, and retain effective principal components; The AI edge box runs a random forest model to monitor the data collected by the air-space-ground stereo perception network after optimizing the data layer quality in real time. When the water level exceeds the warning value, it triggers an audible and visual warning to avoid uploading invalid data to the cloud.

[0011] As a further improvement scheme: The specific process of aligning the data collected by the air-space-ground stereo perception network in terms of time consistency is as follows: Using the Beidou-3 PPS second pulse signal as the time reference, calculate the deviation between the clocks of the lidar, millimeter-wave radar, and 4K high-definition camera and UTC. Suppress noise through moving window averaging, synchronize the lidar, millimeter-wave radar, and 4K high-definition camera to UTC, ensure time consistency, and achieve precise alignment at the second level.

[0012] As a further improvement: The specific process of using the spatio-temporal calibration algorithm to perform spatial calibration on the data coordinates collected by the air-ground-space stereo perception network is as follows: Convert polar coordinates to Cartesian coordinates. The specific conversion process is as follows: Conversion of the abscissa: , conversion of the ordinate: , where is the distance between the lidar and the section, is the angle between the radar beam and the water flow direction; after conversion, it is unified with the CGCS2000 coordinate system of the ground sensor; Convert the pixel coordinates of the 4K high-definition camera to geodetic coordinates, and perform the conversion through camera calibration and the Bursa seven-parameter transformation: , where , , are the translation parameters, is the scaling factor, , , are the rotation angles, which are calibrated by field control points.

[0013] As a further improvement: The specific process of using digital twin-driven Kalman filtering to achieve dynamic fusion of the physical model and the spatially calibrated data and improve the prediction accuracy of the physical model is as follows: Based on the linearization and discretization of the Saint-Venant equations, construct the state vector and state equation respectively. Since there are errors in the physical model prediction, it is necessary to use the real-time data collected by the air-ground-space stereo perception network for correction. As a fusion tool for model prediction and real-time observation, Kalman filtering can, while retaining the mechanism of the physical model, use sensor data to calibrate the state deviation in real time. Its core is achieved through the construction of the observation equation and two-step iterative update. The observation equation establishes the mapping between the system state and the real-time data collected by the air-ground-space stereo perception network; If the accuracy of the air-ground-space stereo perception network is high, the observation data dominates the correction. If the physical model is reliable, the model prediction dominates the correction.

[0014] As a further improvement: The specific process of constructing a hybrid model based on long short-term memory network and graph neural network to solve the physical model prediction error is as follows: Construct a hybrid model of LSTM (Long Short-Term Memory) network and GNN (Graph Neural Network). Among them, the LSTM network is used to process the time series of data collected by the space-air-ground three-dimensional perception network to capture the time dependence of flood evolution, and the GNN processes the spatial topology to capture the spatial propagation of flood evolution. Use the simulation results of the physical model of historical flood events and actual monitoring data as the training set to train the model. The loss function uses the mean absolute error, and the formula is: , where is the actual monitored water level, is the water level predicted by the model, is the sample size; When the sudden change of the downstream water level is caused by the flood discharge of the upstream reservoir, the LSTM-GNN model can quickly identify the change of the relationship between flow and water level and correct the prediction error of the physical model.

[0015] As a further improvement scheme: the specific process of realizing the minute-level prediction of flood evolution is as follows: Extract real-time features, where the real-time features include: current water level, water level rising rate, flow rate, rainfall, and historical features. The historical features are the evolution paths of similar flood events; Use the CART decision tree to train the relationship between real-time features and thresholds, and update the dynamically adjusted dynamic thresholds in real time. Use LiDAR point cloud and InSAR interferometric measurement to construct the three-dimensional terrain of the basin, and use the Shader to simulate the flow, reflection, and inundation effects of floods. Based on the water level data calculated by the physical model, update the water surface height and flow velocity vector in real time to realize the minute-level prediction of flood evolution.

[0016] As a further improvement scheme: the specific process of constructing elastic computing power is as follows: By having the edge side undertake real-time lightweight tasks and the cloud side handle batch calculations, flexible scheduling of computing power is achieved. The cloud side undertakes the flood evolution simulation of the entire basin and the training of machine learning models, and dynamically adjusts strategies. The edge side is triggered by the load threshold and offloads tasks to the cloud side.

[0017] As a further improvement scheme: the specific process of fault assessment is as follows: Construct a Bayesian network structure, define the nodes as devices and functions, use historical fault data to calculate the conditional probability table, and the edge side collects the state of the space-air-ground three-dimensional perception network in real time as evidence input to the Bayesian network. Through joint probability inference, calculate the probability of fault occurrence.

[0018] In the second aspect, the present invention provides a smart water conservancy digital twin simulation method based on multi-source data, including: S1: Construct a space-air-ground three-dimensional perception network, and combine edge intelligence preprocessing to achieve minute-level data collection and transmission; S2: Align the data collected by the space-air-ground three-dimensional perception network in terms of time consistency, perform spatial calibration on the data coordinates collected by the space-air-ground three-dimensional perception network using a spatio-temporal calibration algorithm, and implement dynamic fusion of the physical model and the spatially calibrated data using digital-twin-driven Kalman filtering to improve the prediction accuracy of the physical model; S3: Construct a hybrid model based on long short-term memory network and graph neural network to solve the prediction error of the physical model, and use a lightweight architecture of physical model and machine learning, a dynamic threshold decision tree, and digital-twin visualization technology to achieve minute-level prediction of flood evolution; S4: Construct an elastic computing power and self-healing mechanism, and ensure dynamic scheduling of system computing power and self-healing of faults through edge-cloud collaborative computing and Bayesian network fault assessment.

[0019] The beneficial effects of the present invention are as follows: Through the space-air-ground three-dimensional perception network and edge intelligent preprocessing, minute-level data collection and transmission are realized. Multi-source sensors such as millimeter-wave radar and lidar improve the comprehensiveness of data, wavelet denoising and anomaly detection ensure data quality, and the Beidou / 5G dual link ensures the continuity of data in extreme weather. Using the spatio-temporal calibration algorithm and Kalman filtering, the spatio-temporal asynchrony problem of multi-source data is solved, second-level alignment is achieved, and the combination of blockchain evidence storage ensures the credibility and traceability of data. The hybrid simulation model accelerates the calculation of the Saint-Venant equation and corrects the roughness error through the fusion of the physical model and machine learning. The dynamic threshold decision tree adapts to complex flood scenarios, and digital-twin visualization improves decision-making efficiency. The elastic computing power mechanism realizes edge-cloud collaborative scheduling, and Bayesian network fault assessment ensures system self-healing, overall improving the accuracy of water conservancy monitoring and prediction and the emergency response ability. Description of the Drawings

[0020] The present invention will be further described below with reference to the drawings.

[0021] Figure 1 It is the system module diagram of Embodiment 1 of the present invention; Figure 2 It is the step flow diagram of Embodiment 2 of the present invention. Detailed Embodiments

[0022] In order to make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0023] Embodiment 1 As Figure 1 shown, the intelligent water conservancy digital-twin simulation system based on multi-source data described in the embodiment of the present invention includes: Edge intelligent preprocessing module: Construct a space-air-ground three-dimensional perception network, and realize minute-level data collection and transmission in combination with edge intelligent preprocessing; Specifically, the specific process of constructing the space-air-ground three-dimensional perception network is as follows: On the basis of traditional sensors, such as water level gauges and rain gauges, other types of sensors are added. For example, millimeter-wave radars are synchronously arranged at river cross-sections and water level monitoring points, lidar is arranged within the flooded area of the river and the basin terrain, and 4K high-definition cameras are arranged at the water edge line and overtopping identification area of the river; Among them, the millimeter-wave radar at the river cross-section and water level monitoring point is used to measure the water flow velocity, and the water flow velocity is calculated through the Doppler frequency shift formula. The specific formula is: , where is the Doppler frequency shift, is the water flow velocity, is the angle between the radar beam and the water flow direction, is the radar wavelength; The lidar is used to construct a three-dimensional model of the water surface. The three-dimensional model of the water surface is constructed through a point cloud fitting algorithm. The specific formula is: , where is the point cloud weight, is the i-th lidar point cloud data, is the elevation; The 4K high-definition camera is used to identify the water edge line and overtopping, and the U-Net image segmentation model is used to extract the water boundary; Build a Beidou-3 short message and 5G-MEC dual link to optimize the transmission layer delay. The edge node is deployed within 5 kilometers of the water conservancy facility, and the transmission is optimized through an end-to-end delay model. The specific formula is: , where is the transmission delay, is the edge processing delay, is the queue delay; When the 5G signal strength is less than -110 dBm, it automatically switches to the Beidou-3 short message, and the Beidou-3 short message protocol is used to transmit key data to ensure the continuity of data transmission in extreme weather; The local networking of sensors and edge nodes is realized through the LoRaWAN protocol. After the edge node aggregates the data, it is uploaded via 5G / Beidou to reduce the air interface occupancy; The specific process of realizing minute-level data acquisition and transmission by combining edge intelligence preprocessing is as follows: Deploy an AI edge box at the sensor terminal to optimize the data layer quality; Since the noise of the data collected by the space-air-ground three-dimensional perception network mainly comes from burst pulse noise and Gaussian white noise, it is necessary to adopt wavelet basis and threshold strategy for noise reduction processing for burst pulse noise and Gaussian white noise; The specific noise reduction process is as follows: The Daubechies-4 wavelet is used to decompose the time series data collected by the air-ground-space stereoscopic perception network into 5 layers, balancing the amount of calculation and the denoising effect. The soft threshold is used to shrink and avoid the signal mutation of the hard threshold. The soft threshold is adaptively calculated through the Stein unbiased risk estimation: ,in, is the i-th original data, is the denoised data of the i-th data, is the noise variance, is the wavelet coefficient; Furthermore, the quality of the data layer is optimized; Identify and process invalid and abnormal data in the data collected by the air-space-ground stereoscopic perception network; Among them, the abnormal data types include: gradual abnormality and sudden abnormality. Gradual abnormality is mainly sensor drift, and sudden abnormality is mainly data jump, such as electromagnetic interference causing invalid value of instantaneous output of lidar; The isolation forest anomaly detection method is used to calculate the anomaly score of the data collected by the air-ground-air stereoscopic perception network and identify sensor drift. The specific calculation formula is: ,in, is the path length of the data collected by the air-space-ground-space stereoscopic perception network in the tree, is the average path length, is the amount of data collected by the air-space-ground stereoscopic perception network, Score for data anomaly; set anomaly values based on the data collected by the historical space-ground stereoscopic perception network. When it is greater than 0.6, it is an outlier; The PCA feature compression method is used to perform principal component analysis on the data collected by the space-ground-air stereoscopic perception network, remove abnormal mutation data, and retain effective principal components. The AI edge box runs a random forest model to monitor the data collected by the air-space-ground-air stereoscopic perception network after data layer quality optimization in real time. When the water level exceeds the warning value, an audible and visual warning is triggered to avoid uploading invalid data to the cloud and shorten the response chain. Trusted Verification Module: Align the data collected by the space-air-ground-earth-space stereoscopic perception network for temporal consistency, use the space-time calibration algorithm to perform spatial calibration on the data coordinates collected by the space-air-ground-earth-space stereoscopic perception network, and use the digital twin-driven Kalman filter to achieve dynamic fusion of the physical model and the spatially calibrated data, thereby improving the prediction accuracy of the physical model. Specifically, the space-time calibration algorithm solves the temporal asynchrony and spatial heterogeneity problems of LiDAR, millimeter-wave radar, and 4K high-definition cameras in the space-ground-air stereoscopic perception network; Using the Beidou-3 PPS second pulse signal as the time reference, calculate the deviation between the clocks of the lidar, millimeter-wave radar, and 4K high-definition camera and UTC. Suppress the noise through moving window averaging, synchronize the lidar, millimeter-wave radar, and 4K high-definition camera to UTC, ensure time consistency, and achieve precise alignment at the second level; Furthermore, use the spatio-temporal calibration algorithm to perform spatial calibration on the data coordinates collected by the air-space-ground three-dimensional perception network; Since the coordinate representation methods of the millimeter-wave radar and 4K high-definition camera in the air-space-ground three-dimensional perception network are different, the lidar and millimeter-wave radar use polar coordinates, and the 4K high-definition camera uses pixel coordinates, which cannot be directly superimposed. Therefore, coordinate system conversion and registration are required; Convert the polar coordinates to Cartesian coordinates. The specific conversion process is as follows: Conversion of the abscissa: , Conversion of the ordinate: , where is the distance between the lidar and the cross-section, is the angle between the radar beam and the water flow direction; after conversion, it is unified with the CGCS2000 coordinate system of the ground sensor; Convert the pixel coordinates of the 4K high-definition camera to geodetic coordinates, and perform the conversion through camera calibration and the Bursa seven-parameter transformation: , where , , is the translation parameter, is the scaling factor, , , is the rotation angle, calibrated through field control points; For the data collected by the air-space-ground three-dimensional perception network after spatial calibration, adopt digital twin-driven Kalman filtering to realize the dynamic fusion of the physical model and the data after spatial calibration, and improve the prediction accuracy of the physical model; Dynamically fuse the physical model prediction and the data after spatial calibration to improve the prediction accuracy of the physical model; Based on the linearization and discretization of the Saint-Venant equation, construct the state vector and state equation respectively; State vector: , where is the water level at time is the flow velocity at time is the time step; State equation: , where is the state transition, and the expression is , is the average water level, is the time step, is the space step, and the linear approximation Saint-Venant continuity equation, is the control matrix, corresponding to the gate opening control quantity , , is the flow velocity opening conversion coefficient, is the process noise, is the system state vector at time is the system state vector at time is the state transition matrix at time is the control input at time There are errors in the physical model prediction, and it is necessary to correct it using the real-time data collected by the space-air-ground three-dimensional perception network. As a fusion tool for model prediction and real-time observation, Kalman filtering can retain the physical model mechanism while using sensor data to calibrate the state deviation in real time. Its core is achieved through the construction of the observation equation and two-step iterative update: The specific processes of the observation equation and Kalman update are as follows: The observation equation establishes the mapping between the system state and the real-time data collected by the space-air-ground three-dimensional perception network. For the core state water level H, flow velocity V, and flow rate Q of the water conservancy system, the observation equation describes the direct relationship between the real-time data collected by the space-air-ground three-dimensional perception network and the system state: The observation equation is updated to: , where , is the physical model prediction object at time is the observation noise, is the data collected by the space-air-ground three-dimensional perception network at time Kalman filtering gradually optimizes the state estimation and reduces the model error through two steps from prior prediction to posterior correction; Use the state transition relationship of physical models such as the Saint-Venant equation to describe the evolution laws of the water level H, flow velocity V, and flow rate Q, and predict the state and uncertainty at the next moment: The Kalman wave iterative update is: Prediction step: , , is the state transition relationship, that is, the evolution laws of the water level H, flow velocity V, and flow rate Q, is the state predicted by the physical model at time is The fusion of the physical model and observational data at a moment, the corrected system state is the uncertainty predicted by the physical model at a moment is the posterior state covariance matrix at a moment is the transpose of the state transition relation and is the process noise covariance matrix, which is used to describe the uncertainty of the physical model prediction; The weights of the model prediction and the real-time data collected by the space-air-ground stereo perception network are balanced through the Kalman gain to correct the state estimation: Update step: , , is the Kalman gain is the uncertainty of the model prediction at a moment is the observation matrix is the observation noise covariance is the posterior state estimation at a moment is the state predicted by the model at a moment is the real-time data collected by the space-air-ground stereo perception network at a moment; If the accuracy of the space-air-ground stereo perception network is high approaches 1, and the observational data dominates the correction; If the physical model is reliable approaches 0, and the model prediction dominates the correction; The posterior state data corrected by the Kalman filter is the core basis for flood evolution simulation and emergency decision-making. It is necessary to conduct full-link evidence preservation on the posterior state data corrected by the Kalman filter to ensure its credibility: In the existing technology, the efficiency of uploading all data to the chain is low, and it is necessary to screen the core data to reduce the amount of data uploaded; First, screen the posterior state data corrected by the Kalman filter. The specific screening rules are as follows: The posterior state data corrected by the Kalman filter is hereinafter referred to as data; Only upload extreme value data, key data after fusion, and abnormal events; Generate a unique hash value for the screened data using the SHA-256 algorithm, and build a water conservancy alliance chain based on Hyperledger Fabric to ensure the traceability and non-tamperability of the data after uploading to the chain; In the water conservancy alliance chain, the nodes include basin management agencies, equipment manufacturers, and third-party inspection agencies; the authorized equipment is: only the sensors within the whitelist can submit data to the chain, and the data of other equipment is automatically marked as unauthorized; When it is necessary to verify whether the data has been tampered with, query the hash value of the data from the blockchain browser and compare it with the hash calculated from the original data stored locally. If the comparison result is consistent, the data has not been tampered with; Intelligent decision-making module: Build a hybrid model based on long short-term memory network and graph neural network to solve the prediction error of the physical model. Adopt the lightweight architecture of physical model and machine learning, dynamic threshold decision tree, and digital twin visualization technology to achieve minute-level prediction of flood evolution; In the existing technology, the calculation process of the Saint-Venant equation is relatively slow and it is difficult to support minute-level response. Use the Preissmann four-point implicit difference to discretize the Saint-Venant equations to solve the problem of slow calculation process of the Saint-Venant equation; The specific calculation formula is: , where, is the cross-sectional area of flow, is the flow rate, is the friction slope, is the bottom slope, is the spatial step, is the gravity, is the time layer, is the spatial node, is the th spatial node, is the time layer; Deploy the matrix operation of the difference equation to the GPU and use the CUDA kernel function for parallel computing to compress the time step of the operation; In the existing technology, the physical model depends on roughness and bottom slope parameters, and the robustness of the model is insufficient, and errors are difficult to avoid; The specific process to solve the error of the physical model is: Build a hybrid model of LSTM long short-term memory network and GNN graph neural network. Among them, the LSTM long short-term memory network is used to process the time series of data collected by the space-air-ground three-dimensional perception network to capture the time dependence of flood evolution, and the GNN processes the spatial topology to capture the spatial propagation of flood evolution; Use the simulation results of the physical model of historical flood events and actual monitoring data as the training set to train the model. The loss function uses the mean absolute error, and the formula is: , where, is the actual monitored water level, is the water level predicted by the model, is the sample size; Update the model parameters using the data collected by the latest space-air-ground three-dimensional perception network, and correct the roughness error of the physical model. When the sudden change of the downstream water level is caused by the flood discharge of the upstream reservoir, the LSTM-GNN model can quickly identify the change of the relationship between the flow rate and the water level, and correct the prediction error result of the physical model; In the prior art, the fixed threshold cannot adapt to complex flood scenarios. Through feature engineering extraction and dynamic threshold training, the threshold is updated in real time; Specifically, by extracting real-time features, the real-time features include: current water level, water level rising rate, flow rate, rainfall, and historical features. The historical features are the evolution paths of similar flood events; Use the CART decision tree to train the relationship between real-time features and thresholds, and update the adjusted dynamic thresholds in real time; In the prior art, the abstract data of flood evolution is difficult to intuitively understand, which affects the decision-making efficiency; To solve the problem that the abstract data of flood evolution is difficult to intuitively understand and affects the decision-making efficiency, use LiDAR point cloud and InSAR interferometric measurement to construct the three-dimensional terrain of the basin, use the Shader shader to simulate the flow, reflection, and inundation effects of floods, and based on the water level data calculated by the physical model, update the water surface height and velocity vector in real time to achieve minute-level prediction of flood evolution; System self-healing module: Build an elastic computing power and self-healing mechanism, and ensure the dynamic scheduling and fault self-healing of system computing power through edge-cloud collaborative computing and Bayesian network fault assessment; In the prior art, the computing power of the water conservancy scenario cannot dynamically adjust the computing power mechanism according to the demand, resulting in too much computing volume during the rapid growth period of floods. By having the edge side undertake real-time lightweight tasks and the cloud handle batch calculations, flexible scheduling of computing power is achieved; The edge side undertakes tasks with high real-time performance and low weight; Exemplarily, it processes the data collected by the space-air-ground three-dimensional perception network; The cloud undertakes batch and high-complexity tasks; Exemplarily, it processes the flood evolution simulation of the entire basin and the training of machine learning models; Dynamic adjustment strategy, the edge side is triggered by the load threshold and unloads tasks to the cloud; Exemplarily, during the heavy rain period, the rainwater rises rapidly, and the real-time computing volume of the edge side is automatically unloaded to the cloud due to insufficient computing power, and the cloud GPU is used to accelerate the calculation to reduce the latency; Build a Bayesian network structure, define the nodes as devices and functions, and the edges as correlations. Exemplarily, if the 5G module fails, it will cause data transmission delay, which will in turn affect the flood simulation service. Use historical fault data to calculate the conditional probability table; The edge side collects the state of the space-air-ground three-dimensional perception network in real time, inputs it as evidence into the Bayesian network, and calculates the probability of failure through joint probability inference; The technical solution of the embodiment of the present invention is as follows: Build a multi-source sensor network. On the basis of traditional water level gauges and rain gauges, millimeter-wave radars, lidar, and 4K high-definition cameras are added; Beidou-3 short message and 5G-MEC dual-link transmission are adopted. Edge nodes are deployed within 5 kilometers of water conservancy facilities. Through the end-to-end delay model, transmission is optimized. When the 5G signal is weak, it automatically switches to the Beidou link, and local networking is achieved in combination with the LoRaWAN protocol; An AI box is deployed on the edge side. Using Daubechies-4 wavelet decomposition for 5 layers combined with soft threshold shrinkage for noise reduction, anomalies are determined through the Isolation Forest algorithm and invalid data is removed by PCA. Taking the Beidou PPS second pulse as the time reference, the UTC time synchronization of lidar, millimeter-wave radar, and camera is achieved through moving window averaging; Polar coordinates are converted to Cartesian coordinates, and the pixel coordinates of the 4K camera are converted to CGCS2000 geodetic coordinates through the Bursa seven-parameter transformation; A Kalman filter state model is constructed based on the Saint-Venant equation. Through the observation equation and the two-step iteration of prediction-update, the physical model prediction and real-time monitoring data are fused; The core data is hashed by SHA-256 and then uploaded to the Hyperledger Fabric water conservancy alliance chain. Only sensors on the whitelist can submit data, and the immutability is verified through hash comparison. The Saint-Venant equations are discretized using the Preissmann four-point implicit difference method, and a GPU is deployed to accelerate parallel computing using CUDA; An LSTM-GNN hybrid model is constructed. LSTM processes time series, and GNN captures spatial topology. The roughness error is corrected in combination with historical data; The dynamic threshold is trained through the CART decision tree, and real-time features such as the current water level and rising rate are fused with the historical flood path; A three-dimensional terrain is constructed using LiDAR point cloud and InSAR, and the Shader shader simulates the flood flow effect, and the water surface height and velocity vector are updated in real time. The edge side undertakes lightweight tasks such as data preprocessing, and the cloud is responsible for the whole-basin simulation and model training. When the edge computing power is insufficient during the rainstorm period, tasks are automatically offloaded to the cloud GPU; A Bayesian network is constructed, device correlation nodes are defined, the conditional probability table is calculated based on historical fault data, and the device status is collected in real time to infer the fault probability to achieve system self-healing.

[0024] Embodiment 2 Based on Embodiment 1, the present invention provides a smart water conservancy digital twin simulation method based on multi-source data, including: S1: Build an air-space-ground three-dimensional perception network, and realize minute-level data collection and transmission in combination with edge intelligent preprocessing; S2: Align the data collected by the air-space-ground three-dimensional perception network in terms of time consistency, perform spatial calibration on the data coordinates of the air-space-ground three-dimensional perception network using a spatio-temporal calibration algorithm, and achieve dynamic fusion of the physical model and the spatially calibrated data by means of digital-twin-driven Kalman filtering to improve the prediction accuracy of the physical model; S3: Construct a hybrid model based on long short-term memory network and graph neural network to solve the prediction error of the physical model, and use a lightweight architecture of physical model and machine learning, dynamic threshold decision tree and digital-twin visualization technology to achieve minute-level prediction of flood evolution; S4: Construct an elastic computing power and self-healing mechanism, and ensure dynamic scheduling of system computing power and self-healing of faults through edge-cloud collaborative computing and Bayesian network fault assessment.

[0025] The above has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A smart water conservancy digital twin simulation system based on multi-source data, characterized in that: Including: Edge intelligence preprocessing module: Construct an air-ground-space three-dimensional perception network, and combine edge intelligence preprocessing to achieve minute-level data collection and transmission; Trusted verification module: Align the time consistency of the data collected by the air-ground-space three-dimensional perception network, use the spatio-temporal calibration algorithm to perform spatial calibration on the data coordinates collected by the air-ground-space three-dimensional perception network, and use the Kalman filter driven by digital twin to realize the dynamic fusion of the physical model and the data after spatial calibration, improving the prediction accuracy of the physical model; Intelligent decision-making module: Construct a hybrid model based on long short-term memory network and graph neural network to solve the prediction error of the physical model, and use the lightweight architecture of physical model and machine learning, dynamic threshold decision tree and digital twin visualization technology to achieve minute-level prediction of flood evolution; System self-healing module: Construct an elastic computing power and self-healing mechanism, and through edge-cloud collaborative computing and Bayesian network fault assessment, ensure the dynamic scheduling of system computing power and fault self-healing.

2. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of constructing the air-ground-space three-dimensional perception network is as follows: Synchronously deploy millimeter-wave radars at the river cross-section and water level monitoring points, deploy lidar within the flooded area of the river and the basin terrain, and deploy 4K high-definition cameras at the water edge line and overtopping identification area of the river.

3. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of combining edge intelligence preprocessing to achieve minute-level data collection and transmission is as follows: Deploy an AI edge box at the sensor terminal to optimize the data layer quality; Use Daubechies-4 wavelet to decompose the time-series data collected by the air-ground-space three-dimensional perception network into 5 layers, balance the calculation amount and denoising effect, use soft threshold shrinkage to avoid signal mutation of hard threshold, and the soft threshold is adaptively calculated by Stein's unbiased risk estimation to identify and process invalid data and abnormal data in the data collected by the air-ground-space three-dimensional perception network; Use the isolation forest anomaly detection method to calculate the anomaly score of the data collected by the air-ground-space three-dimensional perception network, identify sensor drift, use the PCA feature compression method to perform principal component analysis on the data collected by the air-ground-space three-dimensional perception network, remove mutation abnormal data, and retain effective principal components; The AI edge box runs a random forest model to monitor the data collected by the air-ground-space three-dimensional perception network after optimizing the data layer quality in real time. When the water level exceeds the warning value, it triggers an audible and visual alarm to avoid uploading invalid data to the cloud.

4. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, wherein: The specific process of aligning the time consistency of the data collected by the air-ground-space three-dimensional perception network is as follows: Use the Beidou-3 PPS second pulse signal as the time reference, calculate the deviation between the clocks of the lidar, millimeter-wave radar and 4K high-definition camera and UTC, suppress the noise through moving window averaging, synchronize the lidar, millimeter-wave radar and 4K high-definition camera to UTC, ensure time consistency, and achieve second-level precise alignment.

5. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of using the spatio-temporal calibration algorithm to perform spatial calibration on the data coordinates collected by the air-ground-space three-dimensional perception network is as follows: Convert polar coordinates to Cartesian coordinates. The specific conversion process is as follows: Abscissa transformation: , Ordinate transformation: , where is the distance between the lidar and the cross-section, is the angle between the radar beam and the water flow direction; after transformation, it is unified with the CGCS2000 coordinate system of the ground sensor; Convert the pixel coordinates of 4K high-definition video to geodetic coordinates, and perform the conversion by calibrating the camera with the Bursa seven parameters: , where , , are translation parameters, is the scaling factor, , , are rotation angles, which are calibrated by field control points.

6. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of using digital twin-driven Kalman filtering to achieve dynamic fusion of the physical model and spatially calibrated data and improve the prediction accuracy of the physical model is as follows: Based on the linearization and discretization of the Saint-Venant equations, the state vector and state equations are constructed respectively. Since there are errors in the physical model prediction, real-time data collected by the space-air-ground stereo perception network need to be used for correction. As a fusion tool for model prediction and real-time observation, Kalman filtering can calibrate the state deviation in real time using sensor data while retaining the mechanism of the physical model. Its core is achieved through the construction of the observation equation and two-step iterative update. The observation equation establishes the mapping between the system state and the real-time data collected by the space-air-ground stereo perception network; If the space-air-ground stereo perception network has high accuracy, the observation data dominates the correction. If the physical model is reliable, the model prediction dominates the correction.

7. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of constructing a hybrid model based on long short-term memory network and graph neural network to solve the physical model prediction error is as follows: Construct a hybrid model of LSTM long short-term memory network and GNN graph neural network. Among them, the LSTM long short-term memory network is used to process the time series of data collected by the space-air-ground stereo perception network to capture the time dependence of flood evolution, and the GNN processes the spatial topology to capture the spatial propagation of flood evolution; Using the physical model simulation results and actual monitoring data of historical flood events as the training set to train the model, the loss function is the mean absolute error, and the formula is: , where is the actual monitored water level, is the water level predicted by the model, is the sample size; When the sudden change of the downstream water level is caused by the flood discharge of the upstream reservoir, the LSTM-GNN model can quickly identify the change of the relationship between flow and water level and correct the prediction error of the physical model.

8. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of achieving minute-level prediction of flood evolution is as follows: By extracting real-time features, the real-time features include: current water level, water level rising rate, flow rate, rainfall, and historical features. The historical features are the evolution paths of similar flood events; Use the CART decision tree to train the relationship between real-time features and thresholds, and update the dynamically adjusted thresholds in real time. Use LiDAR point cloud and InSAR interferometry to construct the three-dimensional terrain of the basin. Use the Shader shader to simulate the flow, reflection, and inundation effects of floods. Based on the water level data calculated by the physical model, update the water surface height and velocity vector in real time to achieve minute-level prediction of flood evolution.

9. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of constructing elastic computing power is as follows: By having the edge side undertake real-time lightweight tasks and the cloud handle batch calculations, flexible scheduling of computing power is achieved. The cloud undertakes the flood evolution simulation of the entire basin and the training of machine learning models, and dynamically adjusts the strategy. The edge side is triggered by the load threshold and offloads tasks to the cloud.

10. The intelligent water conservancy digital twin simulation system based on multi-source data according to claim 1, characterized in that: The specific process of the fault assessment is as follows: Construct a Bayesian network structure, define the nodes as devices and functions, use historical fault data to calculate the conditional probability table, and the edge side collects the state of the space-air-ground stereo perception network in real time as evidence input to the Bayesian network. Through joint probability inference, calculate the probability of fault occurrence.

Images