An underwater path planning method and system based on data fusion

Abstract

This invention relates to an underwater path planning method based on data fusion, belonging to the field of intelligent inspection technology for special equipment. The method uses a multi-source heterogeneous sensor array to collect real-time hull status data and construct a comprehensive environmental dataset. The dataset is then differentiated, divided into subsets based on material, deposit density, and water flow disturbance characteristics, and compensation factors are generated. These subsets and compensation factors are input into a graph structure learning network path planning model, outputting cleaning paths and equipment action commands. A multi-dimensional attribution analysis engine is used to locate the main causes of cleaning efficiency deviations and generate compensation strategies. After the strategies are converted into control commands and executed, the collected effect data is fed back to the differentiation processing stage to correct the subset division and compensation factors, forming a continuous evolutionary closed loop. This application improves the targeting and operational efficiency of path planning, reduces energy consumption and repetitive operation rates, promotes the transformation of underwater inspection towards intelligence and adaptability, and thus improves the accuracy of underwater inspection of special equipment.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent detection of special equipment, and in particular to an underwater path planning method and system based on data fusion. Background Technology

[0002] In the field of underwater special equipment inspection, existing technologies mainly rely on limited environmental information acquired by a single sensor (such as an optical camera or side-scan sonar), combined with manually preset fixed path patterns for operation. Traditional methods typically employ simple algorithms or manually planned straight-line scanning paths, controlling the underwater robot's movement through preset waypoint sequences. This lacks the comprehensive perception capability for multi-dimensional operational conditions such as underwater visibility, water flow disturbances, and differences in structural materials. A single data source cannot fully perceive the complex underwater environment, making it difficult to effectively integrate the underwater environment and actual conditions for real-time response. This results in numerous blind spots in structurally complex areas, reducing the accuracy of underwater inspection of special equipment. Therefore, improvements are needed. Summary of the Invention

[0003] To improve the accuracy of underwater inspection of special equipment, this application provides an underwater path planning method and system based on data fusion.

[0004] Firstly, the above-mentioned inventive objective of this application is achieved through the following technical solution: An underwater path planning method based on data fusion, the method comprising the following steps: Based on the real-time acquisition of underwater hull status data using a multi-source heterogeneous sensor array, the hull status data is spatiotemporally aligned and standardized and fused to construct a comprehensive environmental dataset. The comprehensive environmental dataset is processed differentially, and the dataset is divided into multiple job subsets according to the preset feature classification rules. For each job subset, an appropriate cleaning strategy compensation factor is generated independently. The operation subset and the cleaning strategy compensation factor are input into a pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, outputting a cleaning path trajectory sequence and a cleaning equipment action command sequence. The cleaning path trajectory sequence is input into the multidimensional attribution analysis engine, which uses a machine self-learning algorithm to locate the main cause category of the cleaning efficiency deviation and generates equipment travel speed compensation strategy and cleaning intensity compensation strategy. The compensation strategy is converted into control commands for the underwater robot's thrusters and the cleaning device, which are then issued and executed, and an execution log is recorded after execution. The cleaning effect improvement rate is calculated by re-acquiring images of the ship's surface and feeding the improvement effect evaluation results back to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0005] By adopting the above technical solution, the graph structure learning network and compensation factor mechanism effectively address the problems of path deviation and detection blind spots that easily occur in environments with low visibility and complex flow fields, significantly reducing the rate of repetitive work and energy consumption. At the same time, the closed-loop feedback enables the model to self-evolve, greatly reducing the need for manual intervention. This provides a feasible and replicable intelligent solution for the underwater special equipment inspection industry, improving the safety and economy of deep-water operations. The closed-loop feedback enables the model to self-evolve, giving special equipment an adaptive function when working underwater. Compared with the single working mode of existing technologies, this application effectively improves the accuracy of underwater inspection of special equipment.

[0006] In a preferred example, this application can be further configured as follows: the step of differentially processing the comprehensive environmental dataset, dividing the dataset into multiple job subsets according to preset feature classification rules, and independently generating an appropriate cleaning strategy compensation factor for each job subset includes the following steps: Based on the corrosion resistance characteristics of the ship hull material, the comprehensive environmental dataset is divided into a steel structure subset and a non-metallic subset. A rust enhancement detection factor is generated for the steel structure subset, and a biofilm texture enhancement factor is generated for the non-metallic subset. Based on the density distribution characteristics of the attachments, the dataset is divided into a high-density attachment area subset and a low-density area subset. A spiral full-coverage path compensation factor is generated for the high-density area, and a linear scanning path compensation factor is generated for the low-density area. Based on the characteristics of water flow disturbance intensity, the dataset is divided into a strong flow disturbance subset and a still water subset. A countercurrent compensation factor is generated for the strong flow disturbance subset to correct the thruster power distribution.

[0007] In a preferred example, this application can be further configured to add a physical constraint embedding mechanism to the path planning model, including the following steps: The hull structure strength parameters are transformed into load-bearing capacity feature vectors, which are used as prior attribute inputs to the nodes of the graph structure learning network to constrain the cleaning path trajectory to not exceed the safe operating area of ​​the hull. The kinematic characteristics of the underwater robotic arm are converted into edge weight attenuation factors, so that the turning cost is automatically increased in the joint limit area of ​​the path trajectory. The characteristics of the water flow vector field are used as dynamic correction coefficients to compensate for the drift of the cleaning equipment's travel speed, thus eliminating trajectory deviation caused by tidal currents.

[0008] In a preferred example, this application can be further configured to include the following steps in the process of inputting the clean path trajectory sequence into the multidimensional attribution analysis engine: Construct a health status degradation characteristic curve for equipment, and use the time-series change rate of pusher wear and cleaning brush fatigue as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging. Extract the deviation characteristics of the attachment removal efficiency, calculate the deviation ratio between the actual cleaning coverage and the standard coverage, and trigger the cleaning intensity compensation strategy to be executed first when the deviation exceeds the preset threshold. Identify the bottleneck features of the hull structure, calculate the spatial matching degree between the reach of the cleaning arm and the complexity of the hull surface, and when the matching degree is lower than the preset threshold, increase the weight of the main cause of the structural cleaning blind spot and generate hull segment cleaning suggestions.

[0009] In a preferred example, this application can be further configured to include the following steps after the step of converting the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and recording the execution log after execution: The residual between the cleaning effect improvement rate after execution and the expected target value is calculated to construct an effect deviation evaluation vector; Based on the deviation evaluation vector, attribution and source analysis is performed to identify the key influencing factors that lead to the deviation as improper setting of cleaning strategy compensation factors or outdated operation subset division standards. Incremental learning algorithms are used to dynamically adjust the subset partitioning boundary conditions in the differentiation process and to automatically correct the weight coefficients in the compensation factor generation logic. The revised subset partitioning criteria and compensation factors are fed back to the path planning model to achieve adaptive optimization of the next generation of clean paths and control parameters, forming a continuous evolutionary closed loop.

[0010] In a preferred embodiment, this application may be further configured to include the following steps: synchronously acquiring high-resolution images of the hull surface, side-scan sonar echo signals, and Doppler current profile data through underwater edge computing nodes, and constructing a multimodal state vector of the hull; By utilizing a preset synchronization triggering mechanism, the multimodal state vector is aligned with the common time axis at the millisecond level, eliminating the sampling delay difference of underwater sensors; The aligned data stream is then subjected to sliding window filtering and abnormal pulse removal to eliminate underwater acoustic multipath interference and equipment mechanical vibration noise.

[0011] Secondly, the above-mentioned inventive objective of this application is achieved through the following technical solutions: An underwater path planning device based on data fusion, the device comprising: The comprehensive environmental dataset construction unit is used to collect underwater hull status data in real time based on a multi-source heterogeneous sensor array, and to perform spatiotemporal alignment and standardization fusion of the hull status data to construct a comprehensive environmental dataset. The cleaning strategy compensation factor generation unit is used to differentiate the comprehensive environmental dataset. According to the preset feature classification rules, the dataset is divided into multiple job subsets, and an appropriate cleaning strategy compensation factor is generated independently for each job subset. A cross-modal path optimization unit is used to input the operation subset and cleaning strategy compensation factors into a pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factors for cross-modal path optimization, outputting a cleaning path trajectory sequence and a cleaning equipment action command sequence. The compensation strategy generation unit is used to input the cleaning path trajectory sequence into the multidimensional attribution analysis engine. The multidimensional attribution analysis engine locates the main cause category of the cleaning efficiency deviation based on the machine self-learning algorithm and generates the equipment travel speed compensation strategy and the cleaning intensity compensation strategy. The control command execution unit is used to convert the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and to record the execution log after execution. The optimized feedback correction unit is used to re-acquire images of the ship's surface to calculate the improvement rate of cleaning effect, and feeds back the improvement effect evaluation results to the differentiation processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0012] Thirdly, the above-mentioned objectives of this application are achieved through the following technical solutions: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described underwater path planning method based on data fusion.

[0013] Fourthly, the above-mentioned objectives of this application are achieved through the following technical solutions: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described underwater path planning method based on data fusion. Attached Figure Description

[0014] Figure 1 This is a flowchart of an underwater path planning method based on data fusion in one embodiment of this application; Figure 2 This is a schematic diagram of an underwater path planning device based on data fusion in one embodiment of this application; Figure 3This is a schematic diagram of an electronic device according to an embodiment of this application.

[0015] Icon labels: 1. Comprehensive environmental dataset construction unit; 2. Cleaning strategy compensation factor generation unit; 3. Cross-modal path optimization unit; 4. Compensation strategy generation unit; 5. Control instruction execution unit; 6. Optimization feedback correction unit. Detailed Implementation

[0016] The present application will be further described in detail below with reference to the accompanying drawings.

[0017] In one embodiment, taking the inspection of the underwater structure of a hydroelectric power station dam by a deep-water operation robot as an example, such as... Figure 1 As shown, this application discloses an underwater path planning method based on data fusion, which specifically includes the following steps: S10: Real-time acquisition of underwater hull status data based on a multi-source heterogeneous sensor array, spatiotemporal alignment and standardization fusion of the hull status data to construct a comprehensive environmental dataset; The hull status data includes hull surface image data, hull attachment type characteristics, underwater current field distribution data, and underwater visibility data. When a deep-water robot inspects the underwater structure of a hydroelectric dam, its onboard sensor array collects four types of core data in real time: high-resolution images of the dam's concrete surface (identifying cracks and erosion in localized areas), structural defect type characteristic data (determining the crack width to be moderate through image texture analysis), underwater velocity field distribution data (Doppler current meter detected complex turbulence near the dam's spillway, with an asymmetrical velocity distribution), and underwater visibility data (optical sensors indicated low visibility in the work area due to sediment). Using the underwater positioning system's UTC timestamp as a benchmark, the system performs millisecond-level spatiotemporal alignment of the four heterogeneous data types, eliminating inherent sampling delay differences from sonar, optical, and current sensors. After unifying the data format and coordinate system, a comprehensive environmental dataset is constructed, providing a three-dimensional and highly reliable input foundation for subsequent path planning.

[0018] Furthermore, high-resolution images of the hull surface, side-scan sonar echo signals, and Doppler current profile data are simultaneously acquired through underwater edge computing nodes to construct a multimodal state vector of the hull. By utilizing a preset synchronization triggering mechanism, the multimodal state vector is aligned with the common time axis at the millisecond level, eliminating the sampling delay difference of underwater sensors; The aligned data stream is then subjected to sliding window filtering and abnormal pulse removal to eliminate underwater acoustic multipath interference and equipment mechanical vibration noise.

[0019] S20: Perform differential processing on the comprehensive environmental dataset, divide the dataset into multiple job subsets according to the preset feature classification rules, and generate an appropriate cleaning strategy compensation factor for each job subset independently. Specifically, the system performs differentiated processing on the comprehensive environmental dataset: based on the material characteristics of underwater structures, the dam is divided into two subsets: a "concrete main structure area" and a "metal gate accessory area." A crack detection enhancement compensation factor is generated for the concrete area to highlight minute cracks, and a corrosion detection enhancement compensation factor is generated for the metal area to strengthen corrosion texture recognition. Based on defect density characteristics, the dam surface is divided into a "high-density defect area" and a "sparse intact area." A spiral fine scanning path compensation factor is generated for the high-density area to ensure no missed detections, and a fast linear inspection path compensation factor is generated for the sparse area to improve operational efficiency. Based on the water flow disturbance intensity characteristics, the work area is divided into a "strong flow area" and a "still water area." A dynamic hovering compensation factor is generated for the strong flow area to correct the robot's hovering stability. This differentiated processing avoids the image blurring or path deviation problems caused by traditional single detection modes in complex flow fields, significantly improving the targeting of path planning and operational safety.

[0020] S30: Input the operation subset and cleaning strategy compensation factor into the pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, and outputs a cleaning path trajectory sequence and a cleaning equipment action command sequence. Specifically, the differentiated subset and compensation factors are input into the pre-trained path planning model. The model uses a graph structure learning network to model the dam structure as topological graph nodes (each detection profile and feature point is a node), and the edge weights represent the movement cost of the underwater robot. The network integrates compensation factors to enhance cross-modal features: in the "high-density cracks + strong current" subset, the repeated visit weights of path nodes are automatically enhanced, outputting a spiral progressive detection path trajectory; in the "sparse and intact + still water" subset, a fast linear inspection path is output. Simultaneously, a sequence of underwater robot action commands is generated, including thruster distribution, robotic arm posture angles, and lighting brightness adjustment. Compared to traditional preset paths, this optimized path improves detection coverage and significantly reduces the repetitive path rate, effectively addressing navigation difficulties caused by low underwater visibility and complex flow fields.

[0021] The physical constraint embedding mechanism added to the path planning model includes the following steps: S31: Transform the hull structure strength parameters into a bearing capacity feature vector, and use it as the prior attribute input for the nodes of the graph structure learning network to constrain the cleaning path trajectory to not exceed the safe operating area of ​​the hull. Specifically, in the inspection of concrete structures in hydropower station dams, the system obtains structural strength parameters for different areas of the dam body through historical engineering data: the overflow surface has a higher concrete grade but may suffer internal damage due to long-term water erosion; the gallery area has dense reinforcement but a thin concrete cover. These parameters are transformed into bearing capacity feature vectors, which serve as prior attributes for the nodes of the graph structure learning network, inputting them into the path planning model. When the robot's planned inspection path approaches a weak area in the gallery, this feature vector automatically constrains the path trajectory to not exceed the safe bearing capacity range, forcing the path to detour to a thicker concrete area for operation. This mechanism effectively avoids secondary damage to the dam body caused by the robot arm applying inspection pressure to weak points in the structure, minimizing operational risks from uncontrollable to significant improvements in the inherent safety of underwater special equipment inspection.

[0022] S32: Convert the kinematic characteristics of the underwater robotic arm into a side weight attenuation factor, so that the path trajectory automatically increases the turning cost in the joint limit area; Specifically, in this embodiment, the seven-DOF robotic arm mounted on the underwater inspection robot operates on the curved surface of the dam, with each joint having a clearly defined angular limit. The system converts the kinematic characteristics of the robotic arm (such as the maximum joint angle and arm span) into an edge weight decay factor, which is embedded in the edge weight calculation of the graph structure learning network. When the planned path needs to cross the joint limit area (such as the concave corner of the dam), this factor automatically increases the turning cost, forcing the path planning algorithm to prioritize the trajectory where the robotic arm's posture is relaxed and each joint is far from the limit. Compared with traditional path planning that does not consider kinematic constraints, this embedding mechanism significantly reduces the joint over-limit alarm rate during operation, avoids emergency braking and path replanning caused by mechanical limits, and improves operational smoothness and equipment reliability. S33: Using the characteristics of the water flow vector field as a dynamic correction coefficient, the drift compensation of the cleaning equipment's travel speed is performed to eliminate trajectory deviation caused by the tidal current; Specifically, in this embodiment, the water flow near the dam's spillway is turbulent and its direction is variable. The system collects real-time water flow vector field data using a Doppler current meter. The direction and magnitude of the water flow are used as dynamic correction coefficients to compensate for the drift of the detection robot's travel speed: when an acute angle is detected between the robot's travel direction and the water flow direction, the thruster's reverse thrust is automatically increased to counteract downstream acceleration; when in a counter-current state, the travel speed setpoint is appropriately reduced to maintain posture stability. This compensation eliminates trajectory deviation caused by tidal currents, significantly reducing the deviation between the robot's actual and planned trajectories in strong flow areas, ensuring that the detection sensors can accurately align with the target structural surface, and improving data acquisition quality and the accuracy of detection results.

[0023] S40: Input the cleaning path trajectory sequence into the multidimensional attribution analysis engine. The multidimensional attribution analysis engine locates the main cause category of the cleaning efficiency deviation based on the machine self-learning algorithm and generates equipment travel speed compensation strategy and cleaning intensity compensation strategy. Specifically, the executed detection path trajectory is input into a multi-dimensional attribution analysis engine. The engine integrates structural health characteristics (abnormally high surface roughness in a certain area of ​​concrete), defect distribution topology characteristics (cracks are concentrated in specific high-stress areas), and equipment operating condition characteristics (the robotic arm's positioning accuracy decreases in high-flow areas). The engine identifies the primary cause as "enhanced detection is needed in areas with concentrated structural damage." It then generates equipment travel speed compensation strategies (reducing speed to 60% of standard speed for finer scanning in areas of concentrated damage) and lighting compensation strategies (increasing light source power to improve image clarity). This attribution analysis allows the compensation strategies to directly address the root cause of the problem, avoiding the traditional method of blindly repeating detection of ineffective areas and improving the efficiency of a single operation.

[0024] S50: Converts the compensation strategy into underwater robot thruster control commands and cleaning device control commands, issues them for execution, and records the execution log after execution; Specifically, the compensation strategy is translated into underwater robot thruster control commands (increasing lateral thrust power in strong current areas to maintain hovering stability), robotic arm control commands (automatically adjusting probe angle and focal length in areas of concentrated damage), and lighting control commands (locally enhancing light intensity). These commands are transmitted to the robot for execution via underwater acoustic communication or fiber optic links, while simultaneously recording a complete execution log (including timestamps, position coordinates, and device status). This transformation process converts abstract attribution results into executable control signals, achieving a seamless transition from data analysis to physical operations.

[0025] S60: Reacquire images of the ship's surface, calculate the improvement rate of cleaning effect, and feed the improvement effect evaluation results back to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0026] Specifically, after execution, the robot re-acquired images of the dam surface and calculated the improvement rate of detection effect through image comparison: the number of cracks detected increased, the missed detection rate decreased, and the accuracy of defect identification in high-stress areas was significantly improved. The improvement effect evaluation results were fed back to the S20 differential processing stage. The system identified "high-stress area features" as contributing the most in this operation, so the weight of this feature in subset partitioning was increased in subsequent tasks, and the sensitivity of the damage detection algorithm was optimized. This closed-loop feedback enables the path planning model to continuously learn the damage patterns and flow field characteristics of different underwater structures, forming a continuous evolution capability of "detection-planning-execution-evaluation-correction". It can adapt to new structures and new environmental conditions without manual reprogramming, promoting the intelligent transformation of underwater special equipment inspection from experience-driven to data-driven.

[0027] In summary, compared to traditional underwater inspection methods that rely on manual experience to set fixed paths and single-sensor navigation, this application systematically improves the targeting and operational efficiency of path planning through multi-source data fusion and differentiated processing. Traditional methods are prone to path deviation and blind spots in low visibility and complex flow environments. The graph structure learning network and compensation factor mechanism in this application effectively address these challenges, significantly reducing repetitive work rates and energy consumption. Furthermore, closed-loop feedback enables model self-evolution, greatly reducing the need for manual intervention. This provides a practical and replicable intelligent solution for the underwater special equipment inspection industry, improving the safety and economy of deep-water operations.

[0028] In step S20: Differentiating the comprehensive environmental dataset, dividing the dataset into multiple job subsets according to preset feature classification rules, and independently generating appropriate cleaning strategy compensation factors for each job subset, the steps include the following: S21: Based on the corrosion resistance characteristics of the hull material, the comprehensive environmental dataset is divided into a steel structure subset and a non-metallic subset. A rust enhancement detection factor is generated for the steel structure subset, and a biofilm texture enhancement factor is generated for the non-metallic subset. Specifically, based on the corrosion resistance characteristics of the dam structure materials, the comprehensive environmental dataset is divided into two subsets: the concrete dam body is classified as the concrete structure subset. Because concrete surfaces are prone to microbial growth and have fine cracks, the system generates a crack detection enhancement compensation factor for this subset, automatically adjusting the contrast and edge sharpening algorithms of the underwater camera to highlight hairline cracks and spalling defects on the concrete surface. The metal gates and steel structure accessories of the dam are classified as the metal structure subset. Because metal surfaces are prone to rust and pitting corrosion, the system generates a rust texture enhancement compensation factor for this subset, using multispectral imaging technology to enhance the texture features of rust spots and pits. This differentiated processing avoids the shortcomings of a single detection mode, such as missing fine cracks in concrete areas and misjudging the degree of corrosion in metal areas, significantly improving the targeting and accuracy of defect identification for structures of different materials. S22: Based on the density distribution characteristics of the attachments, the dataset is divided into a high-density attachment area subset and a low-density area subset. A spiral full-coverage path compensation factor is generated for the high-density area, and a linear scanning path compensation factor is generated for the low-density area. Specifically, the system divides the detection area into a high-density defect subset and a low-density intact subset based on the density distribution characteristics of defects on the dam surface. In areas with concentrated scour holes formed by long-term water erosion, the defect density is assessed as "high-density." The system generates a spiral full-coverage path compensation factor, instructing the underwater robot to use a dense spiral scanning path to ensure that each scour hole area is repeatedly inspected from multiple angles, avoiding the omission of deep cracks. In intact areas of the dam body far from the main discharge flow, the defect density is assessed as "low-density." The system generates a linear fast inspection path compensation factor, instructing the robot to use a fast linear scanning path to improve detection efficiency. This classification process allows path planning to invest more resources in critical risk areas and save operation time in non-critical areas, significantly improving overall detection efficiency while reducing robot energy consumption and underwater operation risks.

[0029] S23: Based on the characteristics of water flow disturbance intensity, the dataset is divided into a strong flow disturbance subset and a still water subset, and a countercurrent compensation factor is generated for the strong flow disturbance subset to correct the thruster power distribution; Specifically, the system divides the operating environment into a strong current disturbance subset and a calm water subset based on the intensity characteristics of water flow disturbance in different areas of the dam. Near the spillway outlet, where the water flow is turbulent and directionally unstable, the system generates a dynamic hovering compensation factor to adjust the robot's thruster power distribution in real time, increasing vertical thrust to counteract the buoyancy of the water flow and ensuring stable hovering and precise positioning of the robot in strong currents. In the deep, calm water areas of the dam, where the water flow is gentle, the system generates a standard propulsion compensation factor to maintain normal propulsion power and path speed. This compensation mechanism effectively overcomes the interference of complex underwater flow fields on robot displacement, avoids detection blind spots or equipment collision risks caused by positioning drift in strong current areas, and ensures the safety of the detection operation and the stability of data quality.

[0030] The differentiated processing and compensation factor generation mechanism constructed through S21-S23 systematically addresses the core pain points of traditional underwater inspection, which neglects dynamic environmental factors such as differences in structural materials, uneven defect distribution, and water flow interference. S21 enables the inspection system to adapt to different defect patterns in concrete and metal; S22 enables path planning to intelligently allocate resources between critical risk areas and intact areas; and S23 enables the robot to effectively overcome strong current interference and maintain stable operation. The synergy of these three mechanisms gives path planning scene awareness and adaptive capabilities, shifting from "fixed path scanning" to "intelligent differentiated patrolling," significantly improving the coverage, accuracy, and operational safety of underwater special equipment inspection, and driving the underwater inspection industry's intelligent transformation from experience-driven to data-driven.

[0031] In step S40: Inputting the cleaning path trajectory sequence into the multidimensional attribution analysis engine, the following steps are included: S41: Construct a health status degradation characteristic curve for the equipment, and take the time-series change rate of pusher wear and cleaning brush fatigue as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging. Specifically, after long-term inspection of the dam's concrete stilling basin, the system detected a continuous upward trend in the motor temperature of the underwater robot's thruster via sensors, with the temporal rate of change indicating a significantly faster rate of deterioration than initially observed. Simultaneously, the joint flexibility of the inspection arm decreased, the response latency of the camera gimbal increased, and the fatigue characteristic curve entered an accelerated degradation phase. After constructing the equipment health status degradation characteristic curve, the multi-dimensional attribution analysis engine determined that the robot was in a moderately aging state and automatically increased the attribution weight of "equipment aging" as the primary cause. Based on this judgment, the generated compensation strategy prioritizes preventative maintenance of the robot (replacing the thruster bearings and calibrating the inspection arm) rather than increasing inspection time. This avoids ineffective operations that could lead to equipment failure, ensuring that rectification directly addresses the root cause and significantly reduces inspection interruptions and repetitive work costs caused by equipment misjudgments.

[0032] S42: Extract the deviation characteristics of the attachment peeling efficiency, calculate the deviation ratio between the actual cleaning coverage and the standard coverage, and trigger the cleaning intensity compensation strategy to be executed first when the deviation exceeds the preset threshold. Specifically, during the inspection of the dam's metal gates, the system, through image comparison calculations, discovered a significant deviation between the actual and standard inspection coverage. The calculation results for the inspection efficiency deviation characteristic indicated a "severe deviation" level. The multi-dimensional attribution analysis engine determined "insufficient inspection coverage" as the primary efficiency anomaly and immediately triggered a priority execution queue for compensation strategies: automatically reducing the robot's travel speed, decreasing the inspection path step length, increasing the image acquisition frequency, and instructing the inspection arm to repeatedly scan complex areas such as the gate edges. This priority execution mechanism improved the inspection coverage from a deviation state to a qualified level, ensuring no corrosion defects at the gate edges were missed, and significantly improving inspection completeness and data reliability.

[0033] S43: Identify the bottleneck features of the hull structure, calculate the spatial matching degree between the reach of the cleaning arm and the complexity of the hull surface, and when the matching degree is lower than the preset threshold, increase the weight of the main cause of the structural cleaning blind spot and generate hull segment cleaning suggestions. In this embodiment, the system, through 3D spatial modeling analysis, found that the spatial matching degree between the reachability of the detection arm at the corner of the dam gallery and the complexity of the concrete surface was assessed as "poor". After identifying the structural bottleneck characteristics, the multidimensional attribution analysis engine increased the weight of the main cause "insufficient structural accessibility" and generated segmented detection suggestions: it suggested decomposing the gallery detection task into two sub-tasks, "straight line segment + corner segment", and using miniaturized detection probes or adjusting the robot's water entry angle for the corner segment. This suggestion, based on the actual spatial relationship, accurately locates the structural shortcomings of the detection operation, avoids repeatedly adjusting the robot's posture without fundamentally solving the problem, provides a data-driven decision-making basis for subsequent detection scheme optimization, and significantly improves the detection success rate and operational safety in complex structural areas.

[0034] For steps S41-S43, this application constructs a three-dimensional attribution analysis mechanism for equipment health, inspection efficiency, and structural bottlenecks, systematically addressing the core pain point of traditional underwater inspection that focuses only on defect identification while neglecting equipment status and structural accessibility. The degradation characteristic curve in S41 accurately identifies aging equipment, avoiding misjudgments as environmental interference; the deviation characteristic in S42 triggers dynamic compensation for priority execution, achieving rapid closed-loop correction; and the structural bottleneck characteristic in S43 guides the optimization of operational plans, providing long-term improvement directions. The synergy of these three elements enables the inspection strategy to directly target the root cause, avoiding the high cost of traditional blind and repetitive inspections, significantly improving the accuracy, efficiency, and safety of underwater special equipment inspection, and promoting the transformation of underwater engineering inspection towards a data-driven, intelligent, and autonomous operation mode.

[0035] In S50: after the step of converting the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and recording the execution log after execution, the following steps are included: S501: Calculate the residual between the cleaning effect improvement rate after execution and the expected target value, and construct an effect deviation evaluation vector; Specifically, after a week of continuous underwater structural inspection of the dam's stilling basin, the system compared the inspected images with historical benchmark data and found significant discrepancies between the defect detection rate and the expected target in certain areas. Through residual calculation, the system constructed an effectiveness deviation assessment vector: the defect detection rate deviation was large in the scour area, moderate in the corrosion identification of metal gates, and small in the integrity assessment of straight dam sections. This vector quantification reveals the gap between the current "high-density defect area + strong corrosion compensation" strategy and actual inspection needs, providing precise direction for subsequent attribution analysis and avoiding judgments of inspection effectiveness based solely on subjective experience. S502: Based on the deviation assessment vector, perform attribution analysis to identify the key influencing factors that lead to the deviation as improper setting of cleaning strategy compensation factors or outdated operation subset division standards. Specifically, based on the deviation evaluation vector, the system performed attribution analysis: it was found that the boundary conditions for the "high-density defect area" subset were set too loosely (only areas with defect density exceeding a relatively high threshold were included), resulting in some medium-density defect areas not being included in the enhanced detection range; at the same time, it was found that the weight coefficient of the "rust detection" compensation factor was too low, failing to fully highlight the rust texture characteristics of the metal gate area. After identifying these key influencing factors, the system clarified that the root cause of the deviation lay in the outdated subset division standard and improper compensation factor settings, rather than simply insufficient equipment power, thus pointing the way for accurate correction.

[0036] S503: Utilize incremental learning algorithms to dynamically adjust the subset partitioning boundary conditions in the differentiation processing stage and automatically correct the weight coefficients in the compensation factor generation logic. Specifically, using an incremental learning algorithm, the system dynamically adjusts the differential processing steps: the boundary conditions for dividing the "high-density defect area" subset are tightened from exceeding a high threshold to exceeding a medium threshold, allowing more potential risk areas to be included in the enhanced detection range; simultaneously, the weight coefficient of the "corrosion detection" compensation factor is increased, enhancing the model's sensitivity to metal corrosion features. This adaptive correction makes the subsequent detection strategy more closely match the actual deterioration state of the dam structure, achieving parameter optimization without manual intervention and significantly improving the accuracy and economy of the detection strategy.

[0037] S504: The revised subset partitioning criteria and compensation factors are fed back to the path planning model to achieve adaptive optimization of the next generation of clean paths and control parameters, forming a continuous evolutionary closed loop. The revised subset partitioning criteria and compensation factors are fed back to the path planning model and automatically applied in the next round of periodic dam inspections. The new generation of inspection results shows a significant improvement in the defect detection rate in scour areas and a marked improvement in the accuracy of metal gate corrosion identification; the overall inspection effect meets the expected goals. This closed-loop mechanism enables the system to continuously learn and self-evolve, automatically adapting to the aging changes of underwater structures over time (such as crack propagation and increased corrosion), forming a complete intelligent cycle of "inspection-planning-execution-evaluation-correction," greatly reducing the workload of manual recalibration and scheme adjustments.

[0038] The adaptive adjustment closed-loop optimization mechanism constructed through S501-S504 systematically solves the core pain point of the rigid underwater detection path planning strategy and its inability to self-evolve. This mechanism enables the system to automatically learn and adapt to the deterioration patterns of different underwater structures and environmental changes, continuously optimize detection parameters, reduce reliance on manual adjustments, and promote the evolution of underwater special equipment detection towards a fully autonomous and intelligent automated mode, significantly improving the stability, reliability, and economy of long-term operation.

[0039] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0040] In one embodiment, a data fusion-based underwater path planning device is provided, which corresponds one-to-one with the data fusion-based underwater path planning method described in the above embodiments. For example... Figure 2 As shown, the underwater path planning device based on data fusion includes: The comprehensive environmental dataset construction unit 1 is used to collect underwater hull status data in real time based on a multi-source heterogeneous sensor array, and to perform spatiotemporal alignment and standardization fusion of the hull status data to construct a comprehensive environmental dataset. The cleaning strategy compensation factor generation unit 2 is used to perform differential processing on the comprehensive environmental dataset. According to the preset feature classification rules, the dataset is divided into multiple job subsets, and an appropriate cleaning strategy compensation factor is generated independently for each job subset. The cross-modal path optimization unit 3 is used to input the operation subset and the cleaning strategy compensation factor into the pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, and outputs a cleaning path trajectory sequence and a cleaning equipment action command sequence. The compensation strategy generation unit 4 is used to input the cleaning path trajectory sequence into the multidimensional attribution analysis engine. The multidimensional attribution analysis engine locates the main cause category of the cleaning efficiency deviation based on the machine self-learning algorithm and generates the equipment travel speed compensation strategy and the cleaning intensity compensation strategy. The control command execution unit 5 is used to convert the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and record the execution log after execution; The optimized feedback correction unit 6 is used to re-acquire images of the ship's surface, calculate the improvement rate of the cleaning effect, and feed back the improvement effect evaluation results to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0041] Specific limitations regarding the data fusion-based underwater path planning device can be found in the limitations of the data fusion-based underwater path planning method described above, and will not be repeated here. Each module in the aforementioned data fusion-based underwater path planning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0042] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores the database. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a data fusion-based underwater path planning method.

[0043] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Based on the real-time acquisition of underwater hull status data by a multi-source heterogeneous sensor array, the hull status data includes at least hull surface image data, hull attachment type characteristics, underwater flow velocity field distribution data and underwater visibility data. The above data are spatiotemporally aligned and standardized and fused to construct a comprehensive environmental dataset. The comprehensive environmental dataset is differentiated and divided into multiple operation subsets based on the characteristics of ship hull material, attachment density, and water flow disturbance intensity. For each operation subset, an appropriate cleaning strategy compensation factor is generated independently. The operation subset and the cleaning strategy compensation factor are input into a pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, outputting a cleaning path trajectory sequence and a cleaning equipment action command sequence. The cleaning path trajectory sequence is input into a multidimensional attribution analysis engine. The engine integrates the ship's health status characteristics, the topological characteristics of the attachment distribution, and the equipment operating condition characteristics to locate the main cause category of the cleaning efficiency deviation and generate equipment travel speed compensation strategy and cleaning intensity compensation strategy. The compensation strategy is converted into control commands for the underwater robot's thrusters and the cleaning device, which are then issued and executed, and an execution log is recorded after execution. The cleaning effect improvement rate is calculated by re-acquiring images of the ship's surface and feeding the improvement effect evaluation results back to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0044] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Based on the real-time acquisition of underwater hull status data by a multi-source heterogeneous sensor array, the hull status data includes at least hull surface image data, hull attachment type characteristics, underwater flow velocity field distribution data and underwater visibility data. The above data are spatiotemporally aligned and standardized and fused to construct a comprehensive environmental dataset. The comprehensive environmental dataset is differentiated and divided into multiple operation subsets based on the characteristics of ship hull material, attachment density, and water flow disturbance intensity. For each operation subset, an appropriate cleaning strategy compensation factor is generated independently. The operation subset and the cleaning strategy compensation factor are input into a pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, outputting a cleaning path trajectory sequence and a cleaning equipment action command sequence. The cleaning path trajectory sequence is input into a multidimensional attribution analysis engine. The engine integrates the ship's health status characteristics, the topological characteristics of the attachment distribution, and the equipment operating condition characteristics to locate the main cause category of the cleaning efficiency deviation and generate equipment travel speed compensation strategy and cleaning intensity compensation strategy. The compensation strategy is converted into control commands for the underwater robot's thrusters and the cleaning device, which are then issued and executed, and an execution log is recorded after execution. The cleaning effect improvement rate is calculated by re-acquiring images of the ship's surface and feeding the improvement effect evaluation results back to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

[0045] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0047] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An underwater path planning method based on data fusion, characterized in that, The method includes the following steps: real-time acquisition of underwater hull status data based on a multi-source heterogeneous sensor array, spatiotemporal alignment and standardization fusion of the hull status data, and construction of a comprehensive environmental dataset; The comprehensive environmental dataset is processed differentially, and the dataset is divided into multiple job subsets according to the preset feature classification rules. For each job subset, an appropriate cleaning strategy compensation factor is generated independently. The operation subset and the cleaning strategy compensation factor are input into a pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship's topology and integrates the compensation factor to perform cross-modal path optimization, outputting a cleaning path trajectory sequence and a cleaning equipment action command sequence. The cleaning path trajectory sequence is input into the multidimensional attribution analysis engine, which uses a machine self-learning algorithm to locate the main cause category of the cleaning efficiency deviation and generates equipment travel speed compensation strategy and cleaning intensity compensation strategy. The compensation strategy is converted into control commands for the underwater robot's thrusters and the cleaning device, which are then issued and executed, and an execution log is recorded after execution. The cleaning effect improvement rate is calculated by re-acquiring images of the ship's surface and feeding the improvement effect evaluation results back to the differential processing stage to correct the subset partitioning criteria and compensation factor generation logic.

2. The underwater path planning method based on data fusion according to claim 1, characterized in that, The step of differentially processing the comprehensive environmental dataset, dividing the dataset into multiple job subsets according to preset feature classification rules, and independently generating an appropriate cleaning strategy compensation factor for each job subset includes the following steps: Based on the corrosion resistance characteristics of the ship hull material, the comprehensive environmental dataset is divided into a steel structure subset and a non-metallic subset. A rust enhancement detection factor is generated for the steel structure subset, and a biofilm texture enhancement factor is generated for the non-metallic subset. Based on the density distribution characteristics of the attachments, the dataset is divided into a high-density attachment area subset and a low-density attachment area subset. A spiral full-coverage path compensation factor is generated for the high-density area, and a linear scanning path compensation factor is generated for the low-density area. Based on the characteristics of water flow disturbance intensity, the dataset is divided into a strong flow disturbance subset and a still water subset. A countercurrent compensation factor is generated for the strong flow disturbance subset to correct the thruster power distribution.

3. The underwater path planning method based on data fusion according to claim 1, characterized in that, Adding a physical constraint embedding mechanism to the path planning model includes the following steps: The hull structure strength parameters are transformed into load-bearing capacity feature vectors, which are used as prior attribute inputs to the nodes of the graph structure learning network to constrain the cleaning path trajectory to not exceed the safe operating area of ​​the hull. The kinematic characteristics of the underwater robotic arm are converted into edge weight attenuation factors, so that the turning cost is automatically increased in the joint limit area of ​​the path trajectory. The characteristics of the water flow vector field are used as dynamic correction coefficients to compensate for the drift of the cleaning equipment's travel speed, thus eliminating trajectory deviation caused by tidal currents.

4. The underwater path planning method based on data fusion according to claim 1, characterized in that, The steps involved in inputting the cleaning path trajectory sequence into the multidimensional attribution analysis engine are as follows: Construct a health status degradation characteristic curve for equipment, and use the time-series change rate of pusher wear and cleaning brush fatigue as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging. Extract the deviation characteristics of the attachment removal efficiency, calculate the deviation ratio between the actual cleaning coverage and the standard coverage, and trigger the cleaning intensity compensation strategy to be executed first when the deviation exceeds the preset threshold. Identify the bottleneck features of the hull structure, calculate the spatial matching degree between the reach of the cleaning arm and the complexity of the hull surface, and when the matching degree is lower than the preset threshold, increase the weight of the main cause of the structural cleaning blind spot and generate hull segment cleaning suggestions.

5. The underwater path planning method based on data fusion according to claim 1, characterized in that, After the step of converting the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and recording the execution log after execution, the following steps are included: The residual between the cleaning effect improvement rate after execution and the expected target value is calculated to construct an effect deviation evaluation vector; Based on the deviation evaluation vector, attribution and source analysis is performed to identify the key influencing factors that lead to the deviation as improper setting of cleaning strategy compensation factors or outdated operation subset division standards. Incremental learning algorithms are used to dynamically adjust the subset partitioning boundary conditions in the differentiation process and to automatically correct the weight coefficients in the compensation factor generation logic. The revised subset partitioning criteria and compensation factors are fed back to the path planning model to achieve adaptive optimization of the next generation of clean paths and control parameters, forming a continuous evolutionary closed loop.

6. The underwater path planning method based on data fusion according to claim 1, characterized in that, Based on the real-time acquisition of underwater hull state data by a multi-source heterogeneous sensor array, the hull state data is spatiotemporally aligned and standardized and fused to construct a comprehensive environmental dataset. The steps include: synchronously acquiring high-definition images of the hull surface, side-scan sonar echo signals and Doppler current profile data through underwater edge computing nodes to construct a multimodal hull state vector. By utilizing a preset synchronization triggering mechanism, the multimodal state vector is aligned with the common time axis at the millisecond level, eliminating the sampling delay difference of underwater sensors; The aligned data stream is then subjected to sliding window filtering and abnormal pulse removal to eliminate underwater acoustic multipath interference and equipment mechanical vibration noise.

7. An underwater path planning device based on data fusion, applied to the underwater path planning method based on data fusion as described in any one of claims 1 to 6, characterized in that, The device includes: The comprehensive environmental dataset construction unit (1) is used to collect underwater hull status data in real time based on a multi-source heterogeneous sensor array, and to perform spatiotemporal alignment and standardization fusion of the hull status data to construct a comprehensive environmental dataset. The cleaning strategy compensation factor generation unit (2) is used to perform differential processing on the comprehensive environmental dataset, divide the dataset into multiple job subsets according to the preset feature classification rules, and generate an appropriate cleaning strategy compensation factor for each job subset independently. The cross-modal path optimization unit (3) is used to input the operation subset and the cleaning strategy compensation factor into the pre-trained path planning model. The path planning model uses a graph structure learning network to extract features of the ship topology and integrates the compensation factor to perform cross-modal path optimization, and outputs the cleaning path trajectory sequence and the cleaning equipment action instruction sequence. The compensation strategy generation unit (4) is used to input the cleaning path trajectory sequence into the multidimensional attribution analysis engine. The multidimensional attribution analysis engine locates the main cause category of the cleaning efficiency deviation based on the machine self-learning algorithm and generates the equipment travel speed compensation strategy and the cleaning intensity compensation strategy. The control command execution unit (5) is used to convert the compensation strategy into underwater robot thruster control commands and cleaning device control commands for execution, and to record the execution log after execution. The optimized feedback correction unit (6) is used to re-acquire images of the ship's surface to calculate the cleaning effect improvement rate, and feeds back the improvement effect evaluation results to the differentiation processing stage to correct the subset division criteria and compensation factor generation logic.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the underwater path planning method based on data fusion as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the underwater path planning method based on data fusion as described in any one of claims 1 to 6.

Images