Water conservancy project remote early warning method and system

Abstract

The invention relates to the technical field of water conservancy project monitoring and early warning, and discloses a water conservancy project remote early warning method and system, and the method comprises the steps: 1, synchronously collecting water level data and probe displacement data through a radar sensor and an inertia measurement module, and carrying out the dynamic compensation of the data through an extended Kalman filtering algorithm, generating a data message containing the node identifier, the correction displacement and the water level height; and 2, receiving the data message in the step 1, storing the data message in an annular buffer area, performing accumulative compensation on displacement data based on a sliding mean filtering algorithm, and outputting a calibrated vertex coordinate set in combination with preset topographic data. The technical scheme of radar and IMU data synchronous acquisition and extended Kalman filtering dynamic compensation is adopted, sub-centimeter level displacement compensation precision is achieved, compared with the scheme that probe shaking is counteracted by depending on a redundant radar array in the prior art, the problems of high hardware cost and high deployment complexity are solved, and single-probe high-precision water level monitoring is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy project monitoring and early warning, and specifically to a remote early warning method and system for water conservancy projects. Background Art

[0002] Remote early warning of water conservancy projects is the core link to ensure flood control safety and optimize water resource allocation. The key lies in obtaining high-precision water level data in real time and predicting abnormal trends through intelligent algorithms. The present invention proposes a remote early warning method and system for water conservancy projects, which solves the bottlenecks of traditional solutions in terms of hardware cost, real-time performance, and false alarm rate through multi-source sensor data fusion, edge computing modeling, and confidence adaptive early warning technology, providing a cost-effective remote monitoring solution for small and medium-sized water conservancy facilities.

[0003] The existing technologies represented by the comparative document CN118936596B (a remote early warning method and system for water conservancy projects) mainly adopt the following technical routes:

[0004] By deploying multiple radar devices to form an array, the displacement error of a single probe is offset by spatial redundancy. However, this solution requires additional configuration of a synchronization controller and a data fusion module, resulting in an increase in the hardware cost of a single node by about 60%, and professional teams are required for installation and commissioning, making it impossible to popularize in small and medium-sized reservoirs and rivers.

[0005] The original water level data is transmitted to the cloud server, and a high-precision water level model is constructed based on the finite element algorithm. However, due to the computational complexity of finite element mesh generation, the time-consuming for a single modeling exceeds 5 seconds, and the end-to-end delay cannot meet the second-level early warning requirements.

[0006] A static water level threshold is set, without considering the instantaneous false alarms caused by sensor noise and environmental disturbances. Statistical data shows that under complex working conditions such as rainfall and wind waves, the false alarm rate is as high as 15% - 20%, and frequent false triggers will reduce the trust of operation and maintenance personnel.

[0007] CN118936596B relies on multi-radar spatial redundancy to offset displacement errors, but the radar unit price is high, and a dedicated communication link needs to be configured, resulting in an exponential increase in cost and deployment complexity.

[0008] Although the finite element algorithm has high modeling accuracy, the superposition of its computationally intensive characteristics and cloud resource scheduling delay makes the end-to-end delay exceed 6 seconds, unable to cope with water level mutation scenarios.

[0009] The fixed threshold does not introduce the confidence evaluation of prediction results, and it is easy to misjudge noise as a danger under sensor noise or short-term disturbances.

[0010] Therefore, the present invention proposes a remote early warning method and system for water conservancy projects to solve the above-mentioned problems. Summary of the Invention

[0011] In view of the deficiencies of the prior art, the present invention provides a remote warning method and system for water conservancy projects to solve the problems raised in the above background art.

[0012] To achieve the above objectives, the present invention is realized through the following technical solutions: A remote warning method for water conservancy projects, including:

[0013] Step 1, synchronously collect water level data and probe displacement data through a radar sensor and an inertial measurement module, and use the extended Kalman filter algorithm to dynamically compensate the data to generate a data packet containing node identification, corrected displacement, and water level height;

[0014] Step 2, receive the data packet of Step 1 and store it in a circular buffer, perform cumulative compensation on the displacement data based on the sliding mean filter algorithm, and output a calibrated vertex coordinate set in combination with preset terrain data;

[0015] Step 3, according to the vertex coordinate set, construct a grid water level matrix through the inverse distance weighted interpolation algorithm, detect abnormal grid data and trigger data resampling;

[0016] Step 4, input the water level matrix into a pre-trained 1D convolutional neural network model, and output the predicted water level value and confidence standard deviation at the next moment;

[0017] Step 5, generate a hierarchical warning instruction according to the predicted value and confidence standard deviation, and send it to the execution terminal through a multi-protocol communication link to render a visualization interface.

[0018] Preferably, in Step 1, synchronously collect water level data and probe displacement data through a radar sensor and an inertial measurement module, and use the extended Kalman filter algorithm to dynamically compensate the data to generate a data packet containing node identification, corrected displacement, and water level height, which further includes:

[0019] Step 1.1, synchronously collect data: trigger the radar sensor and the inertial measurement module to synchronously collect data through a hardware interrupt:

[0020] The radar sensor obtains a ranging value h, which represents the water level height;

[0021] The inertial measurement module obtains three-axis acceleration components (a x , a y , a z ) and three-axis angular velocity components (g x , g y , g z );

[0022] Align h, (a x , a y , az ), (g x , g y , g z Synchronized to the microsecond level;

[0023] Step 1.2, Dynamic compensation calculation: Input the synchronization data from Step 1.1 into the Extended Kalman Filter algorithm to calculate the probe displacement compensation amounts Δx, Δy, Δz and the corrected water level height h':

[0024] State equation:

[0025]

[0026] where x k is the probe displacement vector [Δx, Δy, Δz] at time k T ,

[0027] x k-1 is the probe displacement vector at time k - 1,

[0028] v k-1 is the velocity vector at time k - 1 (calculated by integrating the angular velocity g x , g y , g z ),

[0029] a k is the acceleration component [a x , a y , a z , T ,

[0030] Δt is the sampling interval time, w k is the process noise vector;

[0031] Observation equation: z k = h k - H·x k + n k ,

[0032] where z k is the radar ranging observation value, H is the observation matrix, n k is the observation noise vector, n k is the observation noise vector, h k is the original ranging value of the radar sensor at time k;

[0033] Kalman gain update:

[0034] K k = P k|k-1 ·H T ·(H·P k|k-1 ·H T + R k) -1 ,

[0035] where P k|k-1 is the state covariance prediction matrix, R k is the observation noise covariance matrix, K k is the Kalman gain matrix, and H T is the transpose matrix of the observation matrix H;

[0036] Step 1.3: Package the displacement compensation amounts Δx, Δy, Δz and the corrected water level height h′ output in Step 1.2, generate a data packet containing the node identifier nodeID, displacement vector (Δx, Δy, Δz), water level value h′ and timestamp t s . After appending the CRC-16-CCITT check code, it is sent through the LoRaWAN protocol, and the sending parameters include the spreading factor SF10 and the bandwidth 125 kHz.

[0037] Preferably, in Step 2, the data packet in Step 1 is received and stored in a circular buffer, and the displacement data is cumulatively compensated based on the sliding mean filtering algorithm, and the calibrated vertex coordinate set is output in combination with the preset terrain data, which further includes:

[0038] Step 2.1: Receive the packet in Step 1 and write it into the circular buffer:

[0039] Define the capacity of the circular buffer as 60 seconds of historical data, and the stored packet fields include the node identifier nodeID, displacement compensation amounts Δx, Δy, Δz, corrected water level height h′ and timestamp t s ;

[0040] Implement multi-threaded read and write synchronization through a memory lock, and update the write pointer to point to the latest data position;

[0041] Step 2.2: Read the displacement data of the past 5 seconds from the circular buffer and calculate the displacement compensation amount after sliding mean filtering:

[0042] Sliding window mean formula:

[0043]

[0044] Output the mean displacement

[0045] where Δx i , Δy i , Δz i are the output displacement compensation amounts,

[0046] is the sliding window mean displacement compensation amount, N is the sliding window size, and k is the time step index;

[0047] Step 2.3, calibrate the displacement mean according to the preset terrain point set D to generate the vertex coordinate set V t :

[0048] The terrain point set D = {(x d , y d , h d )} contains the preset longitude, latitude and elevation data;

[0049] Plane coordinate constraint formula:

[0050]

[0051] y i ′ = y d corresponding to x i ′ , where x0, y0 are the initial installation coordinates of the probe, and x i ′ , y i ′ are the plane coordinates of the probe after terrain constraint;

[0052] Water level height correction:

[0053] Output vertex set: V t = {(x ′ 1, y ′ 1, h ′ 1),…,(x ′ n , y ′ n , h ′ n )},

[0054] where h i ′ is the corrected water level height, and V t is the vertex coordinate set.

[0055] Preferably, in step 3, according to the vertex coordinate set, a gridded water level matrix is constructed by the inverse distance weighted interpolation algorithm, and grid data anomalies are detected and data resampling is triggered, which further includes:

[0056] Step 3.1, based on the vertex coordinate set V t output in step 2 and the preset terrain point set D, divide the monitoring area into 10 m × 10 m grids, define the number of grid rows M and the number of columns F, and generate the initial water level matrix H t ;

[0057] Step 3.2, for each grid point (x, y), perform inverse distance weighted interpolation to calculate the water level value:

[0058]

[0059] Among them, the Euclidean distance from the grid point (x, y) to the vertex (x i ′ , y i ′ ), h i ′ is the corrected water level height of vertex i, and ∈ is a very small constant to prevent division by zero;

[0060] Step 3.3, detecting abnormal grid data:

[0061] Calculate the mean absolute deviation δ between the grid point H(x, y) and the water level values of its 8 adjacent points:

[0062]

[0063] where H(x, y) is the water level height interpolated at the grid point (x, y), and H(x j , y j ) is the water level height interpolated at the grid point (x j , y j );

[0064] If δ > 0.1·H(x, y), mark H(x, y) as abnormal data;

[0065] If the proportion of abnormal data exceeds 5%, send a resampling request to Step 1.

[0066] Preferably, in the said Step 4, inputting the water level matrix into a pre-trained 1D convolutional neural network model, and outputting the predicted water level value and the confidence standard deviation for the next moment, further includes:

[0067] Step 4.1, constructing a historical water level matrix sequence:

[0068] Extract 5 consecutive frames of historical data from the water level matrix H t output by Step 3 to form an input sequence X: X = [H t-4 , H t-3 , H t-2 , H t-1 , H t ;

[0069] where [H t-4 , H t-3 , H t-2 , H t-1 , H t is the historical water level matrix sequence;

[0070] Perform normalization on the input sequence:

[0071]

[0072] where X norm is the input sequence of the normalized historical water level matrix, μ is the mean of the training dataset, and σ is the standard deviation of the training dataset;

[0073] Step 4.2, perform inference through a 1D convolutional neural network model:

[0074] Model structure:

[0075] Z1 = ReLU(W1 * X norm + b1),

[0076] Z2 = MaxPooling(Z1),

[0077] Z3 = Flatten(Z2),

[0078]

[0079] where Z1 is the output feature map of the convolutional layer, Z2 is the output feature map of the max pooling layer, Z3 is the feature vector after the flattening operation, W1 is the weight matrix of the convolutional layer, W2 is the weight matrix of the fully connected layer, b1 and b2 are bias vectors, is the predicted water level matrix at the next moment output by the model;

[0080] Output: Predicted water level matrix

[0081] Step 4.3, calculate the confidence standard deviation:

[0082] Standard deviation formula:

[0083]

[0084] where is the predicted matrix 's mean, G is the total number of grid points, and σ is the confidence standard deviation of the prediction result;

[0085] If σ > 0.15, trigger the data resampling request in Step 1.

[0086] Preferably, in the said Step 5, generate a hierarchical early warning instruction according to the predicted value and the confidence standard deviation, and send it to the execution terminal through a multi-protocol communication link to render the visualization interface, which further includes:

[0087] Step 5.1, generate a hierarchical early warning instruction:

[0088] Early warning threshold determination formula:

[0089]

[0090] Among them, ΔH is the deviation between the predicted water level peak and the current water level peak, and H t is the initial water level matrix, is the predicted water level matrix at the next moment output by the model, and σ is the standard deviation of the confidence level of the output;

[0091] Step 5.2, issue an instruction through a multi-protocol communication link:

[0092] MQTT protocol parameters:

[0093] broker = iot.example.com, port = 1883, topic = water / alert / {nodeID},

[0094] Among them, broker is the target server address for issuing warning instructions through the MQTT protocol, port is the port configuration of the communication link, and topic is the main body of the instruction publication;

[0095] Enable the QoS1 service quality level;

[0096] Heartbeat detection mechanism: Send a heartbeat packet to the edge node every 30 seconds, and determine that the node is offline if the timeout occurs 3 times;

[0097] Step 5.3, render the visualization interface:

[0098] Three-dimensional water level surface rendering: Render the current water level matrix H output in Step 3 based on the WebGL engine t , and superimpose the predicted water level contour lines in Step 4.2

[0099] Color mapping rule:

[0100]

[0101] The mobile APP synchronously displays the water level curves of key nodes and the warning level.

[0102] Preferably, the data transmission between Step 1 and Step 2 uses the LoRaWAN protocol and is appended with a CRC-16 check code;

[0103] The transmission of the prediction results between Step 4 and Step 5 uses the MQTT protocol and enables the QoS1 service quality level.

[0104] A remote warning system for water conservancy projects, the remote warning system for water conservancy projects includes:

[0105] Sensor node module, integrating radar sensor, inertial measurement module and LoRaWAN communication unit;

[0106] Edge positioning module, equipped with a ring buffer and memory lock management unit;

[0107] Water level modeling module with built-in inverse distance weighted interpolation algorithm and anomaly detection unit;

[0108] Prediction module, deploys 1D convolutional neural network model and confidence calculation unit;

[0109] The early warning visualization module supports hierarchical early warning decision-making, multi-terminal interface rendering and command issuance.

[0110] A terminal device comprises a processor, a memory and a communication interface, wherein the memory stores a computer program and the processor implements a remote early warning method for a water conservancy project when executing the program.

[0111] A storage medium stores computer-executable instructions, which, when executed by a processor, implement a water conservancy project remote early warning method.

[0112] The present invention provides a remote early warning method and system for water conservancy projects. It has the following beneficial effects:

[0113] 1. The present invention adopts the synchronous acquisition of radar and IMU data and the extended Kalman filter dynamic compensation technology to achieve sub-centimeter displacement compensation accuracy. Compared with the existing technology that relies on redundant radar arrays to offset probe shaking, it solves the shortcomings of high hardware cost and large deployment complexity, and realizes high-precision water level monitoring with a single probe.

[0114] 2. The present invention adopts a lightweight inverse distance weighted interpolation modeling technology solution on the edge to achieve a technical effect of a water level matrix generation delay of less than 500 milliseconds. Compared with the existing technology that relies on cloud-based finite element modeling, it solves the shortcomings of large computing resource consumption and high end-to-end delay, and meets the real-time early warning needs of water conservancy projects.

[0115] 3. The present invention adopts a dynamic threshold adjustment technology solution based on the prediction standard deviation to achieve a technical effect of a false alarm rate of less than 5%. Compared with the fixed threshold warning solution in the existing technology, it solves the problem of high false alarm rate in complex hydrological environment and improves the reliability and response efficiency of warning decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Figure 1 is a flow chart of the present invention;

[0117] Figure 2 It is a schematic diagram of the system composition of the present invention;

[0118] Figure 3This is the system flow chart of the present invention. Detailed implementation manners

[0119] To enable those skilled in the art to understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0120] The present invention will be described in detail below with reference to the accompanying drawings:

[0121] Embodiment:

[0122] Please refer to the attached Figure 1 - attached Figure 3 , the embodiment of the present invention provides a remote warning method for water conservancy projects, including:

[0123] Step 1, synchronously collect water level data and probe displacement data through a radar sensor and an inertial measurement module, and use the extended Kalman filter algorithm to dynamically compensate the data to generate a data packet containing node identification, corrected displacement, and water level height;

[0124] Step 1.1, synchronously collect data: trigger the radar sensor and the inertial measurement module to synchronously collect data through a hardware interrupt:

[0125] The radar sensor obtains a ranging value h, which represents the water level height;

[0126] The inertial measurement module obtains three-axis acceleration components (a x , a y , a z ) and three-axis angular velocity components (g x , g y , g z );

[0127] Synchronize h, (a x , a y , a z ), (g x , g y , g z ) to the microsecond level through a timestamp alignment mechanism;

[0128] Step 1.2, dynamic compensation calculation: input the synchronized data in Step 1.1 into the extended Kalman filter algorithm to calculate the probe displacement compensation amounts Δx, Δy, Δz, and the corrected water level height h';

[0129] State equation:

[0130]

[0131] Among them, x k is the probe displacement vector [Δx, Δy, Δz] at time k T ,

[0132] x k-1 is the probe displacement vector at time k - 1,

[0133] v k-1 is the velocity vector at time k - 1 (calculated by integrating angular velocity g x , g y , g z ),

[0134] a k is the acceleration component [a x , a y , a z T ,

[0135] Δt is the sampling interval time, and w k is the process noise vector;

[0136] Observation equation: z k = h k - H·x k + n k ,

[0137] Among them, z k is the radar ranging observation value, H is the observation matrix, and n k is the observation noise vector, n k is the observation noise vector, and h k is the original ranging value of the radar sensor at time k;

[0138] Kalman gain update:

[0139] K k = P k|k-1 ·H T ·(H·P k|k-1 ·H T + R k ) -1 ,

[0140] Among them, P k|k-1 is the state covariance prediction matrix, R k is the observation noise covariance matrix, K k is the Kalman gain matrix, and H T is the transpose matrix of the observation matrix H;

[0141] ​Step 1.3, encapsulate the displacement compensation amounts Δx, Δy, Δz and the corrected water level height h′ output in Step 1.2 to generate a data packet containing the node identifier nodeID, displacement vector (Δx, Δy, Δz), water level value h′ and time stamp t s After appending the CRC-16-CCITT check code, it is sent through the LoRaWAN protocol. The sending parameters include the spreading factor SF10 and the bandwidth 125 kHz;

[0142] Step 2, receive the data packet in Step 1 and store it in the circular buffer, perform cumulative compensation on the displacement data based on the sliding mean filtering algorithm, and output the calibrated vertex coordinate set in combination with the preset terrain data;

[0143] Step 2.1, receive the packet in Step 1 and write it into the circular buffer:

[0144] Define the capacity of the circular buffer as 60 seconds of historical data. The stored packet fields include the node identifier nodeID, displacement compensation amounts Δx, Δy, Δz, corrected water level height h′ and time stamp t s ;

[0145] Implement multi-threaded read and write synchronization through a memory lock, and update the write pointer to point to the latest data position;

[0146] Step 2.2, read the displacement data of the past 5 seconds from the circular buffer and calculate the displacement compensation amount after sliding mean filtering:

[0147] Sliding window mean formula:

[0148]

[0149] Output the mean displacement

[0150] where, Δx i , Δy i , Δz i are the output displacement compensation amounts,

[0151] is the sliding window mean displacement compensation amount, N is the sliding window size, and k is the time step index;

[0152] Step 2.3, perform coordinate calibration on the displacement mean in combination with the preset terrain point set D to generate the vertex coordinate set V t :

[0153] Terrain point set D = {(x d , y d , h d )} contains the preset longitude, latitude and elevation data;

[0154] Plane coordinate constraint formula:

[0155]

[0156] y i ′ = y d corresponding to x i ′ , where x0 and y0 are the initial installation coordinates of the probe, and x i ′ , y i ′ are the plane coordinates of the probe after terrain constraint;

[0157] Water level height correction:

[0158] Output vertex set: V t = {(x ′ 1, y ′ 1, h ′ 1),…,(x ′ n , y ′ n , h ′ n )},

[0159] where h i ′ is the corrected water level height, and V t is the vertex coordinate set;

[0160] Step 3, based on the vertex coordinate set, construct a gridded water level matrix through the inverse distance weighted interpolation algorithm, detect grid data anomalies and trigger data resampling;

[0161] Step 3.1, based on the vertex coordinate set V t output in Step 2 and the preset terrain point set D, divide the monitoring area into 10m×10m grids, define the number of grid rows M and the number of columns F, and generate the initial water level matrix H t ;

[0162] Step 3.2, for each grid point (x, y), perform inverse distance weighted interpolation to calculate the water level value:

[0163]

[0164] where the Euclidean distance from the grid point (x, y) to the vertex (x i ′ , y i ′ ), h i ′is the corrected water level height of vertex i, and ∈ is a very small constant to prevent division by zero;

[0165] Step 3.3, detect abnormal grid data:

[0166] Calculate the mean absolute deviation δ between the grid point H(x, y) and the water level values of its adjacent 8 points:

[0167]

[0168] where H(x, y) is the water level height interpolated at the grid point (x, y), and H(x j , y j ) is the water level height interpolated at the grid point (x j , y j );

[0169] If δ > 0.1·H(x, y), mark H(x, y) as abnormal data;

[0170] If the proportion of abnormal data exceeds 5%, send a resampling request to Step 1;

[0171] Step 4, input the water level matrix into a pre-trained 1D convolutional neural network model to output the predicted water level value and the confidence standard deviation for the next moment;

[0172] Step 4.1, construct a historical water level matrix sequence:

[0173] Extract 5 consecutive frames of historical data from the water level matrix H t output by Step 3 to form an input sequence X: X = [H t-4 , H t-3 , H t-2 , H t-1 , H t ;

[0174] where [H t-4 , H t-3 , H t-2 , H t-1 , H t is the historical water level matrix sequence;

[0175] Perform normalization processing on the input sequence:

[0176]

[0177] where X norm is the normalized input sequence of the historical water level matrix, μ is the mean of the training dataset, and σ is the standard deviation of the training dataset;

[0178] Step 4.2, perform inference through the 1D convolutional neural network model:

[0179] Model structure:

[0180] Z1 = ReLU(W1 * X norm + b1),

[0181] Z2 = MaxPooling(Z1),

[0182] Z3 = Flatten(Z2),

[0183]

[0184] where Z1 is the output feature map of the convolutional layer, Z2 is the output feature map of the max pooling layer, Z3 is the feature vector after the flattening operation, W1 is the weight matrix of the convolutional layer, W2 is the weight matrix of the fully connected layer, b1 and b2 are bias vectors, is the predicted water level matrix at the next moment output by the model;

[0185] Output: Predicted water level matrix

[0186] Step 4.3, calculate the standard deviation of confidence:

[0187] Standard deviation formula:

[0188]

[0189] where, is the prediction matrix 's mean value, G is the total number of grid points, and σ is the standard deviation of confidence of the prediction result;

[0190] If σ > 0.15, trigger the data resampling request in Step 1;

[0191] Step 5, generate a hierarchical early warning instruction based on the prediction value and the standard deviation of confidence, and send it to the execution terminal through a multi - protocol communication link to render the visualization interface;

[0192] Step 5.1, generate a hierarchical early warning instruction:

[0193] Early warning threshold determination formula:

[0194]

[0195] where, ΔH is the deviation between the predicted water level peak and the current water level peak, H t is the initial water level matrix, is the predicted water level matrix at the next moment output by the model, and σ is the standard deviation of confidence of the output;

[0196] Step 5.2, send the instruction through a multi - protocol communication link:

[0197] MQTT protocol parameters:

[0198] broker=iot.example.com,port=1883,topic=water / alert / {nodeID},

[0199] Among them, broker is the target server address for issuing warning instructions under the MQTT protocol, port is the port configuration of the communication link, and topic is the subject of instruction publishing;

[0200] Enable QoS1 quality of service level;

[0201] Heartbeat detection mechanism: Send heartbeat packets to edge nodes every 30 seconds. If the heartbeat packets time out three times, the node is considered offline.

[0202] Step 5.3, render the visualization interface:

[0203] 3D water level surface rendering: The current water level matrix H output in step 3 is rendered based on the WebGL engine t , superimpose the predicted water level contours of step 4.2

[0204] Color mapping rules:

[0205]

[0206] The mobile APP simultaneously displays the water level curve and warning level of key nodes.

[0207] Technical Advantages of Step 1: Through the hardware interrupt synchronization acquisition mechanism between the radar sensor and the inertial measurement module, combined with the extended Kalman filter algorithm, dynamic sub-centimeter compensation of probe displacement is achieved. The radar ranging value and the IMU's three-axis acceleration and angular velocity data are combined through closed-loop iteration of the state equation and observation equation, significantly suppressing probe motion errors caused by water impact and wind and wave turbulence. Compared to traditional redundant radar array solutions, the hardware cost of a single probe is reduced by over 60%. The LoRaWAN protocol also ensures low-power, long-distance transmission, meeting the requirements of long-term deployment in off-grid scenarios. The CRC-16-CCITT checksum further ensures message integrity and prevents data distortion caused by wireless channel interference.

[0208] Technical advantages of Step 2: Based on the multi-thread synchronization mechanism of the circular buffer and memory lock, stable access to historical displacement data is achieved, avoiding data overwrite and loss. The sliding window mean filter smooths short-term high-frequency disturbances and outputs the mean displacement, reducing the impact of single-sampling noise on coordinate calibration. Combining with the planar coordinate constraints of the preset terrain point set D, the probe displacement is mapped to the geographic coordinate system to avoid systematic deviations caused by installation tilt or foundation settlement. The water level height correction formula further eliminates the vertical drift error of the probe, ensuring the spatial consistency of the vertex set and providing highly reliable input for subsequent modeling.

[0209] Technical advantages of Step 3: The anomaly detection mechanism identifies local data mutations through the mean absolute deviation of adjacent 8 points and triggers resampling in combination with a 5% anomaly ratio threshold, effectively filtering outlier values caused by occasional sensor failures or environmental interferences and ensuring the robustness of the model output. This closed-loop self-check mechanism reduces the frequency of manual intervention and lowers the operation and maintenance costs.

[0210] Technical advantages of Step 4: The pre-trained 1D convolutional neural network extracts spatio-temporal features from the historical water level matrix sequence X. Normalization processing eliminates the dimension difference, and while outputting the predicted water level matrix, the confidence standard deviation is calculated. The model is deployed in a lightweight manner at the edge to achieve a second-level inference speed, and the prediction accuracy meets the flood control warning requirements. The confidence threshold dynamically triggers a resampling request to avoid low-quality prediction data misleading decision-making, forming a closed-loop optimization link of "perception - prediction - verification" and significantly improving the system's adaptive ability.

[0211] Technical advantages of Step 5: The hierarchical warning instructions are dynamically adjusted based on the predicted water level deviation and confidence standard deviation. The decision-making logics for red warnings and orange warnings are separated, reducing the false alarm rate to below 5%. The MQTT protocol ensures reliable instruction issuance and supports access by tens of thousands of terminals. The three-dimensional water level surface and the mobile APP visualization interface intuitively display the risk distribution through color mapping rules, assisting management personnel in quickly locating the dangerous areas. The multi-protocol communication and cross-terminal data synchronization mechanism enable global collaborative response to warning instructions.

[0212] A remote warning system for water conservancy projects, the remote warning system for water conservancy projects includes:

[0213] A sensor node module, integrating a radar sensor, an inertial measurement module, and a LoRaWAN communication unit;

[0214] An edge positioning module, equipped with a circular buffer and a memory lock management unit;

[0215] A water level modeling module, built-in with an inverse distance weighted interpolation algorithm and an anomaly detection unit;

[0216] A prediction module, deploying a 1D convolutional neural network model and a confidence calculation unit;

[0217] The early warning visualization module supports hierarchical early warning decision-making, multi-terminal interface rendering, and instruction issuance.

[0218] The sensor node module integrates a radar sensor, an inertial measurement module, and a LoRaWAN communication module, and is responsible for real-time collection of water level and probe displacement data. The technical advantages are that the hardware cost of a single node is reduced by 60%, and multi-source data hardware-level synchronization is achieved through microsecond-level timestamp alignment, providing high-precision input for dynamic compensation.

[0219] The edge positioning correction module, based on the circular buffer and memory lock management module, realizes the cumulative compensation of displacement data and terrain constraint calibration. The technical advantages are that the terrain calibration accuracy reaches ±0.1 meters, and the mean filtering reduces the displacement compensation volatility by 30%, providing stable input for the modeling module.

[0220] The water level modeling module constructs a grid water level matrix based on the inverse distance weight interpolation algorithm and detects data anomalies. The technical advantages are the complexity of the IDW algorithm, the edge-side resource occupancy is reduced by 80%, and the false positive rate of anomaly detection is <3%, ensuring the robustness of the modeling.

[0221] The prediction module predicts the water level trend through a pre-trained 1D convolutional neural network model and evaluates the confidence level. The technical advantages are that the number of model parameters is <100,000, the edge-side inference speed reaches 50 frames per second, and the confidence level threshold dynamically optimizes the prediction reliability.

[0222] The early warning visualization module generates hierarchical early warning instructions and realizes danger response through multi-protocol communication and a visualization interface.

[0223] A terminal device includes a processor, a memory, and a communication interface. The memory stores a computer program, and when the processor executes the program, it implements a remote early warning method for water conservancy projects.

[0224] A storage medium stores computer-executable instructions, and when the instructions are executed by a processor, it implements a remote early warning method for water conservancy projects.

[0225] The advantages of the terminal device include: the integrated design of the radar, inertial measurement module, and communication module reduces the deployment complexity and power supply requirements, and supports deployment in field environments without a network; the processor locally executes the extended Kalman filter, inverse distance weight interpolation, and 1D-CNN prediction algorithms, and the end-to-end delay is <1 second, meeting the real-time response requirements of water level mutation scenarios; the combination of LoRaWAN communication and the processor's dynamic frequency adjustment technology results in a standby power consumption of <1W per single node, and the battery life reaches more than 3 years; the communication interface supports the switching between LoRaWAN and MQTT protocols, adapts to the cloud-edge collaborative architecture, and ensures the reliability of instruction issuance and data backhaul.

[0226] Technical advantages of the storage medium: Support the OTA update warning algorithm, and there is no need to replace the hardware to upgrade the system function; Store the historical water level matrix, prediction results and warning logs, provide a data basis for post-disaster traceability analysis, the storage capacity is ≥ 128GB, support 10-year data retention; Adopt AES-256 encryption and write-protected partitions to prevent malicious tampering with program logic and data leakage.

[0227] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A remote warning method for water conservancy projects, characterized in that, Including: Step 1: Synchronously collect water level data and probe displacement data through a radar sensor and an inertial measurement module, and use the extended Kalman filter algorithm to dynamically compensate the data to generate a data message containing node identification, corrected displacement, and water level height; Step 2: Receive the data message in Step 1 and store it in a circular buffer, perform cumulative compensation on the displacement data based on the sliding mean filter algorithm, and output a calibrated vertex coordinate set in combination with preset terrain data; Step 3: According to the vertex coordinate set, construct a gridded water level matrix through the inverse distance weighted interpolation algorithm, detect abnormal grid data, and trigger data resampling; Step 4: Input the water level matrix into a pre-trained 1D convolutional neural network model to output the predicted water level value and confidence standard deviation at the next moment; Step 5: Generate a hierarchical warning instruction according to the predicted value and confidence standard deviation, send it to the execution terminal through a multi-protocol communication link, and render a visual interface.

2. The remote warning method for a water conservancy project according to claim 1, wherein, In Step 1, synchronously collect water level data and probe displacement data through a radar sensor and an inertial measurement module, and use the extended Kalman filter algorithm to dynamically compensate the data to generate a data message containing node identification, corrected displacement, and water level height, which further includes: Step 1.1: Synchronously collect data: Trigger the radar sensor and the inertial measurement module to synchronously collect data through a hardware interrupt: The radar sensor obtains a ranging value h, which represents the water level height; The inertial measurement module acquires the three-axis acceleration components (a x , a y , a z ) and the three-axis angular velocity components (g x , g y , g z ); Align h, (a x , a y , a z ), (g x , g y , g z ) to the microsecond level through the timestamp alignment mechanism; Step 1.2: Dynamic compensation calculation: Input the synchronous data in Step 1.1 into the extended Kalman filter algorithm to calculate the probe displacement compensation amounts Δx, Δy, Δz, and the corrected water level height h'; State equation: Among them, x k is the probe displacement vector [Δx, Δy, Δz] at time k T , x k-1 is the probe displacement vector at time k-1, v k-1 is the velocity vector at time k-1 (calculated by integrating angular velocity g x , g y , g z ), a k For the acceleration component [a x , a y , a z T ,​ Δt is the sampling interval time, w k is the process noise vector; Observation equation: z k = h k - H·x k + n k , where z k is the radar ranging observation value, H is the observation matrix, and n k is the observation noise vector, and n k is the observation noise vector, and h k is the original ranging value of the radar sensor at time k; Kalman gain update: K k = P k|k-1 ·H T ·(H·P k|k-1 ·H T + R k ) -1 , Among them, P k|k-1 is the state covariance prediction matrix, R k is the observation noise covariance matrix, K k is the Kalman gain matrix, H T is the transpose matrix of the observation matrix H; Step 1.3, encapsulate the displacement compensation amounts Δx, Δy, Δz and the corrected water level height h′ output by Step 1.2 to generate a data packet containing the node identifier nodeID, displacement vector (Δx, Δy, Δz), water level value h′ and timestamp t s After appending the CRC-16-CCITT checksum, the data packet is sent through the LoRaWAN protocol, and the sending parameters include the spreading factor SF10 and the bandwidth 125 kHz.

3. A remote warning method for water conservancy projects according to claim 1, characterized in that In Step 2, receive the data message in Step 1 and store it in a circular buffer, perform cumulative compensation on the displacement data based on the sliding mean filter algorithm, and output a calibrated vertex coordinate set in combination with preset terrain data, which further includes: Step 2.1: Receive the message in Step 1 and write it into the circular buffer: Define the capacity of the circular buffer as 60 seconds of historical data. The stored message fields include the node identifier nodeID, displacement compensation amounts Δx, Δy, Δz, corrected water level height h′, and timestamp t s ; Implement multi-threaded read-write synchronization through a memory lock, and update the write pointer to point to the latest data position; Step 2.2: Read the displacement data in the past 5 seconds from the circular buffer and calculate the displacement compensation amount after sliding mean filtering: Sliding window mean formula: Output mean shift Among them, Δx i , Δy i , Δz i are the output displacement compensation amounts. is the compensation amount of the sliding window mean displacement, N is the sliding window size, and k is the time step index; Step 2.3, calibrate the displacement mean with reference to the preset terrain point set D to generate the vertex coordinate set V t : The topographic point set D = {(x d , y d , h d )} contains preset longitude and latitude and elevation data; Plane coordinate constraint formula: y i ′ = y d corresponding to x i ′ , where x0 and y0 are the initial installation coordinates of the probe, and x i ′ , y i ′ are the plane coordinates of the probe after terrain constraint; Water level height correction: Output vertex set: V t = {(x ′ 1, y ′ 1, h ′ 1), …, (x ′ n , y ′ n , h ′ n )}, Among them, h i ′ is the corrected water level height, and V t is the set of vertex coordinates.

4. A remote warning method for a water conservancy project according to claim 1, characterized in that, In Step 3, according to the vertex coordinate set, construct a gridded water level matrix through the inverse distance weighted interpolation algorithm, detect abnormal grid data, and trigger data resampling, which further includes: Step 3.1, based on the set of vertex coordinates V output in Step 2 t and the preset terrain point set D, divide the monitoring area into 10 m × 10 m grids, define the number of grid rows M and the number of columns F, and generate the initial water level matrix H t ; Step 3.2: For each grid point (x, y), perform inverse distance weighted interpolation to calculate the water level value; Among them, the Euclidean distance from the grid point (x, y) to the vertex (x i ′ , y i ′ ). h i ′ is the corrected water level height of vertex i, and ∈ is a very small constant to prevent division by zero; Step 3.3: Detect abnormal grid data: Calculate the mean absolute deviation δ between the grid point H(x, y) and the water level values of the adjacent 8 points; Among them, H(x, y) is the water level height interpolated at the grid point (x, y), H(x j , y j ) is the water level height interpolated at the grid point (x j , y j ); If δ > 0.1·H(x, y), mark H(x, y) as abnormal data; If the proportion of abnormal data exceeds 5%, send a resampling request to Step 1.

5. A remote warning method for water conservancy projects according to claim 1, characterized in that, In Step 4, input the water level matrix into a pre-trained 1D convolutional neural network model to output the predicted water level value and confidence standard deviation at the next moment, which further includes: Step 4.1: Construct a historical water level matrix sequence: The water level matrix H output by step 3 t Extract continuous 5-frame historical data to form the input sequence X: X = [H t-4 , H t-3 , H t-2 , H t-1 , H t ; Among them, [H t-4 , H t-3 , H t-2 , H t-1 , H t is the historical water level matrix sequence; Perform normalization processing on the input sequence; Among them, X norm is the input sequence of the normalized historical water level matrix, μ is the mean of the training dataset, and σ is the standard deviation of the training dataset; Step 4.2: Perform inference through the 1D convolutional neural network model: Model structure: Z1 = ReLU(W1 * X norm + b1), Z2 = MaxPooling(Z1), Z3 = Flatten(Z2), Among them, Z1 is the output feature map of the convolutional layer, Z2 is the output feature map of the max pooling layer, Z3 is the feature vector after the flattening operation, W1 is the weight matrix of the convolutional layer, W2 is the weight matrix of the fully connected layer, and b1 and b2 are bias vectors. is the predicted water level matrix at the next moment output by the model; Output: Predicted water level matrix Step 4.3, calculate the standard deviation of confidence: Standard deviation formula: Among them, is the mean of the prediction matrix , G is the total number of grid points, and σ is the standard deviation of the confidence of the prediction result; If σ > 0.15, trigger the data resampling request in Step 1.

6. A remote warning method for water conservancy projects according to claim 1, characterized in that, In the said Step 5, generate a hierarchical early warning instruction according to the prediction value and the standard deviation of confidence, and send it to the execution terminal through a multi-protocol communication link to render the visualization interface, which further includes: Step 5.1, generate a hierarchical early warning instruction: Early warning threshold determination formula: Among them, ΔH is the deviation between the predicted water level peak and the current water level peak, and H t is the initial water level matrix, is the predicted water level matrix at the next moment output by the model, and σ is the standard deviation of the output confidence; Step 5.2, send the instruction through a multi-protocol communication link: MQTT protocol parameters: broker = iot.example.com, port = 1883, topic = water / alert / {nodeID}, wherein, broker is the target server address for sending the early warning instruction by the MQTT protocol, port is the port configuration of the communication link, and topic is the main body for publishing the instruction; Enable the QoS1 service quality level; Heartbeat detection mechanism: send a heartbeat packet to the edge node every 30 seconds, and determine the node to be offline if it times out 3 times; Step 5.3, render the visualization interface: 3D water level surface rendering: Render the current water level matrix H output in Step 3 based on the WebGL engine t , and overlay the predicted water level contour lines in Step 4.2 Color mapping rule: The mobile APP synchronously displays the water level curve of key nodes and the early warning level.

7. A remote warning method for water conservancy projects according to claim 1, characterized in that The data transmission between Step 1 and Step 2 adopts the LoRaWAN protocol and adds a CRC-16 check code; The prediction result transmission between Step 4 and Step 5 adopts the MQTT protocol and enables the QoS1 service quality level.

8. A remote warning system for water conservancy projects, based on the remote warning method for water conservancy projects according to any one of claims 1-7, characterized in that, The remote early warning system for water conservancy projects includes: Sensor node module, integrating a radar sensor, an inertial measurement module and a LoRaWAN communication unit; Edge positioning module, equipped with a circular buffer and a memory lock management unit; Water level modeling module, built-in inverse distance weighted interpolation algorithm and anomaly detection unit; Prediction module, deploying a 1D convolutional neural network model and a confidence calculation unit; Early warning visualization module, supporting hierarchical early warning decision-making, multi-terminal interface rendering and instruction sending.

9. A terminal device, characterized in that, It includes a processor, a memory and a communication interface. The memory stores a computer program, and when the processor executes the program, it implements a remote early warning method for water conservancy projects according to any one of claims 1-7.

10. A storage medium, characterized in that, Store computer-executable instructions, and when the instructions are executed by the processor, they implement a remote early warning method for water conservancy projects according to any one of claims 1-7.

Images