Assistant decision-making platform for water conservancy project operation and maintenance based on AI unmanned aerial vehicle

Abstract

The invention relates to the technical field of hydraulic engineering intelligent operation and maintenance, and discloses an AI unmanned aerial vehicle-based auxiliary decision-making platform for hydraulic engineering operation and maintenance, and the platform comprises a data collection module which is used for obtaining multi-modal data of a water conservancy facility through an AI unmanned aerial vehicle, the multi-modal data comprises a visible light image, thermal imaging data, multispectral data and laser radar data; the data processing and fusion module is used for carrying out formatting processing, space-time alignment and feature fusion on the multi-modal data; the health assessment and prediction module is used for assessing the health state of the water conservancy facility based on the processed data and predicting the future change trend of the facility state; and the auxiliary decision-making module is used for generating inspection priority planning of the water conservancy facilities. According to the invention, through multi-source data fusion, dynamic risk assessment and intelligent optimization distribution, the effects of accurate monitoring, efficient operation and maintenance and risk visual management of the whole life cycle of the water conservancy facilities are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of water conservancy projects, and specifically to an auxiliary decision-making platform for the operation and maintenance of water conservancy projects based on AI drones. Background Art

[0002] In the daily operation and maintenance of modern water conservancy projects, the long-term safe operation of water conservancy facilities is directly related to the safety of people's lives and property and the sustainable development of society. These facilities are often widely distributed, large in scale, with complex operating environments, and are susceptible to multiple impacts such as natural disasters and structural aging and material fatigue caused by long-term use. In order to timely discover potential risk points and take scientific and effective maintenance measures, it is necessary to use an intelligent platform that can realize multi-dimensional data collection, dynamic risk assessment and auxiliary decision-making;

[0003] The existing technology has developed water conservancy facility operation and maintenance methods based on single sensor monitoring or rule model analysis. For example, displacement sensors, pressure sensors, and temperature monitoring equipment are widely used to collect specific parameters, which can perform basic monitoring of the operating status of facilities. On this basis, some technologies use preset fixed rule models to issue alarms for monitoring data that exceeds the threshold, providing basic safety tips for operation and maintenance personnel. These methods can reduce the difficulty of operation and maintenance to a certain extent under relatively stable working conditions and improve the daily monitoring capabilities of water conservancy facilities.

[0004] However, existing technologies still have many shortcomings when facing complex and changeable operating environments and high-risk working conditions, and these problems are particularly evident in practical applications. On the one hand, the data acquisition capability of a single sensor is limited, and it is difficult to cover the overall operating status of water conservancy facilities. For example, displacement sensors can only capture deformation information but cannot perceive the depth of crack expansion or the degree of material aging, which leads to significant blind spots in the monitoring range. On the other hand, existing technologies generally adopt static analysis models based on threshold rules. This method lacks the ability to adapt to dynamic changes during risk assessment. Taking the expansion of dam cracks as an example, when the crack width exceeds a certain fixed threshold, the system will issue an alarm, but key information such as the crack expansion rate and acceleration are usually ignored. This delayed risk assessment method may cause operation and maintenance personnel to miss the best time for early intervention, which is particularly prone to uncontrollable safety hazards under high water levels or flood discharge conditions; in addition, task priority sorting and resource allocation mainly rely on the experience and judgment of operation and maintenance personnel, and lack scientific optimization based on data, which leads to inefficient resource scheduling in multi-task scenarios; finally, traditional monitoring platforms usually rely on static tables, two-dimensional charts or single-dimensional data display methods, lack of comprehensive processing and intuitive presentation of multi-source data, especially when faced with complex problems that require simultaneous analysis of temporal changes and spatial distribution, operation and maintenance personnel often find it difficult to quickly locate key risk areas or understand the overall operating status of the facility. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention provides an auxiliary decision-making platform for the operation and maintenance of water conservancy projects based on AI drones, which solves the problems in the existing technology that the monitoring range of a single sensor is insufficient, the static analysis model is difficult to adapt to dynamic changes, the task priority and resource scheduling lack scientific optimization, and the multi-source data is difficult to comprehensively and intuitively present.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drones, the platform comprising:

[0007] A data acquisition module, used to acquire multimodal data of water conservancy facilities through AI drones, wherein the multimodal data includes visible light images, thermal imaging data, multispectral data, and lidar data;

[0008] A data processing and fusion module, used for formatting, spatiotemporal alignment and feature fusion of the multimodal data;

[0009] The health assessment and prediction module is used to assess the health status of water conservancy facilities based on the processed data and predict the future trend of facility status changes;

[0010] A decision-making support module for generating inspection priority planning, maintenance recommendations, and emergency response plans for water conservancy facilities;

[0011] Interaction and visualization module, used to display the status of water conservancy facilities, decision suggestions and risk distribution maps.

[0012] Preferably, the data acquisition module includes:

[0013] A UAV mission planning unit, which is used to generate a dynamic path planning for the UAV based on a reinforcement learning algorithm and a geographic information system, wherein the path planning generates an optimal flight route based on the current facility risk level, terrain obstacles, and inspection mission objectives;

[0014] The multimodal data acquisition unit is used to obtain multimodal data of water conservancy facilities through the visible light camera, thermal imager, multispectral sensor and lidar carried by the drone. The multimodal data acquisition unit completes preliminary data processing in real time through the edge computing module and synchronizes it to the cloud.

[0015] Preferably, the data processing and fusion module includes:

[0016] A data preprocessing unit, used for formatting, time synchronization and spatial alignment of multimodal data, wherein the time synchronization unifies the sampling times of different sensors by an interpolation method, and the spatial alignment maps the data of different sensors to a unified spatial range by a geographic coordinate system;

[0017] The feature fusion unit is used to extract features from multimodal data through a deep learning algorithm and to achieve correlation modeling between data of different modalities through a self-attention mechanism. The feature fusion unit embeds multimodal features into a unified representation vector for subsequent health assessment.

[0018] Preferably, the health assessment and prediction module includes:

[0019] A defect detection unit, used to automatically identify cracks, corrosion and leakage areas of water conservancy facilities based on a target detection algorithm, and the detection unit generates defect type labels and location coordinates;

[0020] A health status scoring unit is used to calculate the health status score of the facility through a scoring model based on the defect data generated by the defect detection unit, combined with the facility's historical monitoring data and real-time collected data;

[0021] The state prediction unit is used to predict the future change trend of the health score of water conservancy facilities based on the time series prediction model and generate a prediction report.

[0022] Preferably, the health status scoring unit includes:

[0023] Scoring factor calculation module, used to calculate individual scores based on influencing factors such as crack width, leakage range and dam stress;

[0024] The comprehensive score calculation module is used to generate a comprehensive health score of the facility based on weighted scoring factors. The weight parameters can be dynamically adjusted according to actual operation and maintenance needs.

[0025] Preferably, the auxiliary decision module includes:

[0026] Inspection priority planning unit, used to generate inspection task priorities based on health scores and facility risk levels;

[0027] Emergency response unit, used to generate the optimal emergency inspection task path in real time based on the disaster scenario;

[0028] The maintenance recommendation generation unit is used to generate maintenance plans and resource allocation plans based on the long-term prediction of the health status score.

[0029] Preferably, the inspection priority planning unit includes:

[0030] Task risk assessment module, which is used to calculate the risk level of inspection tasks based on facility health scores;

[0031] The priority calculation module is used to generate an inspection priority list based on the task risk level and the facility health score, and the priority is dynamically adjusted according to the importance of the inspection target.

[0032] Preferably, the interaction and visualization module includes:

[0033] A data visualization unit, used to display the health status and risk distribution map of water conservancy facilities through three-dimensional GIS technology, wherein the risk distribution map is generated according to the health score of the facilities;

[0034] The augmented reality unit is used to dynamically overlay the health scores and inspection recommendations of water conservancy facilities through on-site scanning and present them in real time on the AR device.

[0035] Preferably, the data visualization unit generates a three-dimensional risk distribution map in combination with the multimodal fusion result, wherein the risk value is mapped to a color gradient range according to the health score of the facility and is displayed on the three-dimensional model in the form of a heat map.

[0036] Preferably, the augmented reality unit comprises:

[0037] Status information overlay module, used to dynamically load facility status data in AR glasses or mobile devices;

[0038] The decision-making recommendation display module is used to display inspection recommendations and emergency response plans in real time in an augmented reality scene.

[0039] The present invention provides an auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drones. It has the following beneficial effects:

[0040] 1. The present invention realizes comprehensive monitoring and accurate quantitative evaluation of the operating status of water conservancy facilities by adopting multi-source data fusion and dynamic risk assessment. Compared with the existing technology that only relies on a single sensor or static analysis, it solves the problems of insufficient data, delayed evaluation and incomplete risk identification in complex environments, and provides more efficient technical support for the health management of facilities.

[0041] 2. The present invention combines time series prediction with optimization allocation algorithm to make trend prediction of key risk points of water conservancy facilities more accurate and resource allocation more efficient. Compared with traditional technology that relies on fixed rules, it improves the ability to perceive potential hidden dangers in advance, solves the problems of resource waste and slow task response, and provides a forward-looking decision-making basis for the safe operation of facilities.

[0042] 3. The present invention designs a module that combines real-time interaction with multi-dimensional visualization, presenting information in the form of heat maps, three-dimensional models and dynamic task lists, etc. Users can directly filter, operate and adjust, thereby solving the problem of complex data and difficult information visualization, and providing operation and maintenance personnel with a more friendly operation experience and more efficient work support. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1is a system structure diagram of the present invention;

[0044] Figure 2 It is a system structure diagram of the data acquisition module of the present invention;

[0045] Figure 3 It is a schematic diagram of the system structure of the data processing and fusion module of the present invention;

[0046] Figure 4 It is a schematic diagram of the system structure of the auxiliary decision module of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Please see attached Figure 1-4 The embodiment of the present invention provides an auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drones, the platform comprising:

[0049] The data acquisition module is used to obtain multimodal data of water conservancy facilities through AI drones. The multimodal data includes visible light images, thermal imaging data, multispectral data, and lidar data;

[0050] This module is responsible for obtaining real operating status data from the site, including the external and internal conditions of water conservancy project facilities, as well as real-time changes in the environment, to provide a basic guarantee for subsequent data processing, analysis, and decision support. Its operating logic is connected with other modules of the platform, forming a close linkage relationship with the path planning, data analysis, and decision support modules. Specifically, the data acquisition module is associated with the flight path of the drone in real time during the inspection process, and is synchronously transmitted to the ground data center, achieving multi-faceted information coverage through multi-dimensional data streams collected by multiple sensors;

[0051] First, the high-definition camera carried by the drone is used to collect images of the surface of water conservancy facilities. Generally, the camera resolution can reach 4K, which meets the needs of capturing detailed features such as cracks and surface peeling of water conservancy facilities. As an option, the camera can also be equipped with a lens with a zoom function for close-up shots of key areas. For example, when an abnormality is found on the surface of the dam during the inspection, the camera can switch to a high-power zoom mode to obtain clearer detailed images.

[0052] Specifically, the images acquired by the camera are compressed in real time during the inspection process to reduce data transmission delays while ensuring that the basic quality of the image is not affected. In some embodiments, for scenes with insufficient light, such as nighttime inspections or underwater inspections, the camera can be used in conjunction with infrared imaging equipment to compensate for the lack of image acquisition in low-light environments.

[0053] Secondly, LiDAR is mainly used to generate three-dimensional spatial models of water conservancy engineering facilities. In this embodiment, LiDAR can accurately capture the geometric shape and subtle deformation of structures such as dams and levees through the collection and analysis of point cloud data. Generally, the scanning frequency of LiDAR can reach 300,000 points per second, and the measurement accuracy can be accurate to the millimeter level, which is capable of fine measurement of complex terrain.

[0054] In one possible implementation, the point cloud data collected by LiDAR is filtered out of noise and corrected geometrically by the preliminary calculation module built into the sensor to ensure the accuracy and consistency of the point cloud data. As an option, the LiDAR can also adjust the scanning range and angle according to specific mission requirements. For example, for wide dam areas, a 360° panoramic scanning mode can be enabled; and for deformation monitoring of specific parts, a local high-precision mode can be used.

[0055] In some embodiments, in order to further improve the acquisition accuracy, the LiDAR module is also linked with the flight control system of the drone. Specifically, the drone adjusts the flight altitude and angle in real time to ensure that the LiDAR always maintains the optimal measurement distance and fully matches the geometric contours of the water conservancy facilities. After the point cloud data is collected, it will be compared with the historical model to detect possible deformation trends.

[0056] The multispectral sensor is an important component of the data acquisition module, and is mainly used to obtain the surface material status and temperature distribution of water conservancy facilities. In this embodiment, the multispectral sensor can perform synchronous imaging in multiple bands such as visible light, infrared light, and ultraviolet light, providing comprehensive data support for material aging analysis and temperature anomaly detection.

[0057] In one possible implementation, the multispectral sensor detects material cracks and water seepage in the dam or dike through the short-wave infrared band (SWIR); at the same time, the long-wave infrared band (LWIR) is used to capture the non-uniform distribution of the dam temperature. For example, in a high-temperature environment, an abnormal increase in the local temperature on the dam surface may indicate a leakage problem or internal defects.

[0058] In general, the data collection frequency of the multispectral sensor can be dynamically adjusted according to the inspection task. For example, when a suspected abnormal area is detected, the sampling frequency can be increased to capture more detailed spectral information. In some embodiments, multispectral data is collected synchronously with LiDAR point cloud data, and the deep combination of geometric and spectral information is achieved through the synergy of sensors.

[0059] In addition, the data fusion unit in the data acquisition module integrates and processes the data from multiple sensors to improve the accuracy and availability of the collected data. Generally, the data collected by multiple sensors are inconsistent in space and time, and direct use may lead to error accumulation. In this embodiment, the data fusion unit performs real-time correction and fusion of multi-dimensional data based on the Kalman filter algorithm, eliminates the influence of noise, and generates a high-precision integrated data stream.

[0060] Specifically, the state update formula of the Kalman filter algorithm is as follows:

[0061] X k =AX k-1 +BU k +w k

[0062] Among them, X k represents the system state at the current moment, A is the state transfer matrix, B is the control input matrix, U k is the sensor input data, w k is the process noise;

[0063] The observation update formula is:

[0064] Z k =HX k +v k

[0065] Among them, Z k is the observation vector, H is the observation matrix, v k is the observed noise, the filter gain X k The calculation formula is:

[0066] K k =P k|k-1 H T (HP k|k-1 H T +R) -1

[0067] Among them, K k is the filter gain, P k|k-1 is the prediction error covariance matrix, H is the observation matrix, R is the observation noise covariance matrix, (HP k|k-1 H T +R) -1is the weighted sum inverse matrix of the covariance matrix, which is used to quantify the relative degree of confidence between the observed values ​​and the predicted values;

[0068] Through the above formula, the data fusion unit can dynamically weight the output data of the camera, LiDAR, and multispectral sensor to ensure that the generated comprehensive data is highly consistent in time and space. For example, when the camera detects a crack area, the data fusion unit will combine the LiDAR point cloud data to further confirm the depth and shape information of the crack, thereby providing a more comprehensive basis for judgment;

[0069] In some embodiments, the filter parameters Q (process noise covariance) and R (observation noise covariance) are dynamically adjusted according to the actual working conditions. For example, in a complex environment with high wind speed, increasing the value of Q can improve the filter's trust in the observed data.

[0070] Data processing and fusion module, used for formatting, spatiotemporal alignment and feature fusion of multimodal data;

[0071] This module works closely with the data acquisition module to integrate data from different devices such as cameras, LiDAR, and multispectral sensors, eliminate noise and redundant information, and generate a data stream with temporal and spatial consistency, providing a solid foundation for the AI ​​analysis module.

[0072] In this embodiment, the data processing and fusion module mainly includes a preprocessing unit, a time-space alignment unit and a fusion algorithm unit.

[0073] First, the data preprocessing unit is used to perform preliminary processing on the original collected data. Generally, the data collected by different sensors have problems such as noise, error and inconsistent timestamps, and direct use will reduce the analysis accuracy. This embodiment adopts a multi-step preprocessing method, including noise filtering, data format conversion and outlier removal.

[0074] Specifically, the image data acquired by the camera will use an edge-preserving filtering algorithm to remove environmental noise while retaining important detail features such as cracks. As an option, the point cloud data collected by LiDAR uses a statistical outlier detection method to remove isolated point cloud noise. For example, the average distance from a point cloud point to its neighboring points is set as a threshold. When the distance exceeds the threshold, it is marked as an abnormal point and removed.

[0075] In some embodiments, the spectral data collected by the multi-spectral sensor needs to be corrected for light intensity to eliminate the influence of ambient light changes. The spectral correction formula is:

[0076]

[0077] Among them, I correctedrepresents the corrected spectral intensity, I measured is the measured spectral intensity, I reference is the spectral intensity of the reference light source;

[0078] The time-space alignment unit is an important part of the data processing and fusion module. Its main function is to synchronize the time and space of the data from different sensors. Generally, there are differences in the sampling frequency and timestamp of different sensors, and the installation position of the sensor also leads to different coordinate systems for data collection. This embodiment achieves consistency in time and space through a time interpolation algorithm and a coordinate transformation matrix.

[0079] Specifically, time synchronization uses a linear interpolation method to interpolate the corresponding time point data of other sensors based on the timestamp of the sensor data with a higher sampling frequency. The spatial matching principle is completed through the coordinate transformation matrix. Assuming that the local coordinate system of the sensor is L and the global coordinate system of the drone is G, the spatial transformation formula of the data is:

[0080] P G =R·P L +T

[0081] Among them, P G is the global coordinate point, P L is the local coordinate point, R is the rotation matrix, and T is the translation vector. In a possible implementation, the values ​​of R and T are determined by a calibration experiment of the sensor installation position;

[0082] The fusion algorithm unit is responsible for deep fusion of multi-source data after time and space alignment. Generally, this embodiment uses the Kalman filter algorithm to dynamically weighted fuse multi-dimensional data to further improve data accuracy;

[0083] As an option, this embodiment may also use a deep learning algorithm for feature-level fusion. For example, the spatial features of the camera data are extracted by a convolutional neural network (CNN), and fused with the geometric features of the LiDAR point cloud data in a fully connected layer to generate higher-dimensional comprehensive features.

[0084] In one possible implementation, the output of the data processing and fusion module includes the registered and fused 3D terrain model, surface crack feature point set, and multi-spectral light intensity distribution map. These data can be directly transferred to the AI ​​analysis module for risk assessment and decision support;

[0085] The health assessment and prediction module is used to assess the health status of water conservancy facilities based on the processed data and predict the future trend of facility status changes;

[0086] This module is closely connected with the aforementioned acquisition module and processing module, and provides comprehensive and reliable evaluation basis and early warning information for subsequent operation and maintenance decisions through in-depth analysis of multi-source data. This module is particularly suitable for monitoring and trend analysis of problems such as crack expansion, deformation increase, and material aging generated during the long-term operation of water conservancy facilities;

[0087] In this embodiment, the health assessment and prediction module includes a real-time assessment unit, a historical data association unit and a prediction analysis unit. The specific functions and technical implementations are as follows:

[0088] First, the real-time assessment unit is responsible for instantly assessing the structural health status of the water conservancy facility based on the currently collected data. Generally, the assessment unit quantitatively describes the current status of the water conservancy facility based on a variety of indicators, including but not limited to crack width, depth, growth rate, degree of dam deformation and leakage area. This embodiment adopts a regularized scoring mechanism to quantify the health status into risk levels. For example, if the crack width W exceeds a certain threshold W, the risk level will be quantified. thresh When the risk level is high, the assessment unit marks the current risk level as "high risk". The risk level calculation formula is as follows:

[0089]

[0090] Where R is the risk level (dimensionless value); W, D and A represent the crack width, depth and leakage area respectively; W thresh , D thresh and A thresh : corresponding threshold; α, β and γ: weight parameters, set according to the specific conditions of the facility;

[0091] As an option, the real-time assessment unit can also analyze the temperature distribution of the leakage point in combination with the thermal imaging data. For example, areas with large temperature differences may correspond to higher leakage risks. Specifically, based on the temperature distribution map generated by the multi-spectral sensor, the assessment unit locates high-risk areas through the thermal map clustering algorithm to further verify the severity of the leakage problem.

[0092] The historical data association unit is used to compare and analyze the current state with the historical operation data. Generally speaking, the health status of water conservancy facilities has a certain time dependence. For example, the expansion rate of cracks is usually affected by the external environment (such as water flow pressure and temperature changes). Therefore, in this embodiment, the historical data association unit compares the real-time evaluation results with the data in the historical database to identify the characteristic points of abnormal changes.

[0093] Specifically, the correlation unit uses a time series regression model to fit the changing trend of key indicators in historical data. For example, the expansion trend of crack width W over time t can be expressed as:

[0094] W(t)=t0+k·t+∈

[0095] Where, W(t): crack width; t: time variable representing crack monitoring, which is the independent variable of the formula; W0: initial crack width; k: crack extension rate; ∈: random noise term;

[0096] By fitting the above model, the association unit can calculate the degree to which the current data deviates from the historical trend. For example, if the crack width at the current moment is significantly higher than the fitting value W(t), it can be preliminarily determined that the crack is at risk of accelerating expansion. In some embodiments, the association unit also performs causal analysis on abnormal changes in combination with the operating conditions of the water conservancy facilities (such as pressure fluctuations during flood discharge) to provide richer contextual information for subsequent prediction modules.

[0097] The prediction and analysis unit is the core of the health assessment and prediction module. Its main function is to infer the future health status of water conservancy facilities based on historical data and real-time data, using time series prediction and machine learning models. In general, this embodiment uses a long short-term memory network (LSTM) model for prediction and analysis, which is particularly suitable for long-term trend prediction under complex working conditions.

[0098] Specifically, the prediction and analysis unit inputs multidimensional data in the form of time series, including characteristics such as crack width, dam deformation, leakage area, and environmental variables (such as water level and temperature). The model structure is as follows:

[0099] Input layer: accepts multidimensional time series features;

[0100] Hidden layer: consists of several LSTM units to capture short-term and long-term dependencies of data;

[0101] Output layer: Generates predicted values ​​for several time steps in the future.

[0102] The training objective of the model is to minimize the mean square error (MSE) between the predicted value and the actual value, and its loss function is defined as:

[0103]

[0104] in, Mean square error; Represents the prediction error of the model at the i-th sample point, that is, the gap between the predicted value and the true value; The squared error of the i-th sample; The predicted value of the model; Y i : actual value; n: sample size; Error accumulation sign.

[0105] In one possible implementation, the prediction and analysis unit combines dynamic working condition information to adjust the model weights in different scenarios. For example, when it is detected that the water conservancy facility is in a flood discharge state, the model will give priority to variables related to water flow pressure and dynamically reduce the weights of other features to improve prediction accuracy;

[0106] A decision-making support module for generating inspection priority planning, maintenance recommendations, and emergency response plans for water conservancy facilities;

[0107] This module is closely connected with the assessment module. Through in-depth analysis of risk levels, forecast trends and operating condition data, it provides real-time operation guidance and decision support for operation and maintenance personnel to ensure the safe and stable operation of water conservancy facilities.

[0108] In this embodiment, the auxiliary decision module mainly includes a priority calculation unit, a decision optimization unit and a task generation unit. The specific functions and implementation methods of each unit are as follows:

[0109] First, the priority calculation unit ranks all current risk events according to the risk level output by the health assessment module and the importance index of the facility. In general, the priority calculation formula is:

[0110] P i =ω1·R i +ω2·I i

[0111] Among them, P i : The priority of event i; R i : Risk level of the event, provided by the assessment module; I i : Facility importance index, which is set based on the functional level of the facility; ω1, ω2: weight parameters used to balance the impact of risk level and facility importance. As an option, this unit supports dynamic adjustment of weight parameters. For example, when the system detects heavy rainfall or flood discharge conditions, the risk level weight ω1 can be appropriately increased to give priority to events related to water flow pressure.

[0112] In a possible implementation, the priority calculation unit can also be dynamically updated in combination with the time dimension. For example, for risk events with a high crack growth rate, the system will gradually increase its priority over time to ensure that the risk will not be aggravated due to delayed processing.

[0113] The decision optimization unit is responsible for generating optimal decisions for maintenance and operation based on the priority results and the current status of the facility. Generally, this unit adopts a multi-objective optimization method, taking into account factors such as resource consumption, risk reduction, and task execution time. The optimization objective function can be defined as:

[0114] Minimize: J=λ1·C+λ2·T+λ3·R residual

[0115] Where J: optimization target value; C: total cost of maintenance task; T: total time of task execution; R residual : residual risk level; λ1, λ2, λ3: weight parameters used to balance the optimization needs of cost, time and risk;

[0116] Specifically, the optimization algorithm uses the spatial distribution of facilities and maintenance resources as constraints, and solves the optimal task allocation plan through dynamic programming or genetic algorithms. For example, in the task of repairing multiple cracks in the dam body, the optimization algorithm will generate the optimal construction path and process arrangement based on the geographical location of the cracks, the expansion rate, and the availability of the construction team;

[0117] In some embodiments, the decision optimization unit can also update the optimization results in real time. For example, when the system detects a new risk event, the current task plan can be dynamically adjusted to give priority to handling sudden problems in high-risk areas;

[0118] The task generation unit is the output interface of the auxiliary decision module. Its main function is to convert the optimization decision into a specific execution task and present it to the operation and maintenance personnel in an intuitive way. In this embodiment, the task generation unit supports the generation of multiple forms of output content, including task lists, work order templates, and visual risk maps.

[0119] As an option, the unit can also automatically generate standardized work order templates for direct use by construction personnel. For example, the work order will list the GPS coordinates of the construction location, specific operation steps and precautions in detail.

[0120] In one possible implementation, the task generation unit also supports visual output. For example, a heat map can be used to visually display high-risk areas of a facility, helping operation and maintenance personnel to quickly locate key problem points. The generation of the heat map is based on the spatial data of the facility and the distribution of risk levels. The darker the color, the higher the risk.

[0121] In some embodiments, the task generation unit can also be integrated with an external scheduling system to achieve automatic task allocation and progress tracking. For example, when a task is marked as "urgent", the system will automatically send a notification to the relevant construction team and monitor the task execution status in real time to ensure timely completion;

[0122] Interaction and visualization module, used to display the status of water conservancy facilities, decision suggestions and risk distribution maps;

[0123] This module presents the task information, risk assessment results and forecast analysis content generated by the above-mentioned decision-making support module to users in an intuitive and easy-to-understand form through a graphical and interactive interface. It is closely connected with the decision-making support module to provide operation and maintenance personnel with real-time data display, task progress tracking and operation feedback channels, thus improving the practicality and operation efficiency of the entire system.

[0124] In this embodiment, the interaction and visualization module mainly includes a data display unit, an interaction control unit, and a dynamic feedback unit. The specific functions and implementation methods of each unit are as follows:

[0125] First, the data display unit is responsible for presenting the analysis results and task information to the user in a visual form. Generally, this unit uses a variety of graphical methods to classify and display data, including line charts, heat maps, bar charts, and three-dimensional models. Specifically:

[0126] The risk assessment results are displayed in the form of a heat map, which intuitively reflects the risk level of water conservancy facilities through the depth of color. For example, the red area on the heat map indicates a high-risk location, yellow indicates a medium risk, and green indicates a low risk. The color mapping relationship is defined based on the numerical range of the risk level R:

[0127]

[0128] Among them, Color is the color result after mapping; R is the risk level; R high and R low are the high and low thresholds of risk level respectively; Map(R) is the mapping function from risk level to color;

[0129] For dynamic changing characteristics such as crack width and deformation rate, the system uses a line graph to display their changing trends over time. For example, the expansion of a crack in a dam body can be plotted as a line graph using the following time series data to track its changes in real time.

[0130] The 3D point cloud data is displayed through 3D model dynamic rendering technology to reflect the overall shape and local deformation of the dam or other facilities. As an option, users can rotate or scale the 3D model by dragging the interface to observe the specific details more clearly.

[0131] In some embodiments, the data display unit also supports the overlay display of multiple layers of information. For example, in a three-dimensional model, the user can overlay heat map data and use colors to mark the location and range of high-risk areas, thereby realizing the combination of space and risk information.

[0132] The interactive control unit is responsible for providing users with data query and operation functions. Generally, this unit designs a variety of interactive tools in the form of buttons, sliders, input boxes, etc., allowing users to adjust the content and scope of data display according to their needs.

[0133] Specifically, this unit supports the following interactive features:

[0134] Users can query data for specific tasks or areas by filtering conditions (such as risk level, facility type). For example, users can choose to display only data for facilities with a risk level higher than R threshold task.

[0135] Users can adjust the time range through the slider to dynamically update the data of the line graph or bar graph. For example, when the slider moves from the current time to the past, the line graph will show the historical trend of crack expansion.

[0136] In the 3D model, users can select an area of ​​interest using the mouse or touchpad, and the system will automatically pop up detailed information about the area, such as crack location, expansion rate, current width, etc.

[0137] As an option, this unit also supports custom task grouping. For example, users can group high priority tasks and set specific color labels for quick identification.

[0138] The dynamic feedback unit is used to respond to user input and system status changes in real time to ensure data accuracy and timeliness of interaction. In this embodiment, the dynamic feedback unit completes the following functions through real-time data communication with the system core module:

[0139] When the user selects a task or facility on the interface, the system will update the interface display in real time. For example, when the user clicks on a high-risk area, the system pops up detailed risk analysis results and maintenance recommendations.

[0140] The dynamic feedback unit can track the progress of task execution in real time and make visual updates on the interface. For example, for crack repair tasks, the system will update the task-completed area in the heat map in real time and change its color from red to green.

[0141] In some embodiments, the dynamic feedback unit can also send reminders to the user through sound or pop-up windows. For example, when the system detects that the risk level of a facility has changed significantly, the system will automatically issue an alarm and highlight the relevant area on the interface.

[0142] In one possible implementation, the interaction and visualization module also supports multi-platform synchronous display functions; for example, users can view key data through the mobile application and keep it synchronized with the desktop system in real time; the mobile interface is simple in design and suitable for quickly browsing task lists and viewing risk heat maps; while the desktop interface supports more complex operations and more detailed data display.

[0143] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drones, characterized in that: The platform includes: A data acquisition module, used to acquire multimodal data of water conservancy facilities through AI drones, wherein the multimodal data includes visible light images, thermal imaging data, multispectral data, and lidar data; A data processing and fusion module, used for formatting, spatiotemporal alignment and feature fusion of the multimodal data; The health assessment and prediction module is used to assess the health status of water conservancy facilities based on the processed data and predict the future trend of facility status changes; A decision-making support module for generating inspection priority planning, maintenance recommendations, and emergency response plans for water conservancy facilities; Interaction and visualization module, used to display the status of water conservancy facilities, decision suggestions and risk distribution maps.

2. According to claim 1, an auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drones is characterized in that: The data acquisition module comprises: A UAV mission planning unit, which is used to generate a dynamic path planning for the UAV based on a reinforcement learning algorithm and a geographic information system, wherein the path planning generates an optimal flight route based on the current facility risk level, terrain obstacles, and inspection mission objectives; The multimodal data acquisition unit is used to obtain multimodal data of water conservancy facilities through the visible light camera, thermal imager, multispectral sensor and lidar carried by the drone. The multimodal data acquisition unit completes preliminary data processing in real time through the edge computing module and synchronizes it to the cloud.

3. According to claim 1, the auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone is characterized in that: The data processing and fusion module includes: A data preprocessing unit, used for formatting, time synchronization and spatial alignment of multimodal data, wherein the time synchronization unifies the sampling times of different sensors by an interpolation method, and the spatial alignment maps the data of different sensors to a unified spatial range by a geographic coordinate system; The feature fusion unit is used to extract features from multimodal data through a deep learning algorithm and to achieve correlation modeling between data of different modalities through a self-attention mechanism. The feature fusion unit embeds multimodal features into a unified representation vector for subsequent health assessment.

4. According to claim 1, the auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone is characterized in that: The health assessment and prediction module includes: A defect detection unit, used to automatically identify cracks, corrosion and leakage areas of water conservancy facilities based on a target detection algorithm, and the detection unit generates defect type labels and location coordinates; A health status scoring unit is used to calculate the health status score of the facility through a scoring model based on the defect data generated by the defect detection unit, combined with the facility's historical monitoring data and real-time collected data; The state prediction unit is used to predict the future change trend of the health score of water conservancy facilities based on the time series prediction model and generate a prediction report.

5. According to the AI ​​drone-based decision-making support platform for water conservancy project operation and maintenance as described in claim 1, it is characterized in that: The health status scoring unit comprises: Scoring factor calculation module, used to calculate individual scores based on influencing factors such as crack width, leakage range and dam stress; The comprehensive score calculation module is used to generate a comprehensive health score of the facility based on weighted scoring factors. The weight parameters can be dynamically adjusted according to actual operation and maintenance needs.

6. According to claim 1, the auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone is characterized in that: The auxiliary decision module comprises: Inspection priority planning unit, used to generate inspection task priorities based on health scores and facility risk levels; Emergency response unit, used to generate the optimal emergency inspection task path in real time based on the disaster scenario; The maintenance recommendation generation unit is used to generate maintenance plans and resource allocation plans based on the long-term prediction of the health status score.

7. According to claim 1, the auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone is characterized in that: The inspection priority planning unit includes: Task risk assessment module, which is used to calculate the risk level of inspection tasks based on facility health scores; The priority calculation module is used to generate an inspection priority list based on the task risk level and the facility health score, and the priority is dynamically adjusted according to the importance of the inspection target.

8. According to claim 1, the auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone is characterized in that: The interaction and visualization module includes: A data visualization unit, used to display the health status and risk distribution map of water conservancy facilities through three-dimensional GIS technology, wherein the risk distribution map is generated according to the health score of the facilities; The augmented reality unit is used to dynamically overlay the health scores and inspection recommendations of water conservancy facilities through on-site scanning and present them in real time on the AR device.

9. The auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone according to claim 8 is characterized in that: The data visualization unit generates a three-dimensional risk distribution map in combination with the multimodal fusion results, in which the risk value is mapped to a color gradient range according to the health score of the facility and displayed on the three-dimensional model in the form of a heat map.

10. The auxiliary decision-making platform for water conservancy project operation and maintenance based on AI drone according to claim 8 is characterized in that: The augmented reality unit comprises: Status information overlay module, used to dynamically load facility status data in AR glasses or mobile devices; The decision-making recommendation display module is used to display inspection recommendations and emergency response plans in real time in an augmented reality scene.

Images