Cable tunnel robot path planning method and electronic equipment

Abstract

The invention discloses a cable tunnel robot path planning method and electronic equipment. The method comprises the following steps: in response to an inspection task request for a cable tunnel, acquiring environment sensing data and robot state data acquired by an inspection robot; based on the environmental perception data, identifying a target risk area in the cable tunnel, the target risk area including a cable overheated area, a voiceprint abnormal area and a partial discharge area; generating a global inspection path based on the target risk area and the inspection robot state data; and issuing the global inspection path to the inspection robot for the inspection robot to execute path planning and local obstacle avoidance. The technical problems of low accuracy and poor safety of path planning of the inspection robot caused by inaccurate risk identification and avoidance in path planning of the inspection robot in the cable tunnel in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of smart grids, and more specifically, to a path planning method and electronic device for a cable tunnel robot. Background Technology

[0002] Cable tunnels, as critical channels for power transmission, have complex internal environments and present various potential risks, such as partial discharge, cable overheating, and abnormal noises. To ensure power transmission safety, automated inspection using inspection robots has become an industry trend. Inspection robots need to plan safe and efficient inspection paths within the tunnel to complete comprehensive inspection tasks.

[0003] Currently, most robot path planning methods in related technologies rely on pre-set maps or single environmental sensor data. While these methods can avoid static obstacles, they struggle to address the complex risks of dynamic changes and multimodal interactions in cable tunnels. In other words, the risk identification and avoidance in path planning for inspection robots in cable tunnels is inaccurate, resulting in low accuracy and poor safety in path planning.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a path planning method and electronic device for a cable tunnel robot, which at least solves the technical problem in the related art that the inaccurate risk identification and avoidance in the path planning of inspection robots in cable tunnels leads to low accuracy and poor safety in the path planning of inspection robots.

[0006] According to one aspect of the embodiments of this application, a path planning method for a cable tunnel robot is provided, comprising: in response to a request for an inspection task of a cable tunnel, acquiring environmental perception data and robot status data collected by the inspection robot, wherein the environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data, and the robot status data includes at least the power information and equipment operating status information of the inspection robot; identifying target risk areas in the cable tunnel based on the environmental perception data, wherein the target risk areas include cable overheating areas, acoustic signature abnormality areas, and partial discharge areas; generating a global inspection path based on the target risk areas and the inspection robot status data; and distributing the global inspection path to the inspection robot for the inspection robot to perform path planning and local obstacle avoidance.

[0007] According to another aspect of the embodiments of this application, a cable tunnel robot path planning device is also provided, comprising: an acquisition module, configured to acquire environmental perception data and robot status data collected by the inspection robot in response to a cable tunnel inspection task request, wherein the environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data, and the robot status data includes at least the inspection robot's power information and equipment operating status information; an identification module, configured to identify target risk areas in the cable tunnel based on the environmental perception data, wherein the target risk areas include cable overheating areas, acoustic signature abnormal areas, and partial discharge areas; a generation module, configured to generate a global inspection path based on the target risk areas and the inspection robot status data; and a sending module, configured to send the global inspection path to the inspection robot for the inspection robot to perform path planning and local obstacle avoidance.

[0008] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, which stores multiple instructions adapted for a cable tunnel robot path planning method to be loaded by a processor and executed at any one of them.

[0009] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the cable tunnel robot path planning methods.

[0010] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of any one of the cable tunnel robot path planning methods.

[0011] In this embodiment, in response to a cable tunnel inspection task request, environmental perception data and robot status data collected by the inspection robot are acquired. The environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data. The robot status data includes at least the robot's power information and equipment operating status information. Based on the environmental perception data, target risk areas in the cable tunnel are identified, including cable overheating areas, acoustic signature abnormality areas, and partial discharge areas. Based on the target risk areas and the inspection robot status data, a global inspection path is generated. The global inspection path is then distributed to the inspection robot. This system, used by inspection robots to perform path planning and local obstacle avoidance, achieves real-time environmental data collection through multi-modal sensors (discharge sensing, infrared thermal imaging, and voiceprint acquisition). Combined with the robot's own power and equipment status monitoring, it accurately identifies potentially high-risk areas and dynamically plans safe paths. This improves the accuracy and comprehensiveness of risk identification during robot inspections, ensuring the feasibility and safety of robot path planning. Furthermore, it solves the technical problem in related technologies where inaccurate risk identification and avoidance in path planning for inspection robots in cable tunnels leads to low accuracy and poor safety in inspection robot path planning. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0013] Figure 1 This is a schematic diagram of the implementation environment of a cable tunnel robot path planning method provided in an embodiment of this application;

[0014] Figure 2 This is a flowchart of a cable tunnel robot path planning method according to an embodiment of this application;

[0015] Figure 3 This is a schematic diagram of a cable tunnel robot path planning device according to an embodiment of this application;

[0016] Figure 4 This is a schematic diagram of the structure of an edge computing node provided in an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] First, to facilitate understanding of the embodiments of this application, some terms or nouns involved in this application will be explained below:

[0020] Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain better results.

[0021] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge sub-models to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence.

[0022] Discharge sensing data refers to raw or pre-processed signal data collected by specialized equipment such as ultra-high frequency (UHF) sensors and ultrasonic sensors, used to characterize partial discharge phenomena inside or on the surface of cables and their accessories. This data typically includes information such as the amplitude, phase, and frequency of the discharge pulses.

[0023] Power information refers to the real-time status parameters of the inspection robot's power system, mainly including the current remaining battery capacity (State of Charge, SOC), the estimated continuous running time under the current load, and the battery health status (State of Health, SOH).

[0024] Equipment operating status information refers to a set of data reflecting the working status of each subsystem of the robot, including but not limited to: the torque and speed of the drive motor, the rotation angle and response status of the servo motor, the load rate of the computing unit, the signal strength of the communication module, and the online / offline status and calibration status of all sensors.

[0025] Path tracking is the core task performed by the local path planner. It refers to the robot calculating and outputting control commands (such as linear velocity and angular velocity) in real time based on the received global path coordinate point sequence through algorithms such as PID control, Pure Pursuit, or Model Predictive Control (MPC), so that its actual motion trajectory can accurately follow the preset global path.

[0026] Temperature gradient features, extracted from infrared thermal imaging data, are used to quantify the degree of spatial temperature variation. They reflect the amount of temperature change per unit distance, with high gradient regions typically corresponding to abnormal heat sources such as overheating cable joints.

[0027] Frequency domain distribution characteristics are obtained through spectral analysis of the time-domain acoustic signal, such as by performing a Fast Fourier Transform (FFT). It describes the energy distribution of the sound signal across different frequency components and can be used to identify the characteristic frequencies corresponding to specific mechanical faults.

[0028] Energy attenuation characteristics characterize how the energy of sound or electromagnetic signals decreases as the propagation distance increases during sound or electromagnetic wave propagation in a cable tunnel. This characteristic helps in locating abnormal sound sources or discharge sources.

[0029] Pulse sequence features, extracted from partial discharge sensing data, describe the distribution patterns of discharge pulses along the time axis. These mainly include pulse repetition rate, pulse time interval distribution, and pulse amplitude distribution, and are used to determine the type and severity of the discharge.

[0030] Discharge intensity characteristics are characteristic parameters used to quantify the activity and energy of partial discharge, and typically include, but are not limited to, statistical indicators such as average discharge quantity, maximum discharge quantity, and discharge power.

[0031] Multimodal feature fusion is a data processing technique that aims to integrate features (such as temperature, frequency, and pulse) from different sensors (such as infrared, acoustic signature, and UHF) with different physical meanings and dimensions into a unified and complementary feature representation through feature-level or decision-level fusion algorithms, in order to provide more comprehensive environmental perception.

[0032] Risk types are classifications of risks based on potential physical failure phenomena. In this application, three basic types are mainly defined: cable overheating zone, usually caused by excessive resistance, poor contact, etc.; abnormal sound pattern zone, usually caused by mechanical vibration, corona, etc.; and partial discharge zone, usually caused by insulation defects.

[0033] Risk coupling strength is a quantitative indicator used to describe the degree of mutual influence and reinforcement between the probability or intensity of occurrence of two or more different risk types in the same spatial area.

[0034] Synergistic risk characteristics refer to new, comprehensive risk patterns that arise from the interaction of multiple risk types, characteristics not present in individual risks. For example, overheating may accelerate insulation aging, thereby exacerbating partial discharge; this "thermal-electrical" synergistic effect is a synergistic risk characteristic.

[0035] A composite risk area is a spatially overlapping region in a cable tunnel where risk mechanisms interact with each other. This region contains two or more types of risks simultaneously, and its overall risk level is higher than the simple superposition of individual risks acting independently due to risk coupling.

[0036] Multimodal feature association model, a data-driven mathematical model (such as one based on the covariance matrix), is used to analyze and quantify the statistical correlation between different modal features.

[0037] The thermoelectric coupling coefficient, a specific value calculated using a multimodal characteristic correlation model, is used to accurately quantify the degree of linear correlation between temperature gradient characteristics and discharge intensity characteristics. A positive value indicates a positive correlation between temperature increase and discharge enhancement.

[0038] The acoustic-electric coupling coefficient is a specific value calculated through a multimodal feature correlation model. It is used to accurately quantify the degree of linear correlation between frequency domain distribution features (acoustic) and pulse sequence features (electric).

[0039] The covariance matrix is ​​a mathematical matrix in which each element represents the covariance between two random variables (in this case, features of different modalities). It measures the direction and strength of the linear correlation between these features and is fundamental to building feature association models.

[0040] Information entropy theory, derived from the mathematical theory of information, is used to measure the uncertainty or randomness of information. In this application, it is applied to evaluate the “disorder” or “uncertainty” of each feature in its spatial distribution.

[0041] Uncertainty measure, a scalar value calculated based on information entropy theory, is used to quantify the discreteness and unpredictability of a feature's values ​​at various points in space. The more uniform and random the distribution of feature values, the higher its uncertainty measure.

[0042] Information entropy weights are calculated based on the uncertainty measure of each feature. Generally, features with lower uncertainty (i.e., more reliable and stable information) are assigned higher weights and contribute more to the fusion process.

[0043] The dynamic adaptive weighting coefficient is a final fusion weight that integrates the coupling coefficients between features (such as thermoelectric and acoustic-electric coupling coefficients) and information entropy weights. It is "dynamic" because it is calculated in real time based on the actual situation of each input data; it is "adaptive" because it can automatically adjust according to feature quality and correlation.

[0044] The comprehensive risk feature map is a two-dimensional raster map generated by applying dynamic adaptive weighting coefficients to weight and fuse all input features. Each pixel (raster) value in this map represents the comprehensive risk probability or intensity at that spatial location.

[0045] Spatial correlation refers to the statistical dependence of characteristic values ​​of different risk types on the spatial distribution of cable tunnels, including linear and nonlinear relationships.

[0046] The strength of linear correlation is mainly measured by indicators such as the Pearson correlation coefficient, reflecting the degree to which two risk characteristics follow a linear relationship in spatial distribution.

[0047] Nonlinear dependency strength is mainly measured by indicators such as mutual information, reflecting the strength of the complex relationship between two risk features in spatial distribution that cannot be described by a linear model.

[0048] Historical risk event data refers to structured data collected from past cable tunnel inspection and maintenance records, including the time of the risk occurrence, the precise location, the type of risk, the final confirmed fault level, the handling measures, and the results.

[0049] Spatial interpolation is an algorithm that estimates the values ​​of unknown regions based on known discrete point data. This application employs methods such as Kriging interpolation to interpolate discrete risk point data into a continuous risk value distribution map.

[0050] The risk value distribution map, a continuous surface plot generated by spatial interpolation, intuitively shows the smooth variation of risk intensity throughout the entire cable tunnel space.

[0051] Contour extraction is an image processing technique that extracts all points with equal risk values ​​from a two-dimensional scalar field such as a risk value distribution map and connects them into lines (i.e., contour lines).

[0052] Multi-level risk thresholds are a set of preset threshold values ​​based on cable tunnel safety standards and operation and maintenance experience, used to classify different risk levels. For example, risk values ​​in the range [0, 0.3) are defined as "low risk", [0.3, 0.7) as "medium risk", and [0.7, 1.0] as "high risk".

[0053] Regional boundaries are defined by extracting contour lines from the risk value distribution map and filtering them according to multi-level risk thresholds, resulting in various closed geometric polygons. These boundaries clearly define the extent of areas with different risk levels.

[0054] Cable tunnel safety standards are mandatory or guiding technical documents and codes of conduct formulated by industry authorities or enterprises regarding the design, construction, operation and maintenance, and safety management of cable tunnels.

[0055] The impact of path planning on quantification is the process of transforming the static attributes (type, level) and dynamic attributes (trend of change) of the identified risk areas into numerical dynamic risk weights that the path planning algorithm can directly understand and process.

[0056] Dynamic risk weights are numerical values ​​assigned to each risk region for path cost calculation. They are not only based on the current risk level but also include predictions of its future evolution trends (such as the rate of expansion), and are therefore "dynamically" changing.

[0057] A risk propagation model is a physical or data-driven model that describes the spatial diffusion and spread of risks (such as overheating and discharge) over time under specific environmental conditions in cable tunnels.

[0058] Spatial range change trend refers to the specific patterns and results of how the area, geometry, and location of a certain risk area will evolve over a future period of time (such as the entire inspection cycle), as predicted by the risk propagation model.

[0059] A* algorithm, a classic graph search algorithm, efficiently finds the optimal path by evaluating the estimated total cost of reaching the target point from the starting point through the current node (f(n) = g(n) + h(n), where g(n) is the actual cost and h(n) is the heuristically estimated cost).

[0060] Path cost, in path planning algorithms, is a quantitative metric used to evaluate the quality of a path. In this application, it is a comprehensive value that includes distance cost, risk cost, and time cost.

[0061] A risk propagation predictor is a specific software module or algorithm instance that encapsulates a risk propagation model. In this application, it specifically refers to a model trained using machine learning methods (such as Recurrent Neural Networks (RNNs) or Long Short-Term Memory Networks (LSTMs)) that is capable of predicting the spatiotemporal evolution of risk based on historical and real-time data.

[0062] Time slicing is an analytical method that divides a continuous, future inspection period into a series of equally spaced, discrete time points (such as t1, t2, t3, …) to independently analyze the predicted state of risk at each time point.

[0063] Risk area prediction boundary: For a specific time slice, the geometric boundary of the spatial range occupied by the risk area predicted by the risk propagation predictor at that moment.

[0064] Range expansion rate, a quantitative indicator, is calculated by comparing the predicted boundaries of risk areas in adjacent time slices (e.g., changes in calculated area), and represents the average rate of expansion of the risk area per unit time.

[0065] Intensity change trend, a quantitative indicator used to describe the direction (increase or decrease) and magnitude of changes in the core intensity of a risk area over a future period. For example, the slope obtained through linear regression analysis.

[0066] Time-series risk intensity data is a sequence of risk intensity values ​​obtained from multiple observations or predictions of the same area, arranged in chronological order.

[0067] A risk distribution raster map is a form of spatial data representation that divides a cable tunnel map into a uniform grid (raster) and assigns a risk intensity value to each grid cell, thereby forming a digital risk map.

[0068] Area change rate, an intermediate measure in calculating the rate of expansion of the area, refers to the relative percentage change in the area of ​​the risk zone between adjacent time slices.

[0069] The centroid movement vector is a quantity that has both magnitude and direction. Its magnitude represents the distance the centroid (geometric center) of the risk region moves between two consecutive time slices, and its direction represents the overall direction of movement of the risk region.

[0070] The core area of ​​a risk zone is the portion of a risk area where the risk intensity is higher than that of the surrounding areas. It is usually defined by setting a higher intensity threshold.

[0071] An intensity characteristic value is a single numerical value used to represent the intensity level of a risk area at a specific moment. Depending on the type of risk, it can be the maximum temperature, the sum of discharge frequencies, or the peak sound energy in that area, etc.

[0072] Intensity change time series is a sequence formed by arranging the intensity feature values ​​of a certain risk area in chronological order across different time slices.

[0073] Trend fitting analysis involves performing mathematical analysis on time series of intensity changes, such as using linear fitting or moving average methods, to extract their inherent trends (such as rising, falling, or stabilizing) and quantify their rate of change.

[0074] The weighting strategy is a set of predefined rules and logic used to assign an initial base weight to a risk area based on its risk type and current risk level.

[0075] Base weights are benchmark weight values ​​assigned to different categories of risk areas based on a weighting allocation strategy, without considering dynamic changes. For example, the base weight for "partial discharge area" can be set to be greater than that for "cable overheating area".

[0076] The dynamic adjustment factor, a multiplicative coefficient, is used to adjust the base weights. It is calculated based on the dynamic attributes of the risk area (such as the rate of expansion and the trend of intensity change); the more drastic the dynamic changes, the larger the factor usually is.

[0077] Distance cost, a component of path cost, is positively correlated with the total length of the planned path. Optimizing this aims to improve inspection efficiency and reduce energy consumption and time.

[0078] Risk cost, a core component of path cost, is determined by the sum of the dynamic risk weights of all risk areas traversed or approached by the path and the robot's exposure time in those areas. Optimizing this aims to maximize inspection safety.

[0079] Time cost, a component of path cost, is related to the estimated total time to complete the entire inspection task. Optimizing this aims to improve task execution efficiency.

[0080] The comprehensive cost function, a mathematical expression, combines distance cost, risk cost, time cost, etc., into a single scalar value. The goal of path planning is to find the path that minimizes this function value.

[0081] A time window refers to a specific period of time, calculated based on the robot's estimated moving speed, that it needs to traverse the neighborhood of a certain risk area.

[0082] Time margin, a safety metric, is calculated as: predicted expansion time of the risk area minus the robot's expected departure time from that area. A positive and larger time margin indicates greater safety; a negative value indicates that the robot may encounter risk expansion.

[0083] Route replanning is a local path optimization mechanism triggered when the system detects insufficient time margin on a path. Specific strategies may include: fine-tuning waypoints within the global path framework to move the path further away from the risk zone (increasing detour distance), or selecting a shorter or faster alternative path at the algorithmic level. Essentially, it reduces travel time by optimizing path geometry.

[0084] Figure 1 This is a schematic diagram illustrating the implementation environment of a cable tunnel robot path planning method provided in this application embodiment. See also... Figure 1 The implementation environment may include an edge computing node 110 and a robot controller 140.

[0085] Edge computing node 110 is an independent physical edge computing node, or an edge computing node cluster or distributed system composed of multiple physical edge computing nodes, or a cloud edge computing node that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms. Edge computing node 110 is connected to robot controller 140 via a wireless network, and edge computing node 110 can interact with robot controller 140 for data exchange.

[0086] The robot controller 140 can directly control the inspection robot. In this embodiment, the robot controller 140 can control the inspection robot to perform inspections in the cable tunnel.

[0087] In this embodiment, since a large number of data processing processes are involved, the computing power of the robot controller 140 may not be able to meet the needs of these data processing processes. Therefore, the edge computing node 110 can perform the above data processing processes, and the robot controller 140 can obtain and execute the results of the data processing.

[0088] In the path planning methods for cable tunnel inspection robots in related technologies, a static path model is constructed based on a pre-set environmental map and single-modal sensor data. However, due to the multimodal risk coupling phenomenon within cable tunnels, including abrupt temperature gradient changes, partial discharge pulses, and acoustic signal anomalies, risk identification mechanisms based on single-source sensing data struggle to accurately decouple the spatial distribution characteristics of complex risks. This results in the path planning module being unable to dynamically respond to the collaborative evolution trends of risk areas. When the inspection robot performs its tasks, the spatiotemporal mismatch between its global path and real-time risk situation triggers a surge in path replanning frequency. Furthermore, the static modeling method based on equipment operating status and power constraints is prone to misjudging path feasibility.

[0089] For example, in tunnel structures containing multi-circuit high-voltage cables, electromagnetic interference caused by partial discharge can alter the temperature sampling accuracy of infrared thermal imaging sensors, while thermal radiation from overheated cable joints can affect the frequency domain feature extraction of acoustic sensors. Traditional path planning systems employ an architecture that independently processes data from each sensor, failing to establish a correlation model between discharge intensity and temperature anomalies. This leads to boundary identification errors in composite risk areas exceeding permissible thresholds. When the inspection robot travels along a preset path, its lidar, while able to avoid physical obstacles, cannot predict the expansion of dynamic risk areas caused by cable insulation deterioration, resulting in the robot repeatedly entering high-risk areas and causing equipment damage. If these issues are not addressed, the failure of spatial decoupling of multimodal risk features will cause the path planning system to continuously generate high-risk inspection routes, increasing the operational risks to both the robot and cable equipment. The lag in identifying composite risk areas will cause delays in dynamic path adjustments, extending inspection cycles and resulting in missing critical equipment status monitoring data. Static modeling deviations between equipment operating status and power constraints may lead to path planning results exceeding the robot's actual movement capabilities, causing task interruptions and emergency return events, severely impacting the reliability and response time of the power tunnel maintenance system.

[0090] To address the aforementioned issues, this application provides an embodiment of a method for cable tunnel robot path planning. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here. Figure 2 This is a flowchart of a cable tunnel robot path planning method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0091] In response, this application proposes the following technical solution: Figure 2 This is a flowchart of a cable tunnel robot path planning method according to an embodiment of this application, such as... Figure 2As shown, taking the execution subject as an edge computing node as an example, the method includes the following steps.

[0092] 201. In response to the request for an inspection task of a cable tunnel, acquire the environmental perception data and robot status data collected by the inspection robot. The environmental perception data includes discharge perception data, infrared thermal imaging data and acoustic signature data. The robot status data includes at least the power information and equipment operating status information of the inspection robot.

[0093] 202. Based on environmental perception data, identify target risk areas in cable tunnels, including cable overheating areas, abnormal acoustic signature areas, and partial discharge areas.

[0094] 203. Based on the target risk area and robot status data, generate a global inspection path. The global inspection path can be optimized based on robot power information and equipment operating status information.

[0095] 204. The global inspection path is sent to the inspection robot for path planning and local obstacle avoidance. During the path planning and local obstacle avoidance process, the local path planner of the inspection robot performs path tracking and dynamically avoids obstacles based on real-time collected LiDAR data to perform the inspection task of the cable tunnel.

[0096] The environmental perception data refers to the data set reflecting the environmental status of the cable tunnel collected by multiple types of sensors. Specifically, it can be implemented using a combination of discharge detection sensors, infrared thermal imagers, and acoustic fingerprint sensors to capture partial discharge, temperature anomalies, and acoustic anomalies. The robot status data refers to the parameter set reflecting the operating status of the inspection robot itself. Specifically, it can be collected using an inertial measurement unit, battery management system, and motor encoder to monitor the robot's position, attitude, remaining energy, and the health of mechanical components in real time. Risk area identification refers to the process of determining potential hazardous areas through multi-source data fusion analysis. Specifically, it can be implemented using feature extraction algorithms and spatial correlation models to comprehensively assess the spatial coupling relationship between cable overheating, acoustic anomalies, and discharge phenomena. Path optimization refers to the calculation process of generating the optimal inspection trajectory based on dynamic constraints. Specifically, it can be implemented using heuristic search algorithms and multi-objective optimization models to balance the relationship between path length, risk exposure, and equipment operating limitations. The local path planner is a motion control module that processes environmental changes in real time. Specifically, it can be implemented using dynamic window algorithms and obstacle prediction models to perform local trajectory correction and emergency obstacle avoidance within the global path framework.

[0097] The core innovation of this embodiment lies in constructing a closed-loop collaborative mechanism for multimodal perception and dynamic path planning. By fusing multi-dimensional perception data from discharge, infrared, and acoustic signatures, and combining this with the robot's real-time state parameters, a two-way coupling of dynamic risk assessment and path optimization is formed. During the global path planning stage, power constraints and equipment state verification are introduced to ensure both mechanical feasibility and task reliability. Simultaneously, the real-time obstacle avoidance capability of the local path planner forms a hierarchical control architecture with a global framework and dynamic adjustments.

[0098] The working process and principle of this embodiment are as follows: The cable tunnel inspection robot first acquires multimodal environmental perception data and robot status data. Environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data, used to comprehensively perceive potential risks within the tunnel. Robot status data includes pose information, power information, and equipment operating status information, used to monitor the robot's capabilities in real time. Based on the acquired multi-source data, risk areas are identified. Overheated areas of the cable are identified by analyzing infrared thermal imaging data, abnormal acoustic signature areas are identified by analyzing acoustic signature data, and partial discharge areas are identified by analyzing discharge perception data. This multimodal data fusion analysis method can accurately capture the spatial distribution characteristics of complex risks. Based on the risk area identification, a global inspection path is generated by combining the robot status data. The path generation process considers robot power information and equipment operating status information to ensure the feasibility of the path. For example, a path of appropriate length is planned based on power information, and complex terrain that is difficult for the robot to traverse is avoided based on equipment status information. The generated global inspection path is issued to the inspection robot for execution. The robot's local path planner is responsible for path tracking, while simultaneously using real-time acquired LiDAR data for dynamic obstacle avoidance. This approach, combining global path planning with local obstacle avoidance, ensures the execution of the overall inspection task while flexibly responding to environmental changes. The solution achieves accurate identification of complex risks through multi-source data fusion, ensures path feasibility by considering robot state constraints, and improves the safety and efficiency of the inspection through the combination of global planning and local obstacle avoidance.

[0099] As a preferred embodiment, the method of this embodiment is implemented as follows: The inspection robot is equipped with multiple sensing devices, including a discharge sensor, an infrared thermal imager, a voiceprint collector, and a lidar. When a cable tunnel inspection task request is received, the robot activates each sensor to collect data. The discharge sensor collects electromagnetic pulse signals, the infrared thermal imager acquires temperature distribution images, and the voiceprint collector records acoustic signals. Simultaneously, the robot obtains its own pose, battery level, and the working status information of each actuator through internal sensors. The collected multimodal data is transmitted to an edge computing node for processing. The edge computing node first performs temperature gradient analysis on the infrared thermal image to identify areas with abnormally high temperatures as cable overheating zones. It then performs spectral analysis on the voiceprint data to detect abnormal frequency components and marks them as voiceprint anomaly areas. Finally, it extracts pulse features from the discharge signal to determine local discharge areas. After risk area identification, the edge computing node combines the robot's status data to generate a global inspection path. The path planning algorithm treats risk areas as high-cost areas, while also considering the robot's remaining battery power and equipment status. For example, when the battery is low, the shortest path is prioritized; when a robotic arm malfunction is detected, areas requiring robotic arm operation are avoided. The generated global path is sent to the inspection robot. The robot's local path planner receives the global path and processes the LiDAR data in real time during execution. When an obstacle is detected, the local planner generates a detour path, avoiding sudden obstacles while ensuring the overall task is executed.

[0100] The above methods enable accurate identification and dynamic response to complex risks within cable tunnels. Multimodal data fusion analysis improves the accuracy of risk identification and overcomes the limitations of single sensors. Path planning that considers robot states enhances task execution reliability and reduces task interruptions due to equipment limitations. The strategy combining global pathfinding and local obstacle avoidance strengthens the system's adaptability to environmental changes, improving inspection efficiency and safety. This solution addresses the challenge of handling dynamic complex risks in cable tunnel inspections, providing reliable technical support for the safe operation and maintenance of power systems.

[0101] In some of the solutions mentioned above in this application, a method for identifying risk areas based on multimodal data was proposed. However, when fusing temperature, acoustic signature and discharge characteristics, due to the dynamic coupling effect between different risk types, the single feature analysis method in related technologies cannot accurately capture the spatial correlation of composite risk areas, resulting in insufficient accuracy in identifying collaborative risk features.

[0102] To address this, there are various ways to identify target risk areas in cable tunnels based on environmental perception data. This application further proposes the following technical solution, taking edge computing nodes as the executing entity as an example, which can identify target risk areas in cable tunnels in the following way: Feature extraction is performed based on environmental perception data to obtain multiple environmental features, including temperature gradient features extracted from infrared thermal imaging data, frequency domain distribution features and energy attenuation features extracted from acoustic signature data, and pulse sequence features and discharge intensity features extracted from discharge perception data; feature fusion is performed based on multiple environmental features to obtain the risk coupling strength between different risk types, where the risk coupling strength is used to indicate the co-evolution trend of different risk types in the cable tunnel, as well as the possible spatial dependence and temporal correlation between different risk types; based on the risk coupling strength, composite risk areas with co-risk characteristics are identified; based on the spatial distribution and risk coupling strength of the composite risk areas in the cable tunnel, the target risk area is determined.

[0103] In this embodiment, temperature gradient features are extracted based on infrared thermal imaging data, frequency domain distribution features and energy attenuation features are extracted based on acoustic signature data, and pulse sequence features and discharge intensity features are extracted based on discharge sensing data. Multimodal feature fusion is performed on the temperature gradient features, frequency domain distribution features, energy attenuation features, pulse sequence features, and discharge intensity features to determine the risk coupling strength between different risk types. Based on the risk coupling strength, composite risk regions with synergistic risk features are identified. Based on the spatial distribution of composite risk regions in the cable tunnel and the corresponding risk coupling strength, target risk regions in the cable tunnel are determined. Specifically, the temperature gradient feature, extracted from infrared thermal imaging data, characterizes the gradient distribution pattern of temperature changes on the cable surface, for example, by calculating the gradient value using the temperature difference between adjacent pixels. The frequency domain distribution feature is extracted using the fast Fourier transform of acoustic signature data, which can capture energy concentration phenomena in specific frequency bands, for example, by decomposing the acoustic signal into multiple frequency bands and calculating the energy proportion of each band. The energy attenuation feature is calculated using an attenuation model of the acoustic propagation path, for example, by deriving the attenuation coefficient based on the relationship between the sound source distance and the sound pressure level at the receiving point. Pulse sequence features are extracted through time-domain waveform analysis of discharge sensing data, such as detecting pulse interval time and amplitude variation patterns. Discharge intensity features are calculated by integrating the energy of the discharge pulse per unit time.

[0104] In the multimodal feature fusion process, the thermoelectric coupling coefficient is used to calculate the correlation between temperature gradient and discharge intensity through the covariance matrix, for example, using the Pearson correlation coefficient to measure the degree of linear correlation between the two. The acoustic-electric coupling coefficient is analyzed using the mutual information method to analyze the nonlinear dependence between frequency domain distribution and pulse sequence. The determination of risk coupling strength involves the superposition of linear correlation strength and nonlinear dependence strength, for example, by normalizing the correlation coefficient and mutual information value and then weighting and summing them. The identification of composite risk areas can be achieved using spatial overlay analysis methods, for example, by rasterizing and overlaying spatial distribution maps of different risk features, and marking a composite area when multiple risk features exceed a threshold within the same raster cell. The spatial distribution of risk areas is achieved through geographic coordinate mapping, for example, by aligning the location information of composite areas with the three-dimensional model of the cable tunnel.

[0105] Specifically, temperature gradient features can be calculated by the temperature difference between adjacent pixels in an infrared image to identify localized overheating areas on the cable surface. For example, when the temperature gradient in a certain area exceeds 3℃ / m, it is identified as a potential overheating risk. Frequency domain distribution features are used to mark areas with an abnormal acoustic signature when the energy proportion of the acoustic signature signal in the 200-500Hz frequency band exceeds 60%. Energy attenuation features are used to identify areas near the sound source when the sound wave propagation attenuation coefficient is less than 0.5dB / m. Discharge intensity features can be used to identify partial discharge activity by detecting areas with pulse energy density exceeding 50pC / ms. Pulse sequence features are used to analyze regular discharge patterns with a standard deviation of less than 0.1ms between discharge pulse intervals. When fusing multimodal features, a dynamic weighting algorithm is used to allocate the weights of each feature. For example, when the thermoelectric coupling coefficient reaches 0.8, the weight of the temperature gradient feature is increased to 0.6. The calculation of risk coupling strength can comprehensively consider linear and nonlinear relationships. For example, the Pearson correlation coefficient of 0.7 and the mutual information value of 0.5 are weighted in a 6:4 ratio to obtain a comprehensive coupling strength of 0.62. The determination of a composite risk area requires that at least two risk features have a coupling strength exceeding 0.5 and a spatial overlap area greater than 0.1 m². The final risk area boundary is generated using a contour line extraction algorithm, for example, by extracting closed regions with risk values ​​exceeding a threshold of 0.75 from the risk value distribution map.

[0106] As a preferred embodiment, the method of this embodiment is implemented as follows: Temperature gradient features are extracted based on infrared thermal imaging data. For example, an edge detection algorithm can be used to process the thermal imaging image, extracting the edges of areas with drastic temperature changes and calculating the temperature gradient values ​​at the edges. Frequency domain distribution features and energy attenuation features are extracted based on acoustic signature data. Specifically, a fast Fourier transform can be performed on the acquired sound signal to obtain a spectrum and analyze the frequency component distribution. Simultaneously, the energy attenuation curve of the sound wave during propagation can be calculated to reflect the sound wave attenuation characteristics. Pulse sequence features and discharge intensity features are extracted based on discharge sensing data. For example, the number, interval, and amplitude distribution of discharge pulses per unit time can be statistically analyzed to form pulse sequence features. Discharge intensity features can be characterized by calculating the average amplitude or cumulative charge of the discharge pulses. The above features are fused using multimodal features, and the risk coupling strength between different risk types is determined. Deep learning models such as multimodal autoencoders can be used for feature fusion to learn the correlation between different modal features. Risk coupling strength can be quantified by calculating the mutual information or correlation coefficient between different features. Based on the risk coupling strength, composite risk regions with synergistic risk features are identified. For example, a coupling strength threshold can be set, and areas exceeding the threshold can be marked as composite risk areas. Based on the spatial distribution of composite risk areas in the cable tunnel and their corresponding risk coupling strength, risk areas in the cable tunnel can be determined. Specifically, a spatial clustering algorithm can be used to merge adjacent composite risk areas to form continuous risk area boundaries.

[0107] The above embodiments enable accurate identification of complex risk areas in cable tunnels. By employing multimodal feature fusion and risk coupling analysis, the dynamic coupling effect between different risk types is captured, improving the accuracy of identifying collaborative risk characteristics. The risk area determination method based on spatial distribution and coupling strength makes the boundaries of risk areas more precise, providing a reliable basis for risk avoidance in subsequent route planning.

[0108] In one optional embodiment, there are multiple ways to obtain the risk coupling strength between different risk types by feature fusion based on multiple environmental features. For example, the risk coupling strength between different risk types can be determined using a breast cancer approach: determining the thermoelectric coupling coefficient between temperature gradient features and discharge intensity features, and the acoustic-electric coupling coefficient between frequency domain distribution features and pulse sequence features. The thermoelectric coupling coefficient indicates the degree of linear influence of temperature gradient features on partial discharge intensity features, and the acoustic-electric coupling coefficient indicates the degree of linear influence of the frequency domain distribution features of the acoustic signature data on the pulse sequence features of the discharge sensing data. The information entropy weights corresponding to each of the multiple environmental features are then determined. The information entropy weight is used to quantify the uncertainty of the spatial distribution of the corresponding environmental features. Based on the thermoelectric coupling coefficient, the acoustic-electric coupling coefficient, and the information entropy weights corresponding to each of the multiple environmental features, the weighting coefficients corresponding to each of the multiple environmental features are determined. Based on the weighting coefficients corresponding to each of the multiple environmental features, the multiple environmental features are weighted and fused to obtain a comprehensive risk feature map. The comprehensive risk feature map is used to indicate at least the spatial correlation between feature values ​​of different risk types. The feature values ​​are used to quantify the risk intensity and trend of change in cable tunnels. Based on the comprehensive risk feature map, the risk coupling strength is obtained, which includes linear correlation strength and nonlinear dependence strength.

[0109] It should be noted that, in order to further address the problems of insufficient modeling of the correlation between multi-source heterogeneous features, static feature weight allocation leading to the failure to capture the dynamic interaction between thermoelectric and acoustic-electric features when calculating risk coupling strength, and the difficulty of traditional feature fusion methods in quantifying the impact of uncertainty in feature spatial distribution on risk assessment, this embodiment uses a multimodal feature correlation model to determine the thermoelectric coupling coefficient between temperature gradient features and discharge intensity features, and the acoustic-electric coupling coefficient between frequency domain distribution features and pulse sequence features. The multimodal feature correlation model is constructed based on the covariance matrix; based on information entropy theory, the temperature gradient features and frequency domain distribution features are determined respectively. The uncertainty of the spatial distribution of features, energy decay features, pulse sequence features, and discharge intensity features is measured to obtain the information entropy weight of each feature. Based on the thermoelectric coupling coefficient, acoustic-electric coupling coefficient, and the information entropy weight of each feature, a weighted fusion algorithm is used to generate dynamic adaptive weighting coefficients. Based on the dynamic adaptive weighting coefficients, the temperature gradient feature, frequency domain distribution feature, energy decay feature, pulse sequence feature, and discharge intensity feature are weighted and fused to obtain a comprehensive risk feature map. Based on the spatial correlation of feature values ​​of different risk types in the comprehensive risk feature map, the risk coupling strength is determined, which includes linear correlation strength and nonlinear dependence strength. The linear correlation strength is calculated based on the thermoelectric coupling coefficient and the acoustic-electric coupling coefficient, while the nonlinear dependence strength is used to quantify the strength of the nonlinear interaction of multiple environmental features in the spatial distribution.

[0110] The construction of the multimodal feature association model employs covariance matrix analysis to examine the covariance relationship between temperature gradient features and discharge intensity features. For example, the thermoelectric coupling coefficient is quantified by calculating the covariance value of the two features on a spatial grid, and the determinant of the covariance matrix can be used to characterize the correlation strength between cross-modal features. The acoustic-electric coupling coefficient between frequency domain distribution features and pulse sequence features is obtained by calculating the mutual information entropy of the two features in the time-frequency domain. A larger mutual information entropy value indicates a stronger dynamic interaction between acoustic and electrical features. The calculation of information entropy weights is based on the probability distribution of each feature on a spatial grid. For example, the information entropy weight of the temperature gradient feature is determined by calculating the ratio of its spatial distribution standard deviation to the information entropy; a larger standard deviation results in a higher information entropy weight. The dynamic adaptive weighting coefficients are generated by performing a normalized product operation on the thermoelectric coupling coefficient, the acoustic-electric coupling coefficient, and the information entropy weights. For example, the softmax function is used to normalize the product result, ensuring that the sum of the weighting coefficients of each feature is 1. The generation of the comprehensive risk feature map employs a spatial overlay algorithm. For example, each feature map is multiplied pixel-by-pixel with its corresponding dynamically adaptive weighting coefficient and then overlaid to form a risk distribution heatmap in a unified spatial coordinate system. Determining the risk coupling strength involves calculating the linear correlation strength using the Pearson correlation coefficient and measuring the nonlinear dependence strength using the maximum information coefficient. For instance, both strength indices are calculated simultaneously within a local window of the comprehensive risk feature map, forming a dual risk coupling measure.

[0111] Specifically, the dynamic correlation between temperature gradient features and discharge intensity features is modeled using a covariance matrix. The covariance matrix is ​​constructed based on synchronous sampling data of the two features on a spatial grid. For example, temperature gradient and discharge intensity values ​​are simultaneously collected within each 10cm × 10cm grid cell, and the covariance of the two feature sequences is calculated as the thermoelectric coupling coefficient. The acoustic-electric coupling coefficient between frequency domain distribution features and pulse sequence features is calculated using mutual information entropy. For example, the frequency domain distribution of the acoustic signature signal is divided into multiple sub-bands, and the joint probability distribution of the energy of each sub-band and the frequency of the pulse sequence is statistically analyzed to calculate the mutual information entropy value. The information entropy weight of each feature is determined by calculating the information entropy of its spatial distribution. For example, the information entropy weight of the temperature gradient feature is determined by the ratio of the entropy value of its spatial distribution histogram to the maximum possible entropy value. A higher ratio indicates greater spatial uncertainty for the feature, requiring a higher weight. In the generation of dynamic adaptive weighted coefficients, the thermoelectric coupling coefficient and the acoustic-electric coupling coefficient reflect the dynamic correlation strength of cross-modal features, respectively, while the information entropy weight reflects the spatial uncertainty of single-modal features. These three are integrated using a weighted fusion algorithm, such as the weighted geometric average method, to fuse them into a dynamic coefficient. During the generation of the comprehensive risk feature map, each feature map is weighted and superimposed according to the dynamic coefficient in the spatial dimension. For example, the temperature gradient feature map is multiplied by its dynamic coefficient, and then added pixel-by-pixel to the weighted result of the discharge intensity feature map to form a fused risk heatmap. The calculation of risk coupling strength employs both linear and nonlinear metrics. For instance, within a local 3×3 window of the comprehensive risk feature map, the Pearson correlation coefficient of each risk type feature value is first calculated as the linear correlation strength, and then the maximum information coefficient is calculated using the kernel density estimation method as the nonlinear dependence strength. Finally, the two strength indicators are fused into a comprehensive risk coupling strength value according to a preset ratio.

[0112] As a preferred embodiment, the specific implementation of the method in this embodiment is as follows: The multimodal feature association model is constructed based on the covariance matrix. First, the covariance matrix between the temperature gradient feature and the discharge intensity feature is calculated, and the off-diagonal elements in the matrix are extracted as thermoelectric coupling coefficients. Similarly, the covariance matrix between the frequency domain distribution feature and the pulse sequence feature is calculated, and the off-diagonal elements are extracted as acoustic-electric coupling coefficients. Further, the uncertainty measure of each feature in spatial distribution is determined based on information entropy theory. For each feature, its spatial distribution is discretized into several intervals, the probability distribution of each interval is calculated, and then the information entropy is calculated. Thus, the information entropy weights of the temperature gradient feature, frequency domain distribution feature, energy decay feature, pulse sequence feature, and discharge intensity feature are obtained. The thermoelectric coupling coefficient, acoustic-electric coupling coefficient, and the information entropy weights of each feature are used to generate dynamic adaptive weighting coefficients through a weighted fusion algorithm. Specifically, the coupling coefficients and information entropy weights are normalized and then multiplied to obtain preliminary weights. Then, the preliminary weights are exponentially smoothed to reflect the dynamic changes in feature importance. The dynamic adaptive weighting coefficients are used to weight and fuse temperature gradient features, frequency domain distribution features, energy decay features, pulse sequence features, and discharge intensity features. The fusion process employs a weighted summation method to obtain a comprehensive risk feature map. For example, the spatial correlation between feature values ​​of different risk types in the comprehensive risk feature map is used to determine the risk coupling strength. Risk coupling strength includes linear correlation strength and nonlinear dependence strength. Linear correlation strength is obtained by calculating the Pearson correlation coefficient, while nonlinear dependence strength is calculated using the mutual information entropy method.

[0113] Through the above embodiments, dynamic fusion of multimodal features and accurate quantification of risk coupling strength can be achieved. The multimodal feature association model captures the interaction relationship between thermoelectric and acoustic-electric features, overcoming the problem of insufficient cross-modal feature association modeling in related technologies. The feature weight allocation method based on information entropy enables the feature importance to be dynamically adjusted with environmental changes, avoiding the lack of adaptability caused by fixed weights. The introduction of dynamic adaptive weighting coefficients enables the feature fusion process to respond in real time to changes in the relative importance of features under different operating conditions. The construction of the comprehensive risk feature map realizes a unified representation of multi-source heterogeneous features in the spatial dimension, laying the foundation for subsequent risk coupling analysis. Finally, through dual measurements of linear correlation strength and nonlinear dependence strength, the interaction mechanism between risk types is comprehensively characterized, providing quantitative support for the accurate identification of composite risk areas. The comprehensive application of this series of technical means improves the accuracy and reliability of composite risk identification in cable tunnels.

[0114] In one optional embodiment, there are multiple ways to determine the target risk area based on the spatial distribution and risk coupling strength of the composite risk areas in the cable tunnel. For example, when there are multiple composite risk areas, the target risk area can be obtained as follows: A weighted summation operation is performed based on the linear correlation strength and nonlinear dependence strength corresponding to each of the multiple composite risk areas, as well as the weight values ​​corresponding to each of the linear correlation strength and nonlinear dependence strength, to obtain the risk value corresponding to each of the multiple composite risk areas; a risk value distribution map is generated using a spatial interpolation method based on the spatial distribution of the multiple composite risk areas in the cable tunnel and the risk values ​​corresponding to each of the multiple composite risk areas, wherein the risk value distribution map is used to reflect the spatial variation of risk values ​​in the cable tunnel; contour lines are extracted from the risk value distribution map based on preset multi-level risk thresholds to obtain the boundaries of areas with different risk levels, wherein the multi-level risk thresholds are set according to the preset safety standards of the cable tunnel; and the target risk area is determined based on the area boundaries.

[0115] This embodiment proposes a method to determine the risk value of each composite risk region based on the risk coupling strength. The risk value is obtained by weighted summation of linear correlation strength and nonlinear dependence strength, with the weights dynamically adjusted based on historical risk event data. A continuous risk value distribution map is generated based on the spatial distribution of composite risk regions using spatial interpolation. Contour lines are extracted from the risk value distribution map, and regional boundaries of different risk levels are extracted based on preset multi-level risk thresholds. The target risk region in the cable tunnel is determined based on the extracted regional boundaries.

[0116] The dynamic adjustment of weights in the weighted summation can be achieved through exponential smoothing, Bayesian update algorithms, or neural network models. For example, a sliding time window can be used to statistically analyze the frequency of linear and nonlinear relationships in historical events, and the frequency ratio can be used as the basis for weight adjustment. Spatial interpolation methods can include Kriging interpolation or inverse distance weighted interpolation. Kriging interpolation models spatial autocorrelation using a semi-variogram, while inverse distance weighted interpolation generates a continuous surface based on the distance decay principle. Contour extraction can employ the Marching Squares algorithm, which generates closed contour lines by traversing grid cells. Multi-level risk thresholds are divided into four levels according to cable tunnel safety standards. For example, the risk value range [0.2, 0.4) is defined as a low-risk zone, [0.4, 0.6) as a medium-risk zone, [0.6, 0.8) as a high-risk zone, and [0.8, 1.0] as an emergency risk zone.

[0117] Specifically, dynamic quantification of risk values ​​is achieved by fusing linear correlation strength and nonlinear dependence strength. Linear correlation strength reflects the spatial covariance between temperature gradient and discharge intensity, while nonlinear dependence strength characterizes the complex interaction between acoustic signature frequency domain features and pulse sequences. A dynamic weight adjustment mechanism allows risk value calculation to adapt to risk evolution patterns over different time periods. For example, when nonlinear events account for more than 60% of historical data, the weight coefficient of nonlinear dependence strength is increased to 0.7. During spatial interpolation, discrete composite risk region coordinates are used to generate a continuous risk surface with 1-meter resolution through Kriging interpolation, eliminating the stepped boundaries caused by traditional grid division. For contour line extraction, the Marching Squares algorithm traverses the risk value distribution map with a step size of 0.1, generating closed contour lines, which are then spatially overlaid with preset four-level risk thresholds to ultimately output graded risk region boundaries that conform to industry standards.

[0118] As a preferred embodiment, the method of this embodiment is implemented as follows: Based on the risk coupling strength, the risk value of each composite risk region is determined. The risk value is obtained by weighted summation of linear correlation strength and nonlinear dependence strength. The weights of the weighted summation are dynamically adjusted based on historical risk event data. For example, an exponentially weighted moving average method can be used to process historical data, giving higher weight to recent events, thus making the risk value calculation more closely reflect the current situation. Further, based on the spatial distribution of the composite risk regions, a continuous risk value distribution map is generated using the Kriging spatial interpolation method. This distribution map reflects the spatial variation of risk values ​​in the cable tunnel. The Kriging interpolation method considers spatial autocorrelation and can more accurately estimate the risk value of unknown points. Therefore, contour lines are extracted from the risk value distribution map. Region boundaries of different risk levels are extracted based on preset multi-level risk thresholds. The multi-level risk thresholds are set according to cable tunnel safety standards. Specifically, three levels of risk thresholds—low, medium, and high—can be set, corresponding to minor, moderate, and severe risk levels, respectively. Finally, based on the extracted region boundaries, the target risk region in the cable tunnel is determined. These target risk areas can be visualized in the 3D model of the cable tunnel, making it easier for managers to intuitively understand the risk distribution.

[0119] Through the above technical solutions, this embodiment achieves dynamic quantitative assessment of complex risk areas and models the continuity of risk spatial distribution. The dynamic calculation of risk values ​​integrates equipment operating patterns and historical experience data, avoiding assessment biases caused by static weights. The application of spatial interpolation methods solves the boundary abruptness problem caused by discretization modeling in related technologies, realistically reflecting the spatial gradual change characteristics of risk intensity in cable tunnels. Contour line extraction technology, combined with the setting of multi-level risk thresholds, ensures that the risk level classification conforms to industry safety standards and achieves accurate extraction of risk boundaries. Finally, risk areas are determined based on scientifically defined regional boundaries, providing accurate spatial constraints for subsequent path planning, thereby improving the safety and efficiency of cable tunnel inspection.

[0120] In one optional embodiment, there are multiple ways to generate a global inspection path based on target risk areas and robot state data. For example, when there are multiple target risk areas, the global inspection path can be generated as follows: predict the spatial range change trends of multiple target risk areas within the inspection cycle; construct a global path optimization objective function with the goal of minimizing the total path risk value and maximizing inspection efficiency, wherein the total path risk value is jointly determined by the risk weight of the neighborhood of the corresponding risk area traversed by the inspection robot and the exposure time of the inspection robot being affected by the corresponding spatial range change trend due to the path being close to the neighborhood of the corresponding risk area, and the exposure time refers to the duration of the potential risk or risk diffusion process in the neighborhood of the corresponding risk area; determine the constraints of path planning based on robot state data; and solve the global path optimization objective function using the A* algorithm based on the constraints of path planning to obtain the global inspection path.

[0121] Optionally, in this embodiment, the impact of path planning on the identified target risk areas is quantified. Dynamic risk weights are assigned to each target risk area based on risk type and risk level. The spatial range change trend of each target risk area within the inspection cycle is predicted based on a risk propagation model. A global path optimization objective function is constructed, aiming to minimize the total path risk value and maximize inspection efficiency. The total path risk value is jointly determined by the risk weights of the neighborhoods of each target risk area traversed by the path and the exposure time of the inspection robot due to the spatial range change trend caused by path proximity. Robot power information and equipment operating status information from the robot status data are used as constraints for path planning. Based on the constraints of path planning, the A* algorithm is used to solve the global path optimization objective function. When determining the path cost, a dynamic risk cost based on dynamic risk weights and exposure time is introduced, and a feasibility check of the equipment operating status is added to obtain the global inspection path.

[0122] The path planning impact quantification uses a risk propagation model to predict the spatial dynamic changes of risk areas. For example, a risk propagation predictor trained with machine learning outputs time slice prediction boundaries within the inspection cycle. Dynamic risk weight allocation is based on multiplying the base weight of the risk type with dynamic adjustment factors for the range expansion rate and intensity change trend. The base weight for cable overheating areas can be set to 0.6, partial discharge areas to 0.8, and acoustic anomaly areas to 0.5. The total path risk value in the global path optimization objective function is calculated by multiplying and accumulating the dynamic risk weights and exposure time. The exposure time is calculated based on the relative relationship between the robot's movement speed and the risk area expansion rate. The comprehensive cost function of the A* algorithm consists of distance cost, risk cost, and time cost. The risk cost calculation incorporates a time margin check of the risk propagation prediction results. For example, when the difference between the predicted boundary expansion time and the robot's passage time window is less than 30 seconds, path replanning is triggered. Feasibility verification of equipment operation includes power accessibility calculation and terrain traversal capability verification. The power consumption model can be established based on path length and terrain complexity. For example, each meter of straight path consumes 0.1% of power, while consumption increases to 0.3% in complex terrain areas.

[0123] Specifically, in the dynamic risk quantification phase, the risk propagation model generates predicted risk area boundaries for different time slices within the inspection cycle by inputting historical evolution data and real-time monitoring data. For example, for a partial discharge area, the model predicts that its range will expand by 15% within 30 minutes, and calculates the exposure time for the robot when passing through the vicinity of this area. During the dynamic risk weight allocation process, risk areas with a range expansion rate exceeding 5% / minute or an intensity enhancement trend exceeding 10% will receive a dynamic adjustment factor of 1.2 times, allowing rapidly spreading high-risk areas to receive a higher avoidance priority in path planning.

[0124] When constructing the objective function for path optimization, the calculation of the total path risk value considers not only static risk weights but also the exposure time factor. For example, if a path segment passes through two target risk areas with dynamic risk weights of 0.8 and 1.2, and exposure times of 20 seconds and 15 seconds respectively, then the risk value of this segment is (0.8 × 20) + (1.2 × 15) = 34 risk units. In the improved A* algorithm, the operating status of equipment is checked synchronously during node expansion. When the remaining power is less than 110% of the power required to reach the node, the node is marked as infeasible. During the dynamic path adjustment phase, curvature optimization is performed on path segments with time margins below the safety threshold. For example, the turning radius is increased from 0.5 meters to 0.8 meters to improve traffic speed and reduce transit time by 20%.

[0125] The global inspection path planning scheme in this embodiment can respond to the dynamic evolution of risk areas in real time, proactively avoid risk spread areas during the path generation stage, and accurately balance risk avoidance requirements with inspection efficiency requirements. The introduction of equipment state constraints ensures that the planned path is within the robot's actual operating capabilities, avoiding path execution failures due to insufficient power or mechanical performance limitations. The dynamic risk cost mechanism extends the traditional single-distance optimization of path search to multi-dimensional comprehensive optimization, achieving a dual improvement in safety and efficiency in complex dynamic environments.

[0126] As a preferred embodiment, the specific implementation of this method is as follows: The impact of path planning on the identified target risk areas is quantified. Dynamic risk weights are assigned to each target risk area based on risk type and risk level. The spatial range change trend of each target risk area within the inspection cycle is predicted based on a risk propagation model. A global path optimization objective function is constructed, aiming to minimize the total path risk value and maximize inspection efficiency. The total path risk value is jointly determined by the risk weights of the neighborhoods of each target risk area traversed by the path and the exposure time of the inspection robot affected by the spatial range change trend due to path proximity. Robot battery information and equipment operating status information are used as constraints for path planning. The A* algorithm is used to solve the global path optimization objective function. When determining the path cost, a dynamic risk cost based on dynamic risk weights and exposure time is introduced, and a feasibility check of the equipment operating status is added to obtain the global inspection path.

[0127] Specifically, the spatial extent of each risk area during the inspection cycle is first predicted based on a risk propagation model. For example, for cable overheating areas, a thermal diffusion model can be used to predict the evolution of the temperature field; for partial discharge areas, an electromagnetic field simulation model can be used to predict the diffusion of discharge intensity. Further, initial weights are assigned to each target risk area according to the risk type and risk level. Cable overheating areas may be assigned higher initial weights because they can directly damage robot hardware. Subsequently, the initial weights are dynamically adjusted based on the predicted spatial extent change trend. For example, the weight of a risk area predicted to expand rapidly may be increased. Thus, a global path optimization objective function is constructed. This function comprehensively considers path length, risk exposure level, and inspection efficiency. The total path risk value can be expressed as the integral of the risk weight at each point on the path and the robot's dwell time at that point. Inspection efficiency can be quantified by path length and estimated completion time. As a preferred implementation, the A* algorithm is used to solve this optimization problem. During node expansion, in addition to the traditional distance cost, a dynamic risk cost is introduced. This cost is determined by the risk weight of the current node and the estimated robot exposure time. Simultaneously, each candidate node undergoes feasibility verification to ensure it meets robot power and equipment performance constraints. The resulting global inspection path avoids high-risk areas while maintaining the integrity and efficiency of the inspection task. For example, for risk areas predicted to expand, the path may detour in advance or increase the passage speed to reduce the robot's risk exposure.

[0128] Through the above technical solutions, the method in this embodiment can achieve a balance between minimizing total risk and inspection efficiency in a dynamic risk environment. By quantitatively assessing and dynamically predicting risk areas, path planning can more accurately reflect real-time risk conditions. Introducing risk costs based on exposure time enables the path to avoid potential threats caused by risk spread. Simultaneously, by using robot state as a constraint, the generated path becomes feasible. This comprehensive optimization method improves the safety and efficiency of inspection tasks, providing strong support for the intelligent operation and maintenance of cable tunnels.

[0129] In one optional embodiment, there are several ways to predict the spatial range change trends of multiple target risk areas within an inspection cycle. For example, the spatial range change trends of multiple target risk areas within an inspection cycle can be predicted as follows: Historical evolution data and real-time monitoring data corresponding to each of the multiple target risk areas are input into a pre-set risk propagation predictor to obtain time-series risk intensity data. The risk propagation predictor is trained using machine learning methods to simulate the diffusion behavior of different risk types in a cable tunnel environment. The time-series risk intensity data indicates the risk intensity distribution of each of the multiple target risk areas at different sampling times. Based on the time-series risk intensity data, the range expansion rate and risk intensity change trend corresponding to each of the multiple target risk areas are determined. According to the risk type, risk level, range expansion rate, and risk intensity change trend corresponding to each of the multiple target risk areas, the dynamic risk weight corresponding to each of the multiple target risk areas is determined. Based on the risk area prediction boundary and dynamic risk weight corresponding to each of the multiple target risk areas, the spatial range change trend is obtained.

[0130] Optionally, this embodiment further proposes to quantify the impact of path planning on identified target risk areas, assign dynamic risk weights to each target risk area according to risk type and risk level, and predict the spatial range change trend of each target risk area within the inspection cycle based on a risk propagation model. Historical evolution data and real-time monitoring data may include temperature change time series, discharge frequency statistics, and acoustic signature energy fluctuation records, which are processed by a pre-built risk propagation predictor. The risk propagation predictor can be constructed using a long short-term memory network or a spatiotemporal graph convolutional neural network. Its input layer receives multidimensional time series data, and its output layer generates a risk diffusion probability distribution map. The time slice can be set to 1 / 10 to 1 / 5 of the inspection cycle length, for example, using 15-minute interval slices in a 2-hour inspection cycle. The range expansion rate can be calculated by the ratio of the difference in boundary area between adjacent time slices to the time interval, and the intensity change trend can be determined based on whether the linear regression slope exceeds a preset threshold. The weight adjustment factor in the weighted decision algorithm can be set to 1.2 to 1.5 times the base weight, with the specific value determined according to the safety specifications corresponding to the risk type. The aggregation of prediction boundaries can be achieved using a grid overlay method, which projects the prediction boundaries of different target risk areas onto the same spatial coordinate system to form a dynamic heat map.

[0131] Specifically, historical evolution data and real-time monitoring data are input into a risk propagation predictor trained by a machine learning model, which outputs the risk intensity distribution at various future time points. For example, for an overheated cable area, the predictor can predict the heat diffusion range for the next 30 minutes based on the temperature change curve of the past 24 hours. By extracting the risk area boundaries from continuous time slices, the area expansion rate of 0.5 square meters per minute can be calculated, and the centroid is detected moving southeast at a speed of 2 centimeters per second. In the calculation of dynamic risk weights, the cable overheated area's base weight of 0.6 is multiplied by an adjustment factor of 1.3 because the slope of the intensity change trend exceeds the threshold, ultimately obtaining a dynamic risk weight of 0.78. After coordinate transformation, the predicted boundaries of all risk areas form a four-dimensional spatial model containing time dimension information. This model is used to calculate the dynamic risk exposure time in path planning. The resulting path planning scheme allows the robot to pass through the danger zone 15 seconds before the risk area reaches the predicted boundary, reducing the impact of dynamic risks.

[0132] As a preferred embodiment, the specific implementation of this method is as follows: First, the historical evolution data and real-time monitoring data of each risk area are input into a pre-set risk propagation predictor. This predictor is trained using machine learning methods to simulate the diffusion behavior of different risk types in a cable tunnel environment. For example, a Long Short-Term Memory (LSTM) network can be used as the core algorithm of the predictor, with inputs including environmental parameters such as temperature, humidity, and airflow speed, as well as historical risk event data, and the output being a spatiotemporal distribution prediction of risk intensity. Next, based on the output of the risk propagation predictor, the predicted boundaries of risk areas under different time slices within the inspection cycle are generated. Specifically, a risk distribution raster map sequence for the next 2 hours can be generated according to a time slice of 10 minutes. By extracting the boundaries of the risk raster map at each moment, the predicted boundaries of the risk areas for the corresponding time slice are obtained. Further, by comparing the predicted boundaries of adjacent time slices, the area change rate and centroid movement vector of each risk area are calculated to determine the range expansion rate and main expansion direction. Then, based on the risk type, current risk level, range expansion rate, and intensity change trend, a weighted decision algorithm is used to determine the dynamic risk weights. For example, base weights of 0.4, 0.35, and 0.25 can be set for cable overheating zones, partial discharge zones, and acoustic anomaly zones, respectively. If the expansion rate of a target risk zone exceeds a preset threshold or its intensity shows an increasing trend, a positive adjustment factor of 1.2 is applied. The final dynamic risk weight is obtained by multiplying the base weight by the adjustment factor and then normalizing. Finally, the predicted boundaries and dynamic risk weights of all target risk zones are aggregated to form a spatial range variation trend. This can be achieved by generating a spatiotemporal cube, where the x and y axes represent spatial coordinates, the z-axis represents time, and the value of each point in the cube represents the risk intensity at that spatiotemporal location.

[0133] The above methods enable dynamic prediction and quantitative assessment of risk areas in cable tunnels. By introducing a machine learning-based risk propagation predictor, the diffusion behavior of different risk types in complex tunnel environments can be simulated more accurately. By generating risk area prediction boundaries in time slices, the dynamic expansion rate and centroid movement direction of the risk range can be quantified, providing a spatiotemporal basis for risk evolution in path planning. A dynamically adjusted risk weight mechanism assigns higher weights to rapidly expanding or intensifying areas, allowing path planning to prioritize avoiding dynamically deteriorating risks. The final aggregated spatial range change trend provides dynamic input parameters for global path optimization, enabling path planning to predict risk evolution and proactively avoid risks, thereby improving the safety and efficiency of inspection robots in dynamic risk environments.

[0134] In one optional embodiment, there are several ways to determine the range expansion rate and risk intensity change trend of each of the multiple target risk areas based on time-series risk intensity data. For example, the range expansion rate and risk intensity change trend of each of the multiple target risk areas can be determined as follows: Based on time-series risk intensity data, generate risk distribution raster maps corresponding to multiple sampling times within the inspection cycle, wherein the multiple sampling times are sampled at preset time intervals within the inspection cycle; extract the boundaries of the risk distribution raster maps corresponding to each of the multiple sampling times to obtain the predicted boundaries of the risk areas corresponding to each of the multiple sampling times; and determine the area change rate and centroid movement of each of the multiple target risk areas by comparing the predicted boundaries of the risk areas at adjacent sampling times. The vectors are used to indicate the overall direction and speed of the risk area's movement in space. Based on the area change rate and the time derivative of the centroid movement vector corresponding to each of the multiple target risk areas, the range expansion rate and main expansion direction of each target risk area are determined. The range expansion rate is used to quantify the expansion speed of the risk area boundary over time, and the main expansion direction is used to indicate the main direction of the risk area boundary expansion. Based on the intensity characteristic values ​​of the core areas corresponding to each of the multiple target risk areas and the time-series risk intensity data, the intensity change trend of each target risk area is obtained. The core area refers to the area where the risk intensity is higher than the preset risk threshold or the degree of change is greater than the preset degree. The intensity characteristic value is used to indicate the characteristic risk intensity inside the risk area.

[0135] Optionally, this embodiment proposes to generate a series of risk distribution raster maps based on the time-series risk intensity data output by the risk propagation predictor, and to extract the boundaries of the risk raster map at each time point to obtain the risk area prediction boundary for the corresponding time slice; by comparing the risk area prediction boundaries of adjacent time slices, the area change rate and centroid movement vector of each risk area are determined; the range expansion rate and main expansion direction are determined based on the time derivative of the area change rate and centroid movement vector; the intensity feature value of the core area of ​​each risk area is extracted, and an intensity change time series sequence is established based on the intensity feature value and the time-series risk intensity data; by performing trend fitting analysis on the intensity change time series sequence, the intensity change trend is obtained; wherein, the intensity feature value corresponds to the maximum temperature, discharge frequency or sound energy peak value according to the risk type.

[0136] The preset time interval can be set from 5 to 30 minutes, for example, 10 minutes as the sampling period for the discretized time dimension. Boundary extraction can use edge detection algorithms, such as the Sobel operator or the Canny operator, to transform continuous pixel regions with risk intensity exceeding the threshold into geometric boundaries. The area change rate is calculated by the change rate of the number of pixels within the boundaries of adjacent time slices, for example, by converting the number of pixels into actual area using the raster pixel resolution. The centroid movement vector is obtained by calculating the difference between the coordinates of the geometric center of the risk region in adjacent time slices, for example, by using a weighted centroid algorithm with the risk intensity value as the pixel weight. Time differentiation processing can use the finite difference method, for example, the first-order forward difference is used to calculate the time derivative of the area change rate, eliminating the interference of random noise on the instantaneous expansion rate. The extraction of intensity feature values ​​requires selecting different physical quantities according to the risk type, for example, the maximum temperature corresponds to the cable overheating zone, the discharge frequency corresponds to the partial discharge zone, and the peak sound energy corresponds to the acoustic anomaly zone. Trend fitting analysis can use linear regression or polynomial fitting, for example, by using the least squares method to establish the trend equation of the intensity change time series and identify the intensity enhancement or decay mode.

[0137] Specifically, the time-series risk intensity data is discretized into raster maps across multiple time slices, with each raster cell storing the quantified risk intensity value for its corresponding spatial location. The boundary extraction process transforms the continuous risk intensity distribution into closed boundaries with well-defined geometric shapes, providing a spatial benchmark for dynamic evolution analysis. The area change rate of adjacent time slice boundaries reflects the rate of expansion or contraction of the risk region, while the centroid movement vector characterizes the overall offset direction of the spatial location. By performing time differentiation on the area change rate and centroid movement vector, the instantaneous change characteristics of the expansion rate can be accurately captured, such as anomalous states of sudden acceleration or deceleration. The extraction of core area intensity feature values ​​focuses on key indicators of the risk's physical mechanism; for example, the maximum temperature in the cable overheating zone directly reflects the thermal runaway risk level. The time-series intensity change sequence separates the long-term trend from the short-term fluctuation component through trend fitting; for example, cubic polynomial fitting is used to identify nonlinear enhancement trends. Different risk types employ differentiated feature values ​​to avoid distortion of composite risks by a single indicator; for example, discharge frequency is more suitable for characterizing the activity level of partial discharge, while peak acoustic energy reflects the severity of acoustic anomalies. Through the synergistic effect of the above steps, a refined model of the spatiotemporal evolution characteristics of risk areas is achieved, providing high-precision risk propagation prediction data support for dynamic path planning.

[0138] As a preferred embodiment, the method of this embodiment is implemented as follows: Based on the time-series risk intensity data output by the risk propagation predictor, a series of risk distribution raster maps are generated according to a preset time interval. For example, a risk distribution raster map can be generated every 5 minutes, with each raster cell corresponding to a spatial location in the cable tunnel. Boundary extraction is performed on the risk raster map at each time point. The Canny edge detection algorithm can be used to obtain the predicted boundary of the risk area for the corresponding time slice. Further, by comparing the predicted boundaries of the risk areas of adjacent time slices, the area change rate and centroid movement vector of each risk area are determined. Specifically, the percentage change in the area of ​​the risk area between two adjacent time slices can be calculated to obtain the area change rate. At the same time, the centroid movement vector is obtained by dividing the difference in centroid coordinates by the time interval. Based on the time derivative of the area change rate and the centroid movement vector, the range expansion rate and the main expansion direction can be determined. Thus, the intensity feature value of the core area of ​​each risk area is extracted. Among them, the intensity feature value corresponds to the maximum temperature, discharge frequency, or sound energy peak value according to the risk type. For example, for cable overheating zones, the highest temperature value in the core area can be extracted; for partial discharge zones, the number of discharge pulses per unit time can be extracted; and for areas with abnormal acoustic signatures, the peak value of acoustic energy can be extracted. A time-series sequence of intensity changes is established based on intensity characteristic values ​​and temporal risk intensity data. Finally, by performing trend fitting analysis on the intensity change time-series sequence, the intensity change trend is obtained. Methods such as multinomial regression or exponential smoothing can be used for trend fitting to predict the future trend of risk intensity changes.

[0139] This embodiment achieves refined modeling of the spatiotemporal evolution characteristics of risk areas. By generating a series of risk distribution raster maps and extracting boundaries, the spatial distribution of risk at each moment can be accurately characterized. Comparing the predicted boundaries of adjacent time slices reveals the dynamic evolution of risk areas from two dimensions: the rate and direction of spatial expansion. Using time differentiation to process the area change rate and centroid movement vector accurately extracts the instantaneous change characteristics of the expansion rate. Core area intensity feature values ​​are extracted for different risk types, and trend fitting is performed to distinguish between gradual and abrupt changes in risk intensity. This multi-dimensional spatiotemporal data analysis and feature extraction method solves the problem of insufficient quantification accuracy in the dynamic evolution of risk areas, providing high-precision risk propagation prediction data support for dynamic path planning.

[0140] In one optional embodiment, there are several ways to determine the dynamic risk weights of multiple target risk areas based on their respective risk types, risk levels, range expansion rates, and risk intensity trends. For example, the dynamic risk weights of multiple target risk areas can be determined as follows: Based on the risk types and risk levels of each target risk area, a weight allocation strategy is determined; based on the corresponding weight allocation strategy, a basic weight is determined for each target risk area, wherein different basic weights are assigned to designated risk areas according to the potential hazard level of the corresponding target risk area to the inspection robot, and the designated risk areas are target risk areas with corresponding risk types of cable overheating, partial discharge, and abnormal acoustic signatures; based on the range expansion rate and risk intensity trends of each target risk area, a dynamic adjustment factor is determined for each target risk area; the basic weights of each target risk area are multiplied by the corresponding dynamic adjustment factors to obtain the preliminary risk weights of each target risk area; the preliminary risk weights of each target risk area are normalized to obtain the dynamic risk weights of each target risk area. Optionally, this embodiment proposes to establish a weight allocation strategy based on risk type and risk level, and assign basic weights to risk areas of different risk types. Cable overheating areas, partial discharge areas, and abnormal acoustic signature areas are assigned different basic weight values ​​according to their potential hazard to the inspection robot. Dynamic adjustment factors for each risk area are determined based on the range expansion rate and intensity change trend. Risk areas with expansion rates exceeding the threshold or with increasing intensity trends are given positive adjustment factors. The basic weights are multiplied by the dynamic adjustment factors to obtain preliminary risk weights. The preliminary risk weights are normalized to obtain the dynamic risk weights for each target risk area.

[0141] In the weighting strategy, the severity of risk types can be quantified using historical fault data. For example, the base weight for cable overheating zones, which have a higher probability of melting due to high temperatures, can be set to 0.5; the risk of partial discharge zones due to insulation degradation can be set to 0.3; and the risk of abnormal acoustic signature zones due to mechanical damage can be set to 0.2. The dynamic adjustment factor is calculated using real-time monitoring data. When the risk area expansion rate exceeds 5% per minute or the intensity characteristic value increases by more than 20% per hour, a positive adjustment factor is triggered, with the adjustment range set to 1.2 to 1.5 times the coefficient. The initial risk weight is calculated using matrix multiplication. After multiplying the base weight vector of each risk area by the dynamic adjustment factor matrix, the weight values ​​are compressed to the 0-1 range using maximum value normalization, making the weights of different risk areas comparable.

[0142] Specifically, during path planning, a basic weight is first assigned based on the inherent attributes of the risk type. For example, the cable overheating zone is given the highest basic weight to reflect the severity of its high-temperature hazard. Then, by monitoring changes in the physical parameters of the risk area in real time, when the discharge frequency of a partial discharge area increases by 25% per hour, a 1.3-fold positive adjustment factor is automatically generated, increasing the weight of that area from the basic value of 0.3 to 0.39. After multiplying the initial weights of all risk areas, a sum-normalization method is used to constrain the global weight sum to 1, preventing a single high-risk area from excessively influencing path planning. This dual-weight adjustment mechanism retains the inherent threat level of the risk type while capturing real-time risk evolution through dynamic factors. This allows the path planning system to respond promptly to rapidly changing risk conditions within the cable tunnel, ensuring that the inspection robot always selects the optimal path with the shortest risk exposure time and the lowest threat level.

[0143] As a preferred embodiment, the specific implementation of this application's solution is as follows: During the cable tunnel inspection path planning process, the determination of dynamic risk weights is achieved through multi-stage calculations. First, a weight allocation strategy is established, where the base weight for the cable overheating zone is set to 0.7, the base weight for the partial discharge zone is 0.5, and the base weight for the acoustic anomaly zone is 0.3. These values ​​are determined based on laboratory test data. For partial discharge zones where the detection range expansion rate exceeds 0.5 m / s, the dynamic adjustment factor is set to 1.3; when the temperature gradient change rate of the cable overheating zone exceeds 10℃ / min for three consecutive sampling periods, the corresponding dynamic adjustment factor is increased to 1.5. The initial risk weight is obtained by multiplying the base weight by the adjustment factor; for example, multiplying the base weight of 0.7 for the cable overheating zone by the adjustment factor of 1.5 yields an initial weight of 1.05. Finally, a summation normalization method is used to process the initial weights of all risk areas, ensuring the comparability of risk weights across regions in the global path optimization.

[0144] This embodiment addresses the technical challenges of real-time performance and accuracy in dynamic risk weighting. By multiplying the base weights and dynamic adjustment factors, it organically integrates the inherent attributes of risk with real-time evolutionary characteristics, ensuring that rapidly expanding or intensifying risk areas receive higher weight allocations promptly. Normalization eliminates the impact of different units on path planning, making the threat levels of each risk area comparable during global optimization and preventing path deviations due to local weight anomalies. This technical solution enables inspection path planning to dynamically respond to changes in risk conditions, improving inspection efficiency while ensuring safety.

[0145] In one optional embodiment, based on the constraints of path planning, the global path optimization objective function is solved using the A* algorithm. There are several ways to obtain the global inspection path. For example, the global inspection path can be obtained as follows: A comprehensive cost function is constructed, including distance cost, risk cost, and time cost. The distance cost is based on the estimated travel distance of the inspection robot from the current node to the next node and the additional distance consumed by complex terrain. The risk cost is determined by the sum of the products of the dynamic risk weights of the areas traversed by the inspection path and the corresponding exposure times. The time cost is based on the inspection robot's movement speed, estimated travel distance, and the time extension caused by avoidance detours. During the node expansion process using the A* algorithm based on the constraints of path planning, any candidate path in the candidate set is verified in real time. The feasibility of equipment status at each path node is assessed. This feasibility includes determining the power accessibility range based on power information and determining the inspection robot's ability to traverse complex terrain areas based on equipment operating status information. If the power information of any candidate path node exceeds the power accessibility range, or if the equipment operating status cannot meet the inspection robot's ability to traverse complex terrain areas, that candidate path node is marked as infeasible and removed from the candidate set. Based on a comprehensive cost function, the path with the minimum comprehensive cost value is selected from the remaining path nodes in the candidate set as the initial inspection path. The initial inspection path is dynamically adjusted based on the risk propagation prediction results to obtain the global inspection path. The risk propagation prediction results are used to indicate the spatial range and risk intensity trends of the target risk area during the inspection cycle. Dynamic adjustments include, but are not limited to, increasing safety distances, optimizing path curvature to shorten estimated travel time, and adjusting path order to avoid high-risk periods, ensuring the inspection robot can safely pass through before the risk area spreads, while maximizing inspection efficiency and minimizing the total risk exposure of the task.

[0146] Optionally, this embodiment further proposes to construct a comprehensive cost-value function that includes distance cost, risk cost, and time cost. The risk cost is determined by the sum of the products of the dynamic risk weights of the areas traversed by the path and the corresponding exposure time. During the node expansion process of the A* algorithm, the feasibility of the device status of candidate path nodes is verified in real time, including determining the power accessibility range based on power information and determining the ability to pass through complex terrain areas based on device operating status information. When any path node is detected to be outside the power accessibility range or the device operating status cannot meet the passage requirements, the path node is marked as infeasible and removed from the candidate set. The remaining feasible path nodes are evaluated based on the comprehensive cost-value function, and the path with the minimum comprehensive cost-value is selected as the initial inspection path. The initial inspection path is dynamically adjusted in combination with the risk propagation prediction results so that the robot can complete the passage before the risk area expands, thus obtaining the global inspection path.

[0147] In the comprehensive cost-benefit function, the risk cost is calculated by multiplying and summing the dynamic risk weight and the exposure time. The dynamic risk weight can be dynamically adjusted based on the type, level, and predicted expansion rate of the risk area. The exposure time is determined by the ratio of the path segment length to the robot's moving speed. During the equipment status feasibility verification process, the calculation of the power accessibility range can be combined with the ratio of the remaining power to the energy consumption per unit distance. For example, when the remaining power is 5000mAh and the energy consumption per unit distance is 200mAh / m, the upper limit of the power accessibility range is 25 meters. Equipment operating status information can include the flexibility of the robotic arm joints or the torque of the moving chassis. If the area where the path node is located requires the robotic arm to lift more than a preset threshold or the tilt angle to move more than the maximum torque support angle of the chassis, it is determined to be an infeasible node. During the dynamic adjustment process, the time margin can be calculated based on the difference between the predicted expansion time of the risk area and the robot's expected arrival time. For example, if a risk area is predicted to expand to the location of the path segment in 120 seconds, but the robot is expected to pass through the area in 100 seconds, the time margin is -20 seconds, triggering path point replanning.

[0148] Specifically, in the path planning process, the overall cost of the path is first quantified through a comprehensive cost-value function. The dynamic weighting mechanism of risk costs reflects the actual threat level of different risk areas, avoiding the problem of static weights failing to adapt to risk propagation. During the node expansion phase, the reachability range of the battery and the operating capacity of the equipment are verified in real time. For example, if the robot's remaining battery is insufficient to reach a candidate node, that node is directly eliminated, ensuring that the path planning conforms to actual execution conditions. For feasible nodes that pass the verification, after selecting the optimal initial inspection path based on the comprehensive cost-value, further dynamic adjustments are made based on the risk propagation prediction results. For example, when the time margin of a path segment is detected to be lower than the safety threshold, a detour path is used to increase the safety distance or the path curvature is optimized to reduce travel time. Simultaneously, the dynamic risk weights of the affected path segments are updated, and the cost-value is recalculated to balance safety and efficiency. The resulting global inspection path can avoid dynamically expanding risk areas under the constraints of robot battery power and equipment capacity, achieving safe and efficient inspection tasks.

[0149] As a preferred embodiment, the specific implementation of this method is as follows: During the global path planning process, the comprehensive cost function consists of distance cost, risk cost, and time cost. The risk cost is calculated as the sum of the products of the dynamic risk weights of the areas traversed by the path and the corresponding exposure times. During node expansion, the device status of candidate nodes is verified in real time. Specifically, the reachability of the power is determined by comparing the remaining power with the estimated total energy consumption of the path, and the conditions for traversing complex terrain are verified based on the device operating parameters. For example, when it is detected that the energy consumption required by a candidate node exceeds 85% of the remaining power, the node is marked as infeasible and removed from the candidate set. After the feasible nodes are evaluated based on the comprehensive cost, an initial inspection path is generated. The initial inspection path is dynamically adjusted based on the risk propagation prediction results. Specifically, by comparing the robot's estimated passage time with the predicted expansion time of the risk area, path segments with insufficient time margin are replanned. For example, if it is predicted that a partial discharge area will spread to the current path segment within 15 minutes, path curvature optimization is used to shorten the passage time to 12 minutes.

[0150] Through the above technical solutions, this application solves the problem of path planning being detached from equipment operating capabilities. Path feasibility is ensured by real-time verification of power availability and terrain traversal capability. A dynamic risk cost calculation mechanism, combined with risk propagation prediction, adjusts the path, enabling the robot to safely traverse risk areas before they expand, avoiding the shortcomings of traditional static weighting methods that cannot adapt to dynamic risk changes. The dynamic path adjustment strategy, through time window comparison and local replanning, maintains inspection efficiency while ensuring safety, achieving globally optimal path generation under multi-objective constraints.

[0151] In one optional embodiment, the initial inspection path is dynamically adjusted based on the risk propagation prediction results to obtain the global inspection path in several ways. For example, the global inspection path can be obtained as follows: Based on the time slice data in the risk propagation prediction results, the time window required for the inspection robot to pass through the neighborhood of each target risk area along the initial inspection path is determined and compared with the predicted expansion time of the corresponding target risk area to obtain the time margin; for high-risk path segments with a time margin lower than a preset safety threshold, a path replanning strategy is adopted to adjust the path. The path replanning strategy includes planning a detour path to increase the safety distance or optimizing the path curvature to reduce the expected travel time, so that the robot can safely pass before the risk area expands; during the path adjustment process, the dynamic risk weight is updated in real time, and the risk cost of the inspection path segment affected by the corresponding target risk area is recalculated so that the adjusted inspection path maintains a balance between distance cost, risk cost, and time cost; based on the comprehensive cost of each adjusted path segment, the inspection path combination with the smallest cost is reselected to generate the global inspection path.

[0152] Optionally, the method in this embodiment proposes to determine the time window required for the inspection robot to pass through the neighborhood of each risk area along the initial inspection path based on the time slice data in the risk propagation prediction results, and compare it with the predicted expansion time of the corresponding risk area to obtain the time margin; for high-risk path segments with a time margin lower than the safety threshold, a path point replanning strategy is adopted for adjustment, including planning detour paths to increase the safety distance, or optimizing the path curvature to reduce the expected travel time; during the dynamic adjustment of the path, the dynamic risk weight is updated in real time, and the risk cost of the affected path segments is recalculated; based on the comprehensive cost of each adjusted path segment, the combination with the smallest cost is reselected to generate the global inspection path.

[0153] The time slice data is divided into discrete time units in seconds, with each unit corresponding to a millimeter-precision predicted value for the risk region boundary expansion. A safety threshold is set at 1.2 times the predicted expansion time; when the time margin falls below this threshold, a path replanning strategy is automatically triggered. During detour path planning, the safety distance increment is controlled within 15% to 25% of the original path length, and path curvature optimization uses a Bézier curve algorithm to reduce the turning radius to 1.1 times the robot's minimum turning radius. The update frequency of the dynamic risk weights is synchronized with the refresh cycle of the risk propagation prediction results. After each update, the risk cost of the affected path segment is recalculated using a sliding window method, with the window length set to 10% of the total path length.

[0154] Specifically, after the initial inspection path is generated, the minute-level time slice data stored in the risk propagation prediction model is first extracted. The robot's kinematics model is used to calculate the specific time window for traversing each risk area's neighborhood. For example, if a path traverses from 08:15:30 to 08:16:15, and the predicted risk area expansion time is 08:16:20, then the time margin is calculated to be 65 seconds. When a path margin is detected to be lower than the preset 78-second safety threshold, the path planning module automatically generates two alternative paths: one is a detour path with a 23% increase in length, expanding the safe distance to 1.5 meters; the other optimizes the path curvature, reducing the turning radius from 0.8 meters to 0.7 meters, and shortening the estimated travel time by 12 seconds. During the adjustment process, the dynamic risk weights are updated every 30 seconds, and the recalculated path segment risk value integrates the distance, risk, and time parameters using a weighted average method. The final global path maintains 83% of the structure of the initial inspection path, while optimizing and adjusting 17% of the high-risk sections, enabling the robot to pass through the risk area 120 seconds before it expands.

[0155] As a preferred embodiment, the method of this embodiment is implemented as follows: In generating the global inspection path, firstly, based on the time slice data in the risk propagation prediction results, the time window for the inspection robot to pass through the neighborhood of each risk area along the initial inspection path is calculated, and compared with the predicted expansion time of the corresponding risk area to generate a time margin index. When the time margin of a path segment is detected to be lower than the safety threshold, a path replanning strategy is triggered. For example, a detour path is planned near the cable overheating area to increase the safety distance, or the path curvature is optimized near the partial discharge area to shorten the expected travel time. During the adjustment process, the dynamic risk weight is updated according to the real-time risk propagation data, and the risk cost of the affected path segment is recalculated through a weighted accumulation method, so that the adjusted path is in a balanced state in terms of cost in the three dimensions of distance, risk, and time. Finally, a globally optimal path is generated through a cost-value minimization algorithm, where the combination selection of path segments comprehensively considers the dynamic matching between real-time risk changes and the robot's travel capability.

[0156] This embodiment addresses the issue of dynamic risk exposure caused by insufficient time margin in the initial inspection path. A dynamic adjustment mechanism ensures the robot's safe passage before the risk area expands. Specifically, the comparison mechanism between the time window and the predicted expansion time accurately identifies high-risk path segments. Detour paths and curvature optimization strategies reduce the probability of risk exposure from two dimensions: spatial avoidance and time compression, respectively. Dynamic weight updates and cost recalculation enable path planning to respond in real-time to the uncertainty of risk propagation, avoiding the insufficient adaptability of traditional static weights. The resulting global path ensures inspection efficiency while improving the safety and reliability of robot operations in complex dynamic environments.

[0157] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0158] Through steps S102 to S108, the real-time acquisition of environmental data by multimodal sensors (discharge sensing, infrared thermal imaging, and acoustic signature acquisition), combined with the robot's own power level and equipment status monitoring, achieves the goal of accurately identifying potential high-risk areas and dynamically planning safe paths. This improves the accuracy and comprehensiveness of risk identification during robot inspection, ensuring the feasibility and safety of robot path planning. It also solves the technical problem in related technologies where inaccurate risk identification and avoidance in path planning for inspection robots in cable tunnels leads to low accuracy and poor safety. Specifically, this embodiment first considers how to collaboratively analyze multimodal environmental perception data and robot dynamic status to solve the coupling problem between composite risk identification and path planning. Traditional methods use an architecture that processes data from each sensor independently, leading to the failure of risk decoupling. This embodiment attempts to establish a correlation model between multi-source data and improve the identification accuracy of risk area boundaries through feature fusion. Simultaneously, addressing the spatiotemporal mismatch between path planning and risk situation, it explores the possibility of combining a risk propagation prediction model with a path optimization algorithm to achieve proactive avoidance of dynamic risk areas. Regarding equipment state constraints, it was found that relying solely on static power estimation could lead to misjudgments of path feasibility. Therefore, a dynamic constraint verification mechanism based on real-time state data was investigated. By comparing different sensor fusion schemes, it was discovered that a multimodal correlation model based on the covariance matrix can capture the nonlinear relationship between temperature gradient and discharge intensity, while feature weight allocation based on information entropy can reduce the impact of sensor noise on risk identification. Ultimately, an architecture combining global path optimization and local obstacle avoidance collaborative control was adopted, ensuring both overall path safety and maintaining flexibility in real-time environmental response.

[0159] This embodiment also provides a cable tunnel robot path planning device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0160] According to an embodiment of this application, an apparatus embodiment for implementing the above-described cable tunnel robot path planning method is also provided. Figure 3 This is a schematic diagram of the structure of a cable tunnel robot path planning device according to an embodiment of this application, as shown below. Figure 3As shown, the above-mentioned cable tunnel robot path planning device includes: an acquisition module 301, an identification module 302, a generation module 303, and a sending module 304. The acquisition module 301 is used to acquire environmental perception data and robot status data collected by the inspection robot in response to a cable tunnel inspection task request. The environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data. The robot status data includes at least the inspection robot's power information and equipment operating status information. The identification module 302, connected to the acquisition module 301, is used to identify target risk areas in the cable tunnel based on the environmental perception data. The target risk areas include cable overheating areas, acoustic signature abnormality areas, and partial discharge areas. The generation module 303, connected to the identification module 302, is used to generate a global inspection path based on the target risk areas and the inspection robot status data. The sending module 304, connected to the generation module 303, is used to send the global inspection path to the inspection robot for path planning and local obstacle avoidance.

[0161] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0162] It should be noted that the acquisition module 301, identification module 302, generation module 303, and sending module 304 mentioned above correspond to steps S201 to S204 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0163] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0164] The aforementioned cable tunnel robot path planning device may also include a processor and a memory. The aforementioned acquisition module 301, recognition module 302, generation module 303, and sending module 304 are all stored in the memory as program modules, and the processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0165] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip. Figure 4This is a schematic diagram of an edge computing node provided in an embodiment of this application. The edge computing node 600 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 601 and one or more memories 602. The one or more memories 602 store at least one computer program, which is loaded and executed by the one or more processors 601 to implement the methods provided in the above-described method embodiments. Of course, the edge computing node 600 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The edge computing node 600 may also include other components for implementing device functions, which will not be elaborated here.

[0166] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program is running, it controls the device where the non-volatile storage medium is located to execute any of the cable tunnel robot path planning methods described above.

[0167] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0168] Optionally, a program that controls the device containing the non-volatile storage medium to execute any of the above-mentioned cable tunnel robot path planning method steps during program execution.

[0169] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described cable tunnel robot path planning methods.

[0170] According to an embodiment of this application, an embodiment of a computer program product is also provided, which, when executed on a data processing device, is adapted to execute a program that initializes the cable tunnel robot path planning method steps described above.

[0171] This application provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the above-described cable tunnel robot path planning methods.

[0172] The order of the embodiments described above is merely for illustrative purposes and does not represent the superiority or inferiority of the embodiments.

[0173] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0174] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0175] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0176] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0177] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0178] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A path planning method for a cable tunnel robot, characterized in that, include: In response to a request for an inspection of a cable tunnel, environmental perception data and robot status data collected by the inspection robot are acquired. The environmental perception data includes discharge perception data, infrared thermal imaging data, and acoustic signature data. The robot status data includes at least the power information and equipment operating status information of the inspection robot. Based on the environmental perception data, target risk areas in the cable tunnel are identified, including cable overheating areas, abnormal acoustic signature areas, and partial discharge areas. Based on the target risk area and the inspection robot status data, a global inspection path is generated; The global inspection path is sent to the inspection robot for the robot to perform path planning and local obstacle avoidance.

2. The method according to claim 1, characterized in that, The process of identifying target risk areas in cable tunnels based on the environmental perception data includes: Based on the environmental perception data, feature extraction is performed to obtain multiple environmental features, including temperature gradient features extracted based on the infrared thermal imaging data, frequency domain distribution features and energy attenuation features extracted based on the acoustic signature acquisition data, and pulse sequence features and discharge intensity features extracted based on the discharge perception data. Based on the multiple environmental features, feature fusion is performed to obtain the risk coupling strength between different risk types. The risk coupling strength is used to indicate the co-evolution trend of different risk types in the cable tunnel, as well as the possible spatial dependence and temporal correlation between different risk types. Based on the aforementioned risk coupling strength, composite risk regions with synergistic risk characteristics are identified; The target risk area is determined based on the spatial distribution of the composite risk area in the cable tunnel and the risk coupling strength.

3. The method according to claim 2, characterized in that, The feature fusion based on the multiple environmental features to obtain the risk coupling strength between the different risk types includes: The thermoelectric coupling coefficient between the temperature gradient feature and the discharge intensity feature, and the acoustic-electric coupling coefficient between the frequency domain distribution feature and the pulse sequence feature are determined. The thermoelectric coupling coefficient is used to indicate the degree of linear influence of the temperature gradient feature on the partial discharge intensity feature, and the acoustic-electric coupling coefficient is used to indicate the degree of linear influence of the frequency domain distribution feature of the acoustic signature data on the pulse sequence feature of the discharge sensing data. Determine the information entropy weights corresponding to each of the plurality of environmental features, wherein the information entropy weights are used to quantify the uncertainty of the corresponding environmental features in spatial distribution; Based on the thermoelectric coupling coefficient, the acoustic-electric coupling coefficient, and the information entropy weights corresponding to the multiple environmental features, the weighting coefficients corresponding to the multiple environmental features are determined. Based on the weighting coefficients corresponding to each of the multiple environmental features, the multiple environmental features are weighted and fused to obtain a comprehensive risk feature map. The comprehensive risk feature map is used to indicate at least the spatial correlation between feature values ​​of different risk types, and the feature values ​​are used to quantify the risk intensity and changing trend in the cable tunnel. Based on the comprehensive risk feature map, the risk coupling strength is obtained, wherein the risk coupling strength includes linear correlation strength and nonlinear dependence strength.

4. The method according to claim 2, characterized in that, When there are multiple composite risk areas, determining the target risk area based on the spatial distribution of the composite risk areas in the cable tunnel and the risk coupling strength includes: The risk value corresponding to each of the multiple composite risk regions is obtained by performing a weighted summation operation based on the linear correlation strength and nonlinear dependence strength corresponding to each of the multiple composite risk regions and the weight values ​​corresponding to the linear correlation strength and nonlinear dependence strength. Based on the spatial distribution of the multiple composite risk areas in the cable tunnel and the risk values ​​corresponding to each of the multiple composite risk areas, a risk value distribution map is generated using a spatial interpolation method. The risk value distribution map is used to reflect the spatial variation of risk values ​​in the cable tunnel. Based on preset multi-level risk thresholds, contour lines are extracted from the risk value distribution map to obtain the boundaries of areas with different risk levels. The multi-level risk thresholds are set according to preset safety standards for cable tunnels. Based on the aforementioned regional boundaries, the target risk area is determined.

5. The method according to claim 1, characterized in that, When there are multiple target risk areas, the step of generating a global inspection path based on the target risk areas and the robot state data includes: Predict the spatial range change trends of multiple target risk areas during the inspection cycle; A global path optimization objective function is constructed with the goal of minimizing the total path risk value and maximizing the inspection efficiency. The total path risk value is determined by the risk weight of the neighborhood of the corresponding risk area traversed by the inspection robot and the exposure time of the inspection robot due to the spatial range change trend of the corresponding risk area neighborhood. The exposure time refers to the duration of the risk or risk diffusion process that may exist in the neighborhood of the corresponding risk area. Based on the robot's state data, the constraints for path planning are determined; Based on the constraints of the path planning, the global path optimization objective function is solved using the A* algorithm to obtain the global inspection path.

6. The method according to claim 5, characterized in that, The prediction of the spatial range change trends of multiple target risk areas within the inspection cycle includes: The historical evolution data and real-time monitoring data corresponding to each of the multiple target risk areas are input into a preset risk propagation predictor to obtain time-series risk intensity data. The risk propagation predictor is trained by machine learning methods and is used to simulate the diffusion behavior of different risk types in the cable tunnel environment. The time-series risk intensity data is used to indicate the risk intensity distribution of each of the multiple target risk areas at different sampling times. Based on the time-series risk intensity data, determine the range expansion rate and risk intensity change trend of each of the multiple target risk areas; Based on the risk type, risk level, scope expansion rate, and risk intensity change trend of each of the multiple target risk areas, determine the dynamic risk weight corresponding to each of the multiple target risk areas; Based on the risk area prediction boundaries and dynamic risk weights corresponding to the multiple target risk areas, the spatial range change trend is obtained.

7. The method according to claim 6, characterized in that, The step of determining the range expansion rate and risk intensity change trend of each of the multiple target risk areas based on the time-series risk intensity data includes: Based on the time-series risk intensity data, risk distribution raster maps corresponding to multiple sampling times within the inspection cycle are generated, wherein the multiple sampling times are sampled at preset time intervals within the inspection cycle; Boundary extraction is performed on the risk distribution raster maps corresponding to each of the multiple sampling times to obtain the predicted risk area boundaries corresponding to each of the multiple sampling times. By comparing the predicted boundaries of the risk areas at adjacent sampling times, the area change rate and centroid movement vector corresponding to each of the multiple target risk areas are determined, wherein the centroid movement vector is used to indicate the overall movement direction and speed of the risk area in space. Based on the area change rate and the time derivative of the centroid movement vector corresponding to each of the multiple target risk areas, the range expansion rate and main expansion direction corresponding to each of the multiple target risk areas are determined. The range expansion rate is used to quantify the expansion speed of the risk area boundary over time, and the main expansion direction is used to indicate the main direction of the risk area boundary expansion. Based on the intensity characteristic values ​​of the core areas corresponding to the multiple target risk areas and the time-series risk intensity data, the intensity change trends corresponding to the multiple target risk areas are obtained. The core area refers to the area where the risk intensity is higher than a preset risk threshold or the degree of change is greater than a preset degree. The intensity characteristic values ​​are used to indicate the characteristic risk intensity within the risk area.

8. The method according to claim 6, characterized in that, The step of determining the dynamic risk weights corresponding to each of the multiple target risk areas based on their respective risk types, risk levels, expansion rates, and risk intensity trends includes: Based on the risk type and risk level corresponding to each of the multiple target risk areas, a weight allocation strategy corresponding to each of the multiple target risk areas is determined. Based on the self-corresponding weight allocation strategy, the basic weights corresponding to the multiple target risk areas are determined. Among them, different basic weights are assigned to the designated risk areas according to the potential hazard level of the corresponding target risk areas to the inspection robot. The designated risk areas are the target risk areas with corresponding risk types of cable overheating area, partial discharge area and acoustic abnormality area. Based on the expansion rate and risk intensity change trend of each of the multiple target risk areas, determine the dynamic adjustment factor corresponding to each of the multiple target risk areas; The basic weights corresponding to each of the multiple target risk areas are multiplied by the corresponding dynamic adjustment factors to obtain the preliminary risk weights corresponding to each of the multiple target risk areas. The initial risk weights corresponding to each of the multiple target risk regions are normalized to obtain the dynamic risk weights corresponding to each of the multiple target risk regions.

9. The method according to claim 5, characterized in that, Based on the constraints of the path planning, the global path optimization objective function is solved using the A* algorithm to obtain the global inspection path, which includes: A comprehensive cost function is constructed, which includes distance cost, risk cost, and time cost. The distance cost is based on the estimated travel distance of the inspection robot from the current node to the next node and the additional distance consumption caused by complex terrain. The risk cost is determined by the sum of the products of the dynamic risk weights of the areas traversed by the inspection path and the corresponding exposure time. The time cost is based on the inspection robot's movement speed, estimated travel distance, and the time extension caused by risk avoidance detours. During the node expansion process using the A* algorithm based on the constraints of the path planning, the feasibility of the device status of any candidate path node in the candidate set is verified in real time. The feasibility of the device status includes determining the range of power reachability based on the power information and determining the ability of the inspection robot to pass through complex terrain areas based on the device operating status information. If the power information of any candidate path node exceeds the power accessibility range, or if the device operating status cannot meet the ability of the inspection robot to pass through complex terrain areas, then any candidate path node will be marked as infeasible and removed from the candidate set. Based on the comprehensive cost function, the path with the smallest comprehensive cost value is selected from the remaining path nodes in the candidate set as the initial inspection path. The initial inspection path is dynamically adjusted based on the risk propagation prediction results to obtain the global inspection path. The risk propagation prediction results are used to indicate the spatial range and risk intensity change trends of the target risk area within the inspection cycle.

10. The method according to claim 9, characterized in that, The process of dynamically adjusting the initial inspection path based on the risk propagation prediction results to obtain the global inspection path includes: Based on the time slice data in the risk propagation prediction results, the time window required for the inspection robot to pass through the neighborhood of each target risk area along the initial inspection path is determined, and compared with the predicted expansion time of the corresponding target risk area to obtain the time margin. For high-risk path segments with a time margin lower than a preset safety threshold, a path replanning strategy is adopted to adjust the path. The path replanning strategy includes planning detour paths to increase the safety distance or optimizing the path curvature to reduce the expected travel time, so that the robot can safely pass through before the risk area expands. During the path adjustment process, the dynamic risk weight is updated in real time, and the risk cost of the inspection path segment affected by the corresponding target risk area is recalculated so that the adjusted inspection path maintains a balance between distance cost, risk cost, and time cost. Based on the adjusted comprehensive cost value of each path segment, the inspection path combination with the minimum cost value is reselected to generate the global inspection path.

11. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the cable tunnel robot path planning method according to any one of claims 1 to 10.

Images