Intelligent water conservancy project patrol inspection method

Through fixed monitoring sites and deep learning models, potential pollution sources are predicted, combined with the collaborative operation of drones and underwater robots, the existing smart water conservancy engineering inspection methods are solved, and the comprehensive and efficient water conservancy engineering projects are achieved , accurate patrols have enhanced safety and management levels.

CN118937244AInactive Publication Date: 2024-11-12李成旭

Patent Information

Application Number
CN202410970671.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing smart water conservancy project inspection and inspection methods are inefficient and have great safety hazards. Traditional drone inspections mainly focus on the water surface, ignore the underwater environment, and make it difficult to effectively discover internal hidden dangers and potential problems of water conservancy projects.

Method used

Fixed monitoring sites are used to analyze historical water quality data in combination with deep learning models, predict potential pollution sources and high-risk areas, combine drones and underwater robots to work together to monitor and clean floating objects in real time, and generate detailed water quality reports through deep learning and big data analysis. .

Benefits of technology

It has achieved comprehensive, efficient and accurate inspections of water conservancy projects, improved the accuracy and efficiency of inspections, effectively discovered internal hidden dangers of water conservancy projects, and enhanced safety and management level.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118937244A_ABST
    Figure CN118937244A_ABST
Patent Text Reader

Abstract

The invention provides an intelligent water conservancy project patrol inspection method, and relates to the technical field of water conservancy projects, a water quality sensor is installed, data is monitored in real time and transmitted to a central control platform, the platform predicts potential pollution sources and pollution high-risk areas through deep learning, and an unmanned aerial vehicle executes refined patrol inspection. An optimal flight path is planned through a genetic algorithm, it is ensured that all important monitoring points and newly-recognized floating object positions are covered, a water area image is analyzed, the floating object positions are recognized and classified, a proper salvage mode is judged, and for floating objects suitable for being salvaged by an unmanned aerial vehicle, an undirected graph model is dynamically constructed by the system; the path is adjusted in real time to efficiently salvage and monitor the load, and the operation safety is ensured. The system integrates the Internet of Things, deep learning, genetic algorithm and real-time monitoring technology, and achieves the efficient and intelligent management of the water conservancy project.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy projects, and in particular to a smart water conservancy project inspection method. Background Art

[0002] Smart water conservancy project inspection refers to the use of intelligent technology and equipment to conduct regular inspections and tests on water conservancy projects to ensure their safe operation and effective management. With the continuous improvement and development of drone technology, drones have begun to be used in river inspections of water conservancy projects.

[0003] The document with prior art publication number CN117191483B provides a smart water conservancy project inspection method. Through the smart water conservancy project inspection system, a set of special mechanical devices is used to clean floating objects on the water surface of the detection point. After the floating objects are cleaned, the system samples the water at the detection point through a sampling mechanism, and the sampled water is sent to a collection mechanism for centralized collection.

[0004] However, this inspection method still has certain limitations. The cleaning of floating objects in the waters can only rely on manual operation or limited automated equipment, which is inefficient and poses safety risks. Traditional drone inspection methods often only focus on the situation on the water surface, while ignoring the importance of the underwater environment. Simple drone inspections can often only obtain surface information, and it is difficult to effectively discover hidden dangers and potential problems inside water conservancy projects.

[0005] In view of this, we proposed a smart water conservancy project inspection method. Summary of the invention

[0006] The present invention provides a smart water conservancy project inspection method, and the technical solution includes the following aspects.

[0007] According to one aspect of an embodiment of the present application, a method for inspecting and patrolling a smart water conservancy project is provided, the method comprising:

[0008] S1. Deploy fixed monitoring stations, install water quality sensors, collect water quality data in real time and transmit it to the central control platform, use deep learning models to analyze historical water quality data, predict potential pollution sources and high-risk pollution areas, and set detection parameters and early warning mechanisms;

[0009] S2. Based on the prediction results of historical water quality data, potential pollution sources and high-risk pollution areas are identified, and water sampling points are set up in these areas. The initial path is planned for the drone to ensure that all high-risk areas and collection points are covered. The drone conducts detailed inspections along the planned path, equipped with high-precision cameras and multi-spectral sensors to confirm the pollution status, and collects water samples in high-risk areas and predetermined collection points;

[0010] S3, based on the water area images obtained by the drone, identify the location of floating objects in the water area and classify them, and determine whether the floating objects are suitable for drone salvage, underwater robot salvage or manual salvage;

[0011] S4. If it is suitable for drone salvage, based on the location of the floating object, an additional dynamic undirected graph model is constructed on the basis of the original path planning, and new floating object nodes are added to re-plan the path. After the floating object is salvaged, the remaining volume and carrying weight are calculated. When the threshold is reached, the drone returns and the path is re-planned based on the remaining node positions;

[0012] S5. If it is suitable for the underwater robot, the underwater robot plans a path according to the salvaged floating objects and their positions, monitors the volume and weight of the salvaged floating objects in real time, and plans the path of the underwater robot in real time;

[0013] S6. After the water samples are collected, they are transported to the laboratory for analysis. The water quality data is uploaded to the central platform via a wireless network. Big data and deep learning models are used to comprehensively analyze historical water quality data and new water quality data, identify high-risk areas for pollution, and generate a detailed water quality report.

[0014] S7. Users can view the patrol status in real time through the platform and adjust the water management strategy or deploy emergency measures based on the analysis results.

[0015] Among them, the deep learning model analyzes historical water quality data:

[0016] Through the integration of the Internet of Things platform and the geographic information system, fixed monitoring stations are deployed, and water quality sensors are installed at the stations to collect water quality data in real time;

[0017] The sensor sends water quality data to the central control platform in real time through the Internet of Things technology, and the collected raw data is processed;

[0018] Use deep neural network models for feature extraction to automatically learn feature vectors of water quality changes from historical water quality data;

[0019] Input the real-time collected water quality data into the trained deep learning model to predict the current water quality status and future change trends;

[0020] The model outputs the predicted water quality value of each monitoring point in a certain period of time in the future, and identifies potential high-risk pollution areas by setting thresholds or using cluster analysis.

[0021] Among them, set the threshold:

[0022] When using set thresholds to identify potential high-risk areas for pollution, determine the safety range and warning level of each water quality indicator based on environmental standards or historical water quality data;

[0023] The water quality data predicted by the model is compared with the set threshold. If the predicted value of a monitoring point in a certain period of time in the future exceeds the warning line, the point will be marked as a potential pollution risk point.

[0024] Among them, when cluster analysis is used to identify potential high-risk areas for pollution, the future water quality data predicted by the model are converted into feature vectors, with each monitoring point corresponding to one feature vector;

[0025] The vector elements are the predicted values ​​of various water quality indicators. The DBSCAN algorithm is used to identify those areas where water quality indicators are abnormally concentrated in the predicted future period;

[0026] Abnormally concentrated areas are high-risk areas for pollution, indicating that water quality parameters are concentrated in abnormally high concentrations or low quality in space.

[0027] Among them, planning the initial path: according to the historical water quality data prediction results and real-time monitoring data in S1, determine the water sample collection points, which cover all high-risk areas and potential pollution sources to ensure that the collected water samples are representative and timely. Use genetic algorithms to plan the path of the patrol drone. For the patrol drone, it is necessary to visit N preset collection points, and randomly generate an initial solution set (population). Each solution (individual) represents a possible flight path sequence. Define a fitness function to evaluate the quality of each individual (path). The fitness function takes into account the flight distance, endurance limit, sampling point reachability, and camera coverage area;

[0028] The fitness function is, F(P) = w1×(1 / TotalDistance(P))-w2×Penalty(P)+w3×CoverageScore(P), where TotalDistance(P) is the total flight distance of the path represented by individual P, Penalty(P) is the penalty term for not covering all points or violating the endurance limit, which is 0 if all conditions are met, and CoverageScore(P) measures the coverage of the water area by the camera's view under path P, which may be achieved by calculating the actual coverage ratio of the grid or directly calculating the penalty for the uncovered area. w1, w2, and w3 are weight factors, balancing the importance of minimizing the distance and satisfying the constraints;

[0029] Individuals are selected proportionally according to their fitness values ​​to enter the next generation. Individuals with high fitness have a higher probability of being selected. This is achieved through the roulette wheel selection method. Two individuals are randomly selected and part of their chromosomes are exchanged to generate new individuals, such as single-point crossover, multi-point crossover, etc. One or several genes of the individual are randomly changed with a small probability to increase diversity. The selection, crossover, and mutation operations are repeated until the predetermined number of iterations is reached or a satisfactory solution is found. The criterion for a satisfactory solution is that the fitness function reaches a certain threshold or there is no significant improvement in several consecutive generations.

[0030] Among them, identify the location and classification of floating objects in the water area: use the target detection model YOLOv4 to determine the location of floating objects from the water area images taken by the inspection drone and classify the floating objects. Collect and annotate a large amount of training data. The data should contain pictures of various floating objects. The floating objects on each picture need to be accurately marked with their positions. Use bounding box coordinates for position annotation and integers to annotate the categories. After training, the model is used for real-time detection. For each detected object, the model outputs its category probability distribution and bounding box position. According to the highest category probability, the bounding box is marked with the corresponding color to achieve classification and positioning of different salvage methods, and the identified floating object location data is transmitted to the central control platform in real time.

[0031] Among them, on the basis of the original path obtained by genetic algorithm, an additional dynamic undirected graph model is constructed to reflect the actual situation of the water area. The dynamic undirected graph model includes the original water sample collection points and the locations of newly appeared floating objects as nodes. The weight of the edge is dynamically calculated based on factors such as the complexity of the water area terrain, the actual flight distance between two points, and the wind direction and speed.

[0032] Assume graph G = (V, E), where V is a node set, including N fixed collection points and dynamically increasing floating object locations;

[0033] For an edge e between any two points u and v uv , whose weight w uv Calculated according to actual flight conditions, w uv =d uv +α·terrainFactor uv +β·windImpact uv , where d uv is the Euclidean distance between two points, t errainFactor uv Reflects the influence of the complexity of water terrain, windImpact uv is the factor affecting the flight efficiency of wind direction and speed, while α and β are adjustment coefficients;

[0034] For the shortest path from the real-time position c of the drone to the next target point t, the Dijkstra algorithm is used to minimize the path cost. The core formula is, dist[u] = min v∈Adj(u) (w uv +dist[v]), for all u≠t, where dist[u] represents the shortest path length from the starting point to node u, and Adj(u) is the set of adjacent nodes of node u;

[0035] The drone’s carrying weight, volume and remaining battery are monitored in real time. When the threshold is reached, the drone returns. Based on the latest water map, the genetic algorithm is used again to plan the flight path for the remaining unvisited points to ensure efficient and continuous execution of the entire mission.

[0036] Among them, sonar or underwater camera equipment is used to conduct preliminary mapping of the underwater environment to collect key information such as terrain, water flow speed and direction, and water depth. The Rao-Blackwellized particle filter (RBPF) is applied to accurately estimate the position of the underwater robot. The Simultaneous Localization and Mapping (SLAM) technology is used to allow the robot to explore and build a map in an unknown environment, and update its own position and surrounding environment model in real time. Pressure sensors, volume sensing devices, and weight sensors are integrated on the underwater robot to monitor the physical properties of the salvaged object in real time. For the floating objects suitable for salvage by underwater robots identified in step S3, after receiving the task, the underwater robot plans the optimal path from the current position to the specified target area based on the D*Lite algorithm. The D*Lite algorithm takes into account the dynamic changes of the underwater environment and can adjust the path in real time to adapt to environmental changes to ensure that the robot can efficiently and safely reach the target area. The underwater robot uses its onboard sonar and underwater camera equipment to accurately locate the exact position of floating objects and conduct detailed identification to prepare for salvage. Based on the identification results, the underwater robot performs salvage operations, which may include using a robotic arm to grab or special salvage tools to fix and lift floating objects. The sensors on the robot monitor its own load and energy status in real time to ensure operational safety. During the salvage process, the underwater robot continuously feeds back the operating status to the central control platform. If encountering special circumstances (such as load approaching the limit, insufficient energy, etc.), the robot will be guided to adjust its strategy based on real-time information and choose the optimal path to return to the base or for temporary charging / unloading.

[0037] Among them, the water samples collected at each sampling point are tested for physical, chemical and biological indicators, and the analysis results are quantified into specific values ​​to form a basic data set. The data obtained from the laboratory analysis are uploaded to the central data platform in real time through wireless network technology. The platform standardizes the data, eliminates dimensional differences, and forms a database in a unified format. Meaningful features are extracted from the original data, and long short-term memory neural networks and convolutional neural networks are used for analysis to determine the level of water quality and the changing trends in time and space. Statistical methods or autoencoders are used to identify abnormal points in water quality data to indicate potential pollution events. According to the analysis results, a detailed report containing water quality overview, changing trends, and abnormal area locations is automatically generated.

[0038] According to one aspect of an embodiment of the present application, a smart water conservancy project inspection system is provided, the system comprising:

[0039] Fixed monitoring station module, which includes multi-parameter water quality sensors and data transmission equipment, used to collect water quality data in real time and transmit it to the central control platform;

[0040] The central control platform module is used to receive, process and analyze data from fixed monitoring sites and inspection equipment, use deep learning models to predict potential pollution sources and high-risk pollution areas, and generate corresponding early warning information;

[0041] The drone module plans routes and performs detailed inspections, water sampling and floating object salvage tasks according to the instructions of the central control platform, and is equipped with high-precision cameras and multi-spectral sensors to confirm pollution conditions;

[0042] The underwater robot module identifies and salvages floating objects in the water according to the instructions of the central control platform;

[0043] The laboratory analysis module is used to conduct laboratory analysis on water samples collected by inspection drones to obtain detailed water quality data;

[0044] The user interface module allows users to view the inspection status, water quality reports and governance suggestions in real time through the central control platform, and adjust the water area management strategy or deploy emergency measures based on the analysis results and suggestions.

[0045] Through the implementation of the above-mentioned smart water conservancy project inspection method and system, the intelligent and refined inspection of water conservancy projects can be realized, the accuracy and efficiency of inspection can be improved, the hidden dangers and potential problems inside the water conservancy projects can be effectively discovered, and strong support can be provided for the safe operation and effective management of water conservancy projects.

[0046] The smart water conservancy project inspection method of the present invention has the following advantages:

[0047] Through the combination of fixed monitoring stations, drone systems and underwater robot systems, comprehensive, efficient and accurate inspections of water conservancy projects are achieved, improving the safety and management level of water conservancy projects. At the same time, deep learning models and big data analysis technologies are used to conduct comprehensive analysis of historical water quality data and new water quality data, predict potential pollution sources and high-risk pollution areas, and provide a scientific basis for water area management. In addition, the present invention also realizes the automatic identification and cleaning of floating objects on the water surface through the collaborative operation of drones and underwater robots, improving the efficiency and intelligence level of inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of a smart water conservancy project inspection method provided for one embodiment of the present application;

[0049] Figure 2 A structural diagram of a smart water conservancy project inspection system provided for one embodiment of the present application; DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0051] Please refer to Figure 1 , which shows a flow chart of a smart water conservancy project inspection method provided by an embodiment of the present application;

[0052] S1. Deploy fixed monitoring stations, install multi-parameter water quality sensors, collect water quality data in real time and transmit it to the central control platform, use deep learning models to analyze historical water quality data, predict potential pollution sources and high-risk pollution areas, and set detection parameters and early warning mechanisms;

[0053] S2. Based on the prediction results of historical water quality data, potential pollution sources and high-risk pollution areas are identified, and water sampling points are set up in these areas. The initial path is planned for the drone to ensure that all high-risk areas and collection points are covered. The drone conducts detailed inspections along the planned path, equipped with high-precision cameras and multi-spectral sensors to confirm the pollution status, and collects water samples in high-risk areas and predetermined collection points;

[0054] S3, based on the water area images obtained by the drone, identify the location of floating objects in the water area and classify them, and determine whether the floating objects are suitable for drone salvage, underwater robot salvage or manual salvage;

[0055] S4. Based on the original path planning, a dynamic undirected graph model is constructed according to the location of floating objects suitable for drone salvage. New floating object nodes are added to re-plan the path. After the floating objects are salvaged, the remaining volume and carrying weight are calculated. When the threshold is reached, the drone returns and the path is re-planned according to the remaining node positions.

[0056] S5. The underwater robot plans a path according to the location and suitable position for the underwater robot to salvage floating objects, monitors the volume and weight of the salvaged floating objects in real time, and plans the path of the underwater robot in real time;

[0057] S6. After the water samples are collected, they are transported to the laboratory for analysis. The water quality data is uploaded to the central platform via a wireless network. Big data and deep learning models are used to comprehensively analyze historical water quality data and new water quality data, identify high-risk areas for pollution, and generate a detailed water quality report.

[0058] S7. Users can view the patrol status in real time through the platform and adjust the water management strategy or deploy emergency measures based on the analysis results.

[0059] As an optional solution of the present application, in step S1, the Internet of Things platform is integrated with GIS (Geographic Information System), and fixed monitoring stations are scientifically deployed according to the terrain of the water area, water flow characteristics and historical pollution records. The stations are equipped with a variety of water quality sensors, such as dissolved oxygen sensors, turbidity sensors, pH sensors, temperature sensors, etc., which can monitor multiple indicators of the water body in real time. Each sensor sends water quality data to the central control platform in real time through the Internet of Things technology. The data includes but is not limited to temperature (T), dissolved oxygen (DO), pH value, conductivity (EC), turbidity (NTU), etc. The collected raw data is cleaned and normalized, and converted into a format suitable for deep learning model input. The normalization formula is x′=x-μ / σ, where x is the raw data, μ is the mean, σ is the standard deviation, and x′ is the normalized data. The deep neural network (DNN) model is used for feature extraction to automatically learn the feature vector of water quality changes from historical water quality data. FeatureVector=f θ (X raw ; W b ), where f θ represents the neural network model, X raw is the original input data, W b is the model weight, θ is the model parameter, the model structure is, the input layer receives multi-dimensional water quality parameter data, the hidden layer contains multiple fully connected layers, the ReLU activation function is used, the output layer predicts the change trend of water quality parameters, the activation function uses the linear activation function f(x) = x, the goal of model training is to minimize the loss function, such as the mean square error (MSE), the formula is: min θL(Y^,Y)=min θ ∑ i=1 N (Y i -Y^ i ) 2 , where Y^ is the model prediction value, Y is the actual observation value, N is the number of samples, and the model parameters are updated using gradient descent. Among them, η is the learning rate. The water quality data collected in real time is input into the trained deep learning model to predict the current water quality status and future change trends. The model outputs the predicted water quality value of each monitoring point in a certain period of time in the future, and identifies potential high-risk pollution areas by setting thresholds or using cluster analysis (such as DBSCAN).

[0060] As an optional solution of the present application, when using set thresholds to identify potential high-risk areas for pollution, the safety range and warning level of various water quality indicators are determined based on environmental protection standards or historical water quality data. For example, dissolved oxygen concentration below 5 mg / L, ammonia nitrogen concentration exceeding 1 mg / L, etc. may be considered as standards for poor water quality. The water quality data predicted by the model is compared with the set threshold. If the predicted value of a monitoring point in a certain period of time in the future exceeds the warning line, the point is marked as a potential pollution risk point, and the system automatically issues an early warning for the marked risk point.

[0061] As an optional solution of the present application, when cluster analysis is used to identify potential high-risk areas for pollution, the future water quality data predicted by the model is converted into a feature vector. Each monitoring point corresponds to a feature vector, and the vector elements are the predicted values ​​of various water quality indicators. The DBSCAN algorithm is used to identify areas where water quality indicators are abnormally clustered in the predicted future period. DBSCAN is a density-based spatial clustering algorithm that does not require the number of clusters to be set in advance. Clusters are divided by defining the number of points in the neighborhood (MinPts) and the distance threshold (ε). Abnormally clustered areas are high-risk areas for pollution, indicating that water quality parameters are concentrated in space with abnormally high concentrations or low quality.

[0062] As an optional solution of the present application, in step S2, water sampling points are determined based on the historical water quality data prediction results and real-time monitoring data in S1. The water sampling points cover all high-risk areas and potential pollution sources to ensure that the collected water samples are representative and timely. A genetic algorithm is used to plan the path of the patrol drone. For the patrol drone, it is necessary to visit N preset collection points. An initial solution set (population) is randomly generated, and each solution (individual) represents a possible flight path sequence. Assume that an individual is represented by a chromosome, where the chromosome length is N, and each gene corresponds to the access order of a collection point, represented by an integer index, P = {p1, p2, ..., p i ,...,pN}, where P is an individual, p i represents the serial number of the i-th visited collection point. The fitness function is defined to evaluate the quality of each individual (path). The fitness function takes into account the flight distance, endurance limit, sampling point reachability and camera coverage area. The camera coverage area uses the overlapping grid method to divide the water area into small grids. The number of grids that can be covered by the perspective of each preset point is calculated to ensure that all grids are covered by at least one or more points. The fitness function is, F(P) = w1×(1 / TotalDistance(P))-w2×P enalty(P)+w3×CoverageScore(P), where TotalDistance(P) is the total flight distance of the path represented by individual P, Penalty(P) is the penalty for not covering all points or violating the endurance limit, which is 0 if all conditions are met. CoverageScore(P) measures the coverage of the water area by the camera's view under path P, which may be achieved by calculating the actual coverage ratio of the grids or directly calculating the penalty for the uncovered area. w1, w2, and w3 are weight factors, balancing the importance of minimizing distance and satisfying constraints. Individuals are selected proportionally according to their fitness values ​​to enter the next generation. Individuals with high fitness have a higher probability of being selected. This is achieved through the roulette wheel selection method, which randomly selects two individuals and exchanges part of their chromosomes to generate new individuals, such as single-point crossover, multi-point crossover, etc., randomly changing one or several genes of the individual with a small probability to increase diversity, and repeating the selection, crossover, and mutation operations until the predetermined number of iterations is reached or a satisfactory solution is found. The criterion for a satisfactory solution is that the fitness function reaches a certain threshold or there is no significant improvement in consecutive generations.

[0063] As an optional solution of the present application, in step S3, the target detection model YOLOv4 is used to determine the location of floating objects from the water images taken by the inspection drone and classify the floating objects. A large amount of training data is collected and annotated. The data should contain pictures of various floating objects. The floating objects on each picture need to be accurately marked with their positions. The bounding box coordinates are used for position annotation, denoted as B = (x1, y1, x2, y2), where (x1, y1) are the coordinates of the upper left corner of the bounding box, and (x2, y2) are the coordinates of the lower right corner. Integers are used to mark the category, such as 1 represents salvage using a salvage drone, 2 represents salvage using an underwater robot, and 3 requires manual salvage. The target detection task involves two main tasks, classification and positioning (bounding box regression). The loss function is a combination of the two losses, L = α·L cls +β·L reg , where L cls For classification loss, we use the cross entropy loss function, L regFor positioning loss, a smooth L1 loss function is used. α and β are weight coefficients that balance classification and positioning losses. The network parameters are updated through back propagation and gradient descent to minimize the loss function L. In each iteration, a batch of data (called mini-batch) is randomly selected, the loss is calculated, and the parameters are updated. After training, the model is applied to real-time detection. For each detected object, the model outputs its category probability distribution and bounding box position. According to the highest category probability, the bounding box is marked with the corresponding color to achieve classification and positioning of different salvage methods, and the identified floating object location data is transmitted to the central control platform in real time.

[0064] As an optional solution of the present application, in step S4, based on the original path obtained by the genetic algorithm, an additional dynamic undirected graph model is constructed to reflect the actual situation of the water area. The dynamic undirected graph model includes the original water sample collection points and the newly appeared floating object positions as nodes. The weight of the edge is dynamically calculated based on factors such as the complexity of the water area terrain, the actual flight distance between the two points, the wind direction and speed, etc. Suppose the graph G = (V, E), where V is a node set, including N fixed collection points and dynamically added floating object positions. For the edge e between any two points u and v uv , whose weight w uv Calculated according to actual flight conditions, w uv =d uv +α·terrainFactor uv +β·windImpact uv , where d uv is the Euclidean distance between two points, terrainFact or uv Wind Impact reflects the impact of the complexity of the water terrain (which can be quantified as the additional cost of crossing a specific area). uv is the influencing factor of wind direction and speed on flight efficiency, while α and β are adjustment coefficients. For the shortest path from the real-time position c of the drone to the next target point t, the Dijkstra algorithm is used to minimize the path cost. The core formula is, dist[u] = min v∈Adj(u) (w uv +dist[v]), for all u≠t, where dist[u] represents the shortest path length from the starting point to node u, and Adj(u) is the set of adjacent nodes of node u. The weight, volume, and remaining power of the drone are monitored in real time. When the threshold is reached, the drone returns. According to the latest water map, the genetic algorithm is used again to plan the flight path of the remaining unvisited points to ensure the efficient and continuous execution of the entire mission.

[0065] As an optional solution of the present application, in step S5, a sonar or underwater camera device is used to perform preliminary mapping of the underwater environment to collect key information such as terrain, water flow speed and direction, and water depth, and a Rao-Blackwellized particle filter (RBPF) is applied to accurately estimate the position of the underwater robot. The algorithm combines the flexibility of the particle filter with the accuracy of the Kalman filter, and is particularly suitable for state estimation in nonlinear and non-Gaussian environments. Mapping (SLAM) technology allows the robot to explore and build maps in an unknown environment, update its own position and surrounding environment model in real time, integrate pressure sensors, volume sensors and weight sensors on the underwater robot, and monitor the physical properties of the salvaged object in real time. For the floating objects suitable for underwater robot salvage identified in step S3, after receiving the task, the underwater robot plans the optimal path from the current position to the specified target area based on the D*Lite algorithm. The D*Lite algorithm considers the dynamic changes of the underwater environment (such as water flow direction, speed, water pressure, obstacles, etc.), and can adjust the path in real time to adapt to environmental changes to ensure that the robot can reach the target area efficiently and safely. Let the state space be S, each state s∈S represents the possible position and explored information of the robot, h(s) estimates the minimum cost from state s to the target state, based on the estimation of Euclidean distance, considering the complexity of the underwater environment, h(s)=d Euclidean (s,Goal)+γterrainFactor(s), where d Euclidean is the Euclidean distance from the current position of the robot to the target point, γ is the terrain complexity factor, and terrainFactor(s) reflects the underwater terrain complexity of the current state position. When the robot moves and the environment changes, g(s) (the actual cost from the starting point to state s) and h(s) are updated according to the observed new information to ensure that the path is always optimal. Sonar echo ranging and underwater camera image recognition technology are used to jointly locate floating objects. Assuming the sonar echo time is t and the sound speed is c, the target distance d = (c·t) / 2. The outline of the floating object recognized by the camera matches the estimated size. IOU (intersection over union) is used as the similarity metric, IOU = |B pred ∩B actual ∣ / ∣B pred ∪B actual ∣, where B pred and B actual are the predicted bounding box and the actually recognized bounding box respectively. Let the current state of the robot be s t , the power and load status is E t ,L t , when E t <E min or L t>L max When the path is re-planned, the nearest charging / unloading point P is found. return ,cost(s t ,P return )=min p ath∈Paths(st,Preturn) ∑ i=1 n g(path i ), where Paths represents the set of all possible paths, n is the number of states in the path, and the path with the minimum total cost is selected.

[0066] As an optional solution of the present application, in step S6, the water samples collected at each sampling point are tested for physical, chemical and biological indicators, including but not limited to temperature, pH value, dissolved oxygen (DO), chemical oxygen demand (COD), ammonia nitrogen (NH3-N), total phosphorus (TP), total nitrogen (TN), etc. The analysis results are quantified into specific values ​​to form a basic data set D = {d1, d2, ..., d n}, where each d i The multi-dimensional water quality index vector representing a sampling point is uploaded to the central data platform in real time through wireless network technology (such as 4G / 5G, LoRaWAN). The platform standardizes the data, eliminates dimensional differences, forms a database in a unified format, extracts meaningful features from the raw data, such as time series trends, spatial distribution patterns, etc., and constructs a feature vector F = {f1, f2, ..., f m}, use long short-term memory neural network and convolutional neural network for analysis, use historical water quality data as training set, label with known pollution events or water quality level, apply loss function for optimization, adjust model parameters through back propagation, determine water quality level, and temporal and spatial variation trend, use statistical methods (such as Z-score) or autoencoder to identify abnormal points in water quality data, indicating potential pollution events, and automatically generate detailed report containing water quality overview, variation trend, and location of abnormal areas based on the analysis results.

[0067] Please refer to Figure 2 , which shows a structure diagram of a smart water conservancy project inspection system provided by an embodiment of the present application;

[0068] Fixed monitoring station module, which includes multi-parameter water quality sensors and data transmission equipment, used to collect water quality data in real time and transmit it to the central control platform;

[0069] The central control platform module is used to receive, process and analyze data from fixed monitoring sites and inspection equipment, use deep learning models to predict potential pollution sources and high-risk pollution areas, and generate corresponding early warning information;

[0070] The drone module plans routes and performs detailed inspections, water sampling and floating object salvage tasks according to the instructions of the central control platform, and is equipped with high-precision cameras and multi-spectral sensors to confirm pollution conditions;

[0071] The underwater robot module identifies and salvages floating objects in the water according to the instructions of the central control platform;

[0072] The laboratory analysis module is used to conduct laboratory analysis on water samples collected by inspection drones to obtain detailed water quality data;

[0073] The user interface module allows users to view the inspection status, water quality reports and governance suggestions in real time through the central control platform, and adjust the water area management strategy or deploy emergency measures based on the analysis results and suggestions.

Claims

1. A smart water conservancy project inspection method, characterized in that: The method comprises: S1. Deploy fixed monitoring stations, install water quality sensors, collect water quality data in real time and transmit it to the central control platform, use deep learning models to analyze historical water quality data, predict potential pollution sources and high-risk pollution areas, and set detection parameters and early warning mechanisms; S2. Based on the prediction results of historical water quality data, potential pollution sources and high-risk pollution areas are identified, and water sampling points are set up in these areas. The initial path is planned for the drone to ensure that all high-risk areas and collection points are covered. The drone conducts detailed inspections along the planned path, equipped with high-precision cameras and multi-spectral sensors to confirm the pollution status, and collects water samples in high-risk areas and predetermined collection points; S3, based on the water area images obtained by the drone, identify the location of floating objects in the water area and classify them, and determine whether the floating objects are suitable for drone salvage, underwater robot salvage or manual salvage; S4. If it is suitable for drone salvage, based on the location of the floating object, an additional dynamic undirected graph model is constructed on the basis of the original path planning, and new floating object nodes are added to re-plan the path. After the floating object is salvaged, the remaining volume and carrying weight are calculated. When the threshold is reached, the drone returns and the path is re-planned based on the remaining node positions; S5. If it is suitable for the underwater robot, the underwater robot plans a path according to the salvaged floating objects and their positions, monitors the volume and weight of the salvaged floating objects in real time, and plans the path of the underwater robot in real time; S6. After the water samples are collected, they are transported to the laboratory for analysis. The water quality data is uploaded to the central platform via a wireless network. Big data and deep learning models are used to comprehensively analyze historical water quality data and new water quality data, identify high-risk areas for pollution, and generate a detailed water quality report. S7. Users can view the patrol status in real time through the platform and adjust the water management strategy or deploy emergency measures based on the analysis results.

2. A smart water conservancy project inspection method according to claim 1, characterized in that: Deep learning models analyze historical water quality data: Through the integration of the Internet of Things platform and the geographic information system, fixed monitoring stations are deployed, and water quality sensors are installed at the stations to collect water quality data in real time; The sensor sends water quality data to the central control platform in real time through the Internet of Things technology, and the collected raw data is processed; Use deep neural network models for feature extraction to automatically learn feature vectors of water quality changes from historical water quality data; Input the real-time collected water quality data into the trained deep learning model to predict the current water quality status and future change trends; The model outputs the predicted water quality value of each monitoring point in a certain period of time in the future, and identifies potential high-risk pollution areas by setting thresholds or using cluster analysis.

3. A smart water conservancy project inspection method according to claim 2, characterized in that: Setting the threshold: When using set thresholds to identify potential high-risk areas for pollution, determine the safety range and warning level of each water quality indicator based on environmental standards or historical water quality data; The water quality data predicted by the model is compared with the set threshold. If the predicted value of a monitoring point in a certain period of time in the future exceeds the warning line, the point will be marked as a potential pollution risk point.

4. A smart water conservancy project inspection method according to claim 2, characterized in that: Cluster analysis: When cluster analysis is used to identify potential high-risk areas for pollution, the future water quality data predicted by the model are converted into feature vectors, with one feature vector corresponding to each monitoring point; The vector elements are the predicted values ​​of various water quality indicators. The DBSCAN algorithm is used to identify those areas where water quality indicators are abnormally concentrated in the predicted future period; Abnormally concentrated areas are high-risk areas for pollution, indicating that water quality parameters are concentrated in abnormally high concentrations or low quality in space.

5. A smart water conservancy project inspection method according to claim 1, characterized in that: Plan the initial path: Determine the water sampling point based on the historical water quality data prediction results and real-time monitoring data in S1; A genetic algorithm is used to plan the path of the patrol drone. For the preset collection points that the patrol drone needs to visit, an initial solution set is randomly generated, and each solution represents a possible flight path sequence; Define a fitness function to evaluate the quality of each individual. The fitness function takes into account the flight distance, endurance limit, sampling point reachability, and camera coverage area. The fitness function is, F = w1 × (1 / TotalDistance-w2 × Penalty + w3 × CoverageScore, where TotalDistance is the total flight distance of the path represented by individual P, Penalty is the penalty term for not covering all points or violating the endurance limit, which is 0 if all conditions are met, and CoverageScore measures the coverage of the water area by the camera view under path P, which may be achieved by calculating the actual coverage ratio of the grid or directly calculating the penalty for the uncovered area. w1, w2, and w3 are weight factors, balancing the importance of minimizing the distance and satisfying the constraints; Individuals are selected proportionally according to their fitness values ​​to enter the next generation. Individuals with high fitness have a higher probability of being selected. This is achieved through the roulette wheel selection method. Two individuals are randomly selected, and part of their chromosomes are exchanged to generate new individuals. The selection, crossover, and mutation operations are repeated until the predetermined number of iterations is reached or a satisfactory solution is found. The criterion for a satisfactory solution is that the fitness function reaches a certain threshold or there is no significant improvement in several consecutive generations.

6. A smart water conservancy project inspection method according to claim 1, characterized in that: Identify and classify floating objects in the water: Use the object detection model YOLOv4 to locate and classify floating objects from images of waters taken by inspection drones; Collect and annotate a large amount of training data. The data should include pictures of various floating objects. The floating objects in each picture need to be accurately marked with their positions, using bounding box coordinates for position annotation and integers for category annotation. After training, the model is used for real-time detection. For each detected object, the model outputs its category probability distribution and bounding box location. Based on the highest category probability, the bounding box is marked with the corresponding color to achieve classification and positioning of different salvage methods, and the identified floating object location data is transmitted to the central control platform in real time.

7. A smart water conservancy project inspection method according to claim 1, characterized in that: The original path planning foundation additionally builds a dynamic undirected graph model: Based on the original path obtained by genetic algorithm, an additional dynamic undirected graph model is constructed to reflect the actual situation of the water area; The dynamic undirected graph model includes the original water sample collection points and the newly appeared floating object locations as nodes, and the edge weights are dynamically calculated based on the complexity of the water terrain, the actual flight distance between the two points, and the wind direction and speed factors; Assume graph G = , where V is a node set, including N fixed collection points and dynamically increasing floating object locations; For an edge e between any two points u and v uv , whose weight w uv Calculated according to actual flight conditions, w uv =d uv +α·terrainFactor uv +β·windImpact uv , where d uv is the Euclidean distance between two points, t errainFactor uv Reflects the influence of the complexity of the water terrain, windImpact uv is the factor affecting the flight efficiency of wind direction and speed, while α and β are adjustment coefficients; For the shortest path from the real-time position c of the drone to the next target point t, the Dijkstra algorithm is used to minimize the path cost. The core formula is, dist[u] = min v∈Adj ,for all u≠t, where dist[u] represents the shortest path length from the starting point to node u, and Adj is the set of adjacent nodes of node u; The drone’s carrying weight, volume and remaining battery are monitored in real time. When the threshold is reached, the drone returns. Based on the latest water map, the genetic algorithm is used again to plan the flight path for the remaining unvisited points to ensure efficient and continuous execution of the entire mission.

8. A smart water conservancy project inspection method according to claim 1, characterized in that: Underwater robot planning path: Use sonar or underwater camera equipment to conduct preliminary mapping of the underwater environment, accurately estimate the position of the underwater robot, and update its own position and surrounding environment model in real time; Integrate pressure sensors, volume sensors and weight sensors on the underwater robot to monitor the physical properties of the salvaged objects in real time; For the floating objects suitable for underwater robot salvage identified in step S3, after receiving the task, the underwater robot plans the optimal path from the current position to the specified target area based on the D*Lite algorithm; The underwater robot uses its onboard sonar and underwater camera equipment to accurately locate the exact location of the floating object and conduct detailed identification to prepare for salvage; The sensors on the robot monitor its own load and energy status in real time to ensure operational safety. During the salvage process, the underwater robot continuously feeds back the operating status to the central control platform. If the load is close to the limit or the energy is insufficient, the robot will be guided to adjust its strategy based on real-time information and choose the optimal path to return to the base or for temporary charging / unloading.

9. A smart water conservancy project inspection method according to claim 1, characterized in that: Comprehensive analysis: The water samples collected at each sampling point are tested for physical, chemical and biological indicators, and the analysis results are quantified into specific values ​​to form a basic data set; Through wireless network technology, the data obtained from laboratory analysis is uploaded to the central data platform in real time, and the platform processes the data; Extract meaningful features from raw data and use long short-term memory neural network and convolutional neural network to analyze and determine the level of water quality and its temporal and spatial trends; Use statistical methods or autoencoders to identify anomalies in water quality data, indicating potential contamination events; Based on the analysis results, a detailed report containing water quality overview, change trends, and abnormal area locations is automatically generated.

10. A smart water conservancy project inspection system, characterized in that: The system comprises: Fixed monitoring station module, which includes multi-parameter water quality sensors and data transmission equipment, used to collect water quality data in real time and transmit it to the central control platform; The central control platform module is used to receive, process and analyze data from fixed monitoring sites and inspection equipment, use deep learning models to predict potential pollution sources and high-risk pollution areas, and generate corresponding early warning information; The drone module plans routes and performs detailed inspections, water sampling and floating object salvage tasks according to the instructions of the central control platform, and is equipped with high-precision cameras and multi-spectral sensors to confirm pollution conditions; The underwater robot module identifies and salvages floating objects in the water according to the instructions of the central control platform; The laboratory analysis module is used to conduct laboratory analysis on water samples collected by inspection drones to obtain detailed water quality data; The user interface module allows users to view the inspection status, water quality reports and governance suggestions in real time through the central control platform, and adjust the water area management strategy or deploy emergency measures based on the analysis results and suggestions.

Citation Information

Patent Citations

  • A smart water conservancy project inspection method

    CN117191483B

Cited By

  • Mobile water quality sampling system for plateau or high mountain area

    CN119198211A

  • Industrial internet security monitoring system and method

    CN119356197A

  • Respiration data reminding system for patient with critical respiratory disease

    CN119655714A

  • Water area intelligent protection monitoring system based on iron tower visual connection platform

    CN119668300A

  • Wetland vegetation intelligent monitoring method and system based on multi-dimensional data fusion

    CN120014461A