A multi-sensor fusion-based comprehensive intelligent perception platform system for underground cable trench

Abstract

The application provides a kind of underground cable trench comprehensive intelligent perception platform system based on multi-sensor fusion, it is related to underground power cable monitoring technical field, including heterogeneous sensing acquisition module, edge computing node, cloud service platform, the heterogeneous sensing acquisition module includes fixed multi-parameter sensor, distributed optical fiber sensor, mobile robot vision thermal image sensing unit;The application adopts cable trench scene multi-source heterogeneous perception system, constructs edge cloud coordination space-time fusion perception architecture, integrates fixed multi-parameter sensor, distributed optical fiber, mobile robot vision thermal image three kinds of heterogeneous sensing equipment, breaks the barrier of traditional single module independent monitoring, through the edge node unified space-time reference, the edge cloud coordination mode of cloud depth fusion calculation, realizes the integration of multi-source, different frequency, different physical mechanism sensing data, improves the integrity, comprehensiveness and data consistency of cable trench scene perception.

Description

Technical Field

[0001] This invention relates to the field of underground power cable monitoring technology, and in particular to an integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. Background Technology

[0002] Existing underground cable trench monitoring technologies mostly use single-type sensors or independent monitoring modules to collect data. They mainly rely on fixed single-parameter sensors, distributed fiber optic sensing devices, or mobile monitoring robots to operate independently. Each type of monitoring device has its own independent data acquisition frequency, physical sensing mechanism, and data output format. At the same time, they use independent edge and cloud computing modes to complete data processing and analysis. They can only collect single-dimensional data on the local electrical, environmental, or structural aspects of the cable trench. They lack the overall construction of a multi-dimensional heterogeneous sensing system, which is the current mainstream technical solution for cable trench monitoring. However, in existing technologies, various types of sensor data are independent of each other, and the spatiotemporal references are not unified, forming serious data silos. The data perception integrity and consistency are poor, and multi-source data correlation analysis cannot be achieved. Multi-source sensor data lacks a high-precision spatiotemporal synchronization and alignment mechanism, and data of different frequencies and different physical mechanisms cannot be accurately matched, resulting in a lack of reliable physical basis for subsequent data analysis. Traditional SLAM mapping and positioning methods rely on a single lidar, which cannot identify non-planar obstacles in cable trenches, resulting in large loop closure positioning errors and low accuracy of robot collaborative perception. Fault monitoring relies on single sensor features, resulting in insufficient positioning accuracy and the absence of a standardized hazard classification and early warning system. The operation and maintenance response lacks hierarchical basis and a quantitative cable health status assessment model, which can only achieve passive fault monitoring and cannot quantify the health status of equipment. The operation and maintenance mode relies on traditional planned maintenance, resulting in low intelligence. Therefore, this invention proposes an integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion to solve the problems existing in the prior art. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. This system employs a multi-source heterogeneous sensing architecture for cable trench scenarios, constructing an edge-cloud collaborative spatiotemporal fusion sensing architecture. It integrates three types of heterogeneous sensing devices: fixed multi-parameter sensors, distributed optical fibers, and mobile robot visual thermal imaging. This breaks down the barriers of traditional single-module independent monitoring. Through a unified spatiotemporal reference at edge nodes and a deep cloud-based fusion computing edge-cloud collaborative mode, it achieves integrated data from multiple sources, different frequencies, and different physical mechanisms. This enhances the completeness, comprehensiveness, and data consistency of cable trench scenario sensing, providing full-dimensional data support for subsequent fault analysis and status assessment.

[0004] To achieve the objectives of this invention, the following technical solution is provided: A comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion, comprising a heterogeneous sensing acquisition module, an edge computing node, and a cloud service platform, characterized in that: the heterogeneous sensing acquisition module includes a fixed multi-parameter sensor, a distributed fiber optic sensor, and a mobile robot visual thermal imaging sensing unit, used to simultaneously acquire three types of heterogeneous monitoring data of cable trench electrical quantities, environmental quantities, and structural quantities; the edge computing node is used to perform nanosecond-level timestamp annotation on the multi-source heterogeneous sensing data, register it with a spatial coordinate system based on the cable trench topology map, and generate a sensing data stream with a unified spatiotemporal label; The cloud service platform incorporates an improved RBPF-SLAM collaborative mapping module, a fault diagnosis and location module, a four-level hierarchical early warning module, and a CHI health assessment module. These modules are used to integrate multiple algorithms to achieve environmental mapping, fault location, hazard early warning, and quantitative assessment of health status. Based on the edge-cloud collaborative architecture, the platform enables intelligent perception and operation and maintenance decision-making for the entire cable trench scenario.

[0005] A further improvement is made in that: the CHI health assessment module incorporates a cable health index (CHI) assessment model, and the calculation formula for the CHI assessment model is as follows: ; The parameters are explained as follows: CHI is the cable health index, with a value range of [0,100]. The larger the value, the better the health status of the cable equipment; n is the total number of sensor types participating in the health assessment; i is the sensor type number, with values ​​of 1, 2, 3...n; wi is the weight coefficient corresponding to the i-th type of sensor, which is obtained by iterative training of the expert system combined with historical cable fault data, and the sum of the weight coefficients is 1. This is the normalization mapping function for the monitoring data of the i-th type of sensor, used to convert the original monitoring data of different dimensions and different numerical ranges into standardized values ​​in the [0,1] interval; This refers to the raw monitoring data collected by the i-th type of sensor.

[0006] A further improvement is that the improved RBPF-SLAM collaborative mapping module incorporates an improved RBPF-SLAM algorithm, and this improved RBPF-SLAM algorithm includes a formula for calculating the joint distribution of observations, used for multi-sensor data fusion modeling. The formula is as follows: ; in, Let be the joint probability distribution of the mobile robot pose and the environment map at time k; represents the continuous pose state variables of the robot from time 1 to k; m represents the constructed grid map model of the cable trench environment. The environmental observation data from multiple sensors at times 1 to k; The odometer control data output by the robot encoder and gyroscope from time 1 to k; Estimating the probability of a map given its pose and observation data; This represents the pose estimation probability based on known observation and control data.

[0007] Further improvements are made in that the iterative operation process of the improved RBPF-SLAM algorithm includes four steps: sampling, weight calculation, resampling, and map estimation, and integrates the ORB feature extraction function of the depth camera; the ORB feature extraction is used to collect feature vectors of special regions in the image inside the cable trench, identify non-planar obstacles such as suspended cables and semi-suspended stones, make up for the detection defects of a single LiDAR, and suppress the positioning error caused by loop feedback.

[0008] A further improvement is made in that the fault diagnosis and location module is equipped with a time-domain differential correction algorithm to eliminate time-series deviations in multi-source fault data. The algorithm formula is as follows: ; in, T0 is the time-domain correction deviation value of a single set of fault data; T0 is the original timestamp of the fault sensing data of the current frame; k is the total number of historical data frames involved in the correction calculation; t is the sequence number of the historical data frame; Tt is the timestamp of the historical fault data of the t-th frame.

[0009] A further improvement is that the fault diagnosis and location module is also equipped with a fault location cross-correction calculation formula, which is used to accurately locate the cable fault point. The formula is as follows: ; Where L represents the final corrected precise location of the cable fault; L1 represents the fault location calculated based on distributed optical fiber temperature field data; L2 represents the fault location calculated based on the time difference of the fault impact from the vibration sensor; L3 represents the fault location calculated based on the amplitude attenuation law of the partial discharge signal; α, β, and γ are the location correction weights corresponding to the three types of fault characteristics, and satisfy the following conditions: The weights are derived from training and optimization using historical fault location data.

[0010] Further improvements are made in the following aspects: The four-level hierarchical early warning module has a built-in four-level hierarchical early warning system, including the prompt level, the warning level, the alarm level, and the emergency level; the prompt level pushes maintenance attention information when parameters show a trend deviation and there are no clear fault characteristics; the warning level automatically generates an inspection work order when there are potential hidden dangers; the alarm level pushes on-site handling notices when a fault is confirmed; and the emergency level automatically triggers the emergency response process in the event of an emergency such as an open flame or water level exceeding the limit.

[0011] Further improvements include: the cloud service platform, based on the CHI health index and time series trend analysis, generates three levels of operation and maintenance decision suggestions: early warning, alarm, and maintenance. It simultaneously outputs a full-area health heat map of the cable trench, enabling visualized monitoring of cable health status and supporting the transformation of power operation and maintenance from a planned maintenance model to a condition-based maintenance model. A further improvement is that the edge computing node is equipped with a three-dimensional spatial interpolation algorithm for accurate registration of spatial coordinates of multi-source sensor data. The algorithm formula is as follows: ; The parameters are defined as follows: S(x,y,z) is the interpolated sensor data at the three-dimensional coordinates (x,y,z) of the cable trench; x, y, and z are the horizontal, vertical, and longitudinal coordinates of the three-dimensional space of the cable trench, respectively; m is the number of effective peripheral sensors involved in the interpolation calculation; j is the serial number of the peripheral sensor. The spatial interpolation weight of the j-th sensor is calculated inversely proportional to the spatial distance between the sensor and the target point. Install coordinates on the j-th sensor. The raw data collected at the location.

[0012] Further improvements are made in the following aspects: the heterogeneous sensing acquisition module collects three types of monitoring data: electrical quantities, environmental quantities, and structural quantities. The electrical quantities include high-frequency partial discharge pulses, sheath circulating current, and cable joint temperature, with an acquisition frequency of 10kHz-50kHz. The environmental quantities include photoelectric point-type water immersion level, MEMS electrochemical harmful gas concentration, and digital temperature and humidity, with an acquisition frequency of 1kHz-5kHz. The structural quantities include MEMS accelerometer micro-vibration data and fiber optic grating displacement and deformation data, with an acquisition frequency of 2kHz-10kHz. The edge computing node uses timestamp anchoring technology to control the time synchronization error of the three types of heterogeneous frequency data to the sub-millisecond level.

[0013] The beneficial effects of this invention are as follows: 1. This invention adopts a multi-source heterogeneous sensing system for cable trench scenarios, constructs an edge-cloud collaborative spatiotemporal fusion sensing architecture, integrates three types of heterogeneous sensing devices: fixed multi-parameter sensors, distributed optical fibers, and mobile robot visual thermal imaging, breaks through the barriers of traditional single-module independent monitoring, and achieves integrated integration of multi-source, different frequency, and different physical mechanism sensing data through edge node unified spatiotemporal reference and cloud deep fusion computing. This improves the integrity, comprehensiveness, and data consistency of cable trench scenario sensing, and provides full-dimensional data support for subsequent fault analysis and status assessment.

[0014] 2. This invention constructs a sub-millisecond-level spatiotemporal synchronization fusion mechanism for three types of data: electrical quantities, environmental quantities, and structural quantities. This mechanism enables precise alignment of multi-source data, solidifies the data foundation for intelligent sensing, and specifically adapts to the acquisition characteristics of the three types of heterogeneous sensor data. Through timestamp anchoring technology and three-dimensional spatial interpolation algorithms, it achieves precise correlation of all monitoring data under a unified spatiotemporal benchmark, eliminating data analysis distortion and fault misjudgment caused by spatiotemporal deviations. This improves the reliability and accuracy of multi-source data fusion analysis and provides precise physical data support for subsequent feature extraction, fault diagnosis, and health assessment.

[0015] 3. This invention employs an improved RBPF-SLAM multi-sensor collaborative mapping algorithm to enhance the localization and mapping accuracy in complex cable trench environments. It integrates multiple sensor information from LiDAR, depth camera, ultrasonic sensors, encoders, and gyroscopes, and adds ORB feature extraction functionality from the depth camera. This enables accurate identification of non-planar obstacles such as suspended cables and loose stones within the cable trench, solving the problems of a single 2D LiDAR failing to detect non-planar obstacles and excessive loopback feedback positioning errors. Through four-step iterative calculations, the robot's pose and environmental map are optimized in real time, adapting to the special scenarios of narrow, complex cable trenches with numerous irregular obstacles, ensuring the stability and accuracy of mobile sensing equipment operations.

[0016] 4. This invention constructs a multi-feature cross-validation fault location mechanism and a four-level hierarchical early warning system to improve fault location accuracy and the standardization of operation and maintenance response. It integrates three core fault features: distributed optical fiber temperature field data, vibration fault impact time difference, and partial discharge signal amplitude attenuation. Through time-domain differential correction and spatial interpolation algorithms, it achieves cross-correction of fault points, strictly controlling the fault location error within 1 meter. At the same time, it establishes a standardized four-level hidden danger hierarchical early warning mechanism, accurately classifying the early warning level based on the characteristics of hidden dangers, parameter deviation, and development trend, and automatically matching the operation and maintenance handling process. This realizes standardized management of the entire process of hidden danger from discovery, investigation, handling to emergency response, and improves the timeliness and professionalism of cable trench fault operation and maintenance.

[0017] 5. This invention constructs a comprehensive health assessment model for underground cable trenches (CHI), enabling quantitative analysis of equipment health status and promoting the intelligent transformation of operation and maintenance models. Sensor weights are determined through joint training with an expert system and historical fault data. A 0-100 point quantitative health assessment system is constructed using a normalized mapping function. Combined with time series trend analysis, three-level operation and maintenance decision suggestions and a visualized heat map are generated. This breaks away from the traditional operation and maintenance model of no quantitative assessment and passive fault repair, accurately predicting the aging and hidden danger development trends of cable equipment. It supports the transformation of power operation and maintenance from traditional planned maintenance to precise condition-based maintenance, reducing cable fault incidence and operation and maintenance costs. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the composition of the present invention. Detailed Implementation

[0019] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0020] Example 1 according to Figure 1 As shown, this embodiment proposes a comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. It is used to specifically realize the multi-source heterogeneous sensor edge-cloud collaborative spatiotemporal fusion sensing architecture for underground cable trenches, and solve the technical problems of independent operation of equipment, data fragmentation, and inability to be unified and integrated in traditional monitoring systems. The system is divided into three core implementation stages: hardware layered deployment, edge spatiotemporal preprocessing, and deep cloud-cloud fusion.

[0021] The first phase involves a layered deployment of heterogeneous sensing hardware across the entire area. This system constructs a three-dimensional sensing hardware system in an underground cable trench scenario. The first layer is a fixed multi-parameter sensor array, deployed along the cable trench at a standard of 5 meters per array, covering all dimensions of cable electrical condition monitoring, trench environment monitoring, and structural condition monitoring, enabling 24 / 7 uninterrupted data acquisition in static scenarios. The second layer is a distributed fiber optic sensing system, with temperature-measuring optical fibers laid synchronously along the cable body to achieve continuous, blind-spot-free distributed monitoring of the cable's temperature field, overcoming the shortcomings of point sensors such as large sampling intervals and discontinuous data. The third layer is a track-mounted mobile inspection robot, equipped with multiple sensing devices including LiDAR, depth cameras, ultrasonic sensors, and infrared thermal imagers. It dynamically inspects along a preset track, completing visualized imaging of the trench environment, obstacle detection, and equipment temperature thermal imaging monitoring, achieving complementary integration of static fixed-point monitoring and dynamic full-area inspection.

[0022] The second stage involves unified spatiotemporal preprocessing at the edge. Embedded edge computing nodes are deployed on-site to aggregate heterogeneous monitoring data from all fixed sensors, distributed optical fibers, and mobile robots. To address the differences in acquisition frequencies, communication protocols, and output formats among different devices, the data format is first standardized and parsed. Then, a nanosecond-level high-precision timestamp is applied to each frame of valid data to achieve a unified time reference for the entire domain. Simultaneously, a three-dimensional topological space coordinate system is constructed based on the as-built topological drawings of the cable trench. All sensor installation points and robot inspection coordinates are uniformly registered to correct spatial errors caused by equipment installation deviations and inspection position offsets, generating a standardized perception data stream with a unified spatiotemporal label.

[0023] The third stage involves deep integration and collaboration between the cloud and the platform. The cloud service platform receives standardized data streams preprocessed by edge nodes in real time, breaking down data barriers between various devices and completing multi-source data feature association, complementary fusion, and global analysis. This enables integrated, comprehensive perception of the electrical operating status, environmental conditions, and structural safety status of the cable trench. This embodiment completely eliminates the data silo problem of traditional solutions, significantly improving the integrity, continuity, and consistency of monitoring data, and providing a complete and reliable data foundation for subsequent fault diagnosis, mapping and location, and health assessment.

[0024] Example 2 according to Figure 1 As shown, this embodiment proposes an integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. Specifically, it realizes a sub-millisecond spatiotemporal synchronization fusion mechanism for multimodal heterogeneous sensors. Addressing the pain points of different physical mechanisms, large differences in sampling frequencies, and inconsistent spatiotemporal references among the three types of sensor data (electrical quantities, environmental quantities, and structural quantities), it constructs an integrated fusion scheme of high-precision time alignment and spatial mapping, which is divided into four implementation steps: data classification and acquisition, high-precision time synchronization, three-dimensional spatial registration, and multi-source data fusion output.

[0025] The first step is to classify and synchronously collect multi-dimensional heterogeneous data. The system starts three types of sensor acquisition tasks in parallel. Electrical data includes high-frequency partial discharge pulses, cable sheath circulating current, and thermocouple temperature data of the joint, with a sampling frequency set to 10kHz-50kHz to adapt to the transient change characteristics of cable electrical faults. Environmental data includes photoelectric water immersion level, MEMS electrochemical harmful gas concentration, and digital temperature and humidity data, with a sampling frequency set to 1kHz-5kHz to adapt to the slow change characteristics of the trench environment. Structural data includes MEMS accelerometer micro-vibration and fiber optic grating displacement deformation data, with a sampling frequency set to 2kHz-10kHz to adapt to the small deformation and vibration response characteristics of cables and structures.

[0026] The second step is sub-millisecond time alignment processing. The high-precision constant-temperature clock built into the edge computing node is used as the unified time reference for the entire domain. The timestamp anchoring technology is used to calibrate the three types of data collected at different frequencies frame by frame. The clock deviation compensation algorithm eliminates the time error caused by the device hardware clock drift and communication transmission delay. The time synchronization error of all heterogeneous data is strictly controlled at the sub-millisecond level to ensure accurate correspondence of multi-dimensional data under the same working conditions and at the same time.

[0027] The third step is three-dimensional spatial coordinate mapping and registration. Based on the three-dimensional topology model of the cable trench, a three-dimensional spatial interpolation algorithm is used to normalize the spatial coordinates of all discretely deployed sensor points. Spatial interpolation is performed to complete the data of sensor gaps and blind spots, transforming single-point discrete monitoring data into continuous spatial distribution data, thus solving the problems of spatial discontinuity and blind spots in traditional point-like monitoring data.

[0028] The fourth step is to unify the benchmark data fusion output. Under a unified spatiotemporal benchmark, the association, matching, and fusion of electrical, environmental, and structural data are completed, achieving seamless integration of multi-source heterogeneous data. Actual testing shows that the spatiotemporal alignment accuracy of multi-source data in this embodiment can reach 99.8%, completely solving the problems of analysis distortion and fault misjudgment caused by traditional data temporal misalignment and spatial mismatch, providing accurate, reliable, and unified physical data support for subsequent intelligent analysis algorithms.

[0029] Example 3 according to Figure 1 As shown, this embodiment proposes a comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. Specifically, it implements a multi-sensor collaborative mapping and loop closure detection method based on an improved RBPF-SLAM. Addressing the shortcomings of traditional single-lidar SLAM algorithms, such as the inability to identify non-planar obstacles, large positioning errors in the complex environment of cable trenches, and severe loop closure drift, this method fuses multiple types of sensor data and introduces visual feature optimization. Specifically, it consists of four core implementation processes: multi-source sensor data preprocessing, RBPF four-step iterative pose calculation, ORB visual feature-assisted optimization, and loop closure detection and map correction.

[0030] The first step is unified preprocessing of multi-source sensor data. The system simultaneously collects data from five types of devices on the robot: lidar, depth camera, ultrasonic sensor, encoder, and gyroscope. To address the issue of inconsistent data dimensions and formats from different sensors, the depth information from the depth camera and the ultrasonic ranging information are uniformly converted into a lidar point cloud data format, completing a coordinate system transformation and achieving homogeneous fusion of multiple types of external observation data. At the same time, encoder odometer data and gyroscope attitude data are collected to construct a robot motion odometry model, providing motion constraints for pose calculation.

[0031] The second step is the four-step iterative operation of the RBPF algorithm. Based on the RBPF-SLAM joint distribution formula, it uses four closed-loop iterations of sampling, weight calculation, resampling, and map estimation to solve the robot's current pose state and the environmental grid map in real time. In the sampling stage, particle pose samples are generated based on the odometry model. In the weight calculation stage, the particle weights are corrected by combining multi-sensor observation data. In the resampling stage, low-weight invalid particles are removed and high-confidence particles are retained. In the map estimation stage, the environmental map is updated iteratively based on the optimal particles to continuously optimize the mapping and positioning accuracy.

[0032] The third step is to optimize the visual features of the ORB, and add an ORB feature extraction mechanism for RGB images of depth cameras. For the unique scene features of cable trenches, the focus is on extracting feature points and feature vectors of special areas such as trench wall texture, cable brackets, suspended cables, suspended gravel, and equipment joints, and build a dedicated cable trench environment feature database to make up for the deficiency of a single 2D LiDAR that can only detect planar obstacles and cannot identify suspended obstacles.

[0033] The fourth step is intelligent loop closure detection and error correction. During the robot's inspection, the features of the current frame image are extracted in real time and matched with the pre-stored global feature library to accurately identify repeated inspection sections, complete loop closure detection, and correct pose drift errors accumulated over long-term inspections in real time, ensuring positioning stability during long-distance inspections. This embodiment can achieve full coverage recognition of all types of obstacles in cable trenches, with a 100% recognition rate for non-planar obstacles. The robot's positioning error is controlled within 3cm, and the mapping accuracy is improved by 60% compared to traditional algorithms. It is fully adaptable to the special operating scenarios of narrow, enclosed, and complex obstacle-prone cable trenches.

[0034] Example 4 according to Figure 1 As shown, this embodiment proposes an integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. Specifically, it realizes multi-feature cross-fault location and four-level hierarchical early warning functions, solving the problems of traditional cable fault monitoring relying on single sensor data, poor location accuracy, chaotic early warning standards, and lack of basis for operation and maintenance. The system is divided into four implementation modules: accurate location of multi-feature faults, fault data correction and optimization, four-level standardized hierarchical early warning, and hierarchical operation and maintenance linkage.

[0035] The first module is multi-dimensional fault feature fusion localization. The system deeply integrates three types of core fault-sensitive features: the continuous distribution characteristics of the temperature field of the entire cable collected by distributed optical fiber, the time difference characteristics of fault impact collected by vibration sensor, and the spatial characteristics of discharge signal amplitude attenuation collected by partial discharge sensor. It abandons the traditional judgment method of single temperature or partial discharge feature, and captures various fault hazards such as cable overheating, insulation aging, partial discharge, and structural loosening in an all-round way through the joint analysis of multi-dimensional fault features.

[0036] The second module is fault data algorithm correction and optimization. To address the issues of temporal deviation and spatial dispersion in multi-source fault data, a time-domain differential correction algorithm is used to unify the time base of various types of fault data, eliminating data deviations caused by transmission delays and sampling time differences. At the same time, a spatial interpolation algorithm is used to complete the discrete data of the fault area, achieving continuous fitting of the fault area data. Then, a multi-feature cross-correction formula is used to iteratively correct the preliminary positioning results, ultimately strictly controlling the cable fault positioning error to within 1 meter, significantly improving the accuracy of fault location.

[0037] The third module is for accurate four-level hazard classification. The system constructs a four-level early warning system with clear hierarchy and quantifiable judgment standards based on the degree of coupling of multi-source features, the magnitude of deviation of operating parameters from the benchmark threshold, and the temporal development trend of hazard parameters. The levels are from low to high as prompt level, early warning level, alarm level, and emergency level. Each level has a dedicated judgment threshold and triggering conditions to avoid the subjectivity and error of human experience judgment.

[0038] The fourth module is a tiered, interconnected operation and maintenance (O&M) system. Different warning levels automatically match corresponding O&M procedures: The alert level indicates a slow, gradual deviation in parameters without clear fault characteristics; the system only pushes a notification to the O&M terminal, requiring no on-site intervention. The warning level indicates the detection of stable potential hazards and minor parameter anomalies; the system automatically generates a fixed-point inspection work order, assigning O&M personnel to conduct on-site investigation and verification within a specified timeframe. The alarm level indicates the confirmation of a clear fault hazard and impaired equipment performance; an alarm notification is immediately pushed to the dedicated O&M team, requiring prompt on-site handling and rectification. The emergency level indicates the detection of emergencies such as open flames in the trench, excessive water levels, or severe partial discharge; it directly triggers the entire emergency response process and activates the emergency response plan. This embodiment achieves precise cable fault location, tiered hazard control, and closed-loop implementation of handling procedures. Fault O&M response efficiency is improved by 80% compared to traditional solutions, and the standardization and professionalism of O&M handling are significantly enhanced.

[0039] Example 5 according to Figure 1 As shown in the figure, this embodiment proposes a comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion. Specifically, it realizes the comprehensive evaluation function of cable health index (CHI) driven by multi-source data, which solves the industry pain points of traditional cable operation and maintenance, such as the lack of quantitative evaluation standards, only passive fault repair, and strong blindness in planned maintenance. It constructs a complete quantitative evaluation and intelligent decision-making system for the health status of underground cable trenches, which is divided into five implementation steps: evaluation parameter selection, model weight training, data normalization processing, CHI index calculation, intelligent decision-making and visualization output.

[0040] The first step is to select multi-dimensional evaluation parameters, screening eight core key monitoring parameters across three dimensions—electrical, environmental, and structural—as model inputs. These parameters cover all dimensions of cable operation electrical safety, trench environmental conditions, and equipment structural safety, ensuring the comprehensiveness and systematic nature of the evaluation system.

[0041] The second step is intelligent training of model weights. A weight training mechanism that integrates expert experience and historical data is established. Based on the experience of experts in the power cable operation and maintenance industry, the basic weights of various parameters are initially calibrated. Then, combined with more than ten years of cable historical fault data, defect rectification data, and equipment aging data, iterative training and optimization are carried out to finally determine the optimal weight coefficient of each type of sensor parameter, ensuring that the weight allocation fits the actual operation and maintenance scenario.

[0042] The third step is the normalization preprocessing of monitoring data. To address the issues of different dimensions, large differences in numerical ranges, and inability to directly calculate weighted values ​​for various monitoring parameters, a dedicated normalization mapping function is constructed to uniformly convert all original monitoring data into standardized values ​​in the [0,1] interval, eliminating dimensional interference and ensuring the accuracy of model calculations.

[0043] The fourth step is to quantify and calculate the CHI health index. Based on the weighted coefficients and normalized data after training, the comprehensive health index of the cable is calculated using the CHI weighted summation formula. The index value range is fixed at [0, 100]. The higher the value, the better the overall health status of the cable equipment. The lower the value, the more serious the aging of the equipment and the hidden dangers.

[0044] The fifth step is trend analysis and intelligent operation and maintenance decision-making. The system combines long-term CHI index change curves to analyze the rate of equipment health status decay and the development trend of potential hazards, automatically generating intelligent operation and maintenance decision-making suggestions at three levels: early warning, alarm, and maintenance. Simultaneously, based on the global cable CHI index distribution, it automatically generates a cable trench health heat map, visually displaying the health status of cables in different sections and accurately locating weak points in the equipment. This embodiment achieves a breakthrough from qualitative judgment to quantitative assessment of cable health status, with a condition-based maintenance decision-making accuracy rate exceeding 95%. It completely changes the traditional extensive planned maintenance model and effectively supports the comprehensive transformation of power operation and maintenance towards precise and intelligent condition-based maintenance.

[0045] Comparative Example: Traditional Single-Sensor Monitoring System for Underground Cable Trench This comparative example uses an existing conventional underground cable trench monitoring technology to compare its performance with the technology of this invention. The specific implementation of the traditional solution is as follows: The system only independently deploys distributed fiber optic temperature measurement equipment and simple partial discharge monitoring equipment, lacking a three-dimensional multi-source heterogeneous sensing system and environmental and structural monitoring sensors, thus failing to achieve full-dimensional data acquisition; the hardware devices are independent of each other without data interaction, each device carries independent data acquisition and analysis programs, lacks an edge computing spatiotemporal preprocessing module, and the time and spatial references of various monitoring data are completely inconsistent, forming serious data silos; the monitoring robot is only equipped with a basic 2D lidar and uses a traditional single-laser SLA. The M-algorithm for mapping and localization lacks a depth camera visual feature fusion mechanism, making it unable to identify non-planar obstacles such as suspended cables and loose pebbles in cable trenches. This results in significant drift errors during long-distance inspections and loopback positioning. Fault determination relies solely on single temperature or partial discharge data, lacking multi-feature cross-correction algorithms. Fault localization depends entirely on single parameter thresholds, lacking a standardized, hierarchical hazard warning system, leading to a high probability of missed or false hazard detections. Furthermore, traditional solutions lack quantitative health assessment models, failing to digitally and quantitatively evaluate the health status of cable equipment. Maintenance relies entirely on fixed-cycle manual inspections, allowing only reactive repairs after obvious equipment failures, resulting in extremely low levels of intelligence and precision in maintenance. Actual testing revealed that traditional solutions suffer from a series of technical shortcomings, including limited monitoring dimensions, lack of data fusion capabilities, poor mapping and localization accuracy, large fault location errors, a lack of early warning systems, and outdated maintenance models. Overall, their monitoring performance and intelligence level are far inferior to this invention.

[0046] Validation data: To systematically verify the practical application performance, technical feasibility, and engineering superiority of this invention, a 100-meter standard real underground cable trench experimental test scenario was constructed. This scenario replicated the cable laying method, trench structure, and operation and maintenance conditions of an actual project. The invention was then compared with traditional monitoring solutions across all dimensions, and the quantitative test results are as follows: 1. Spatiotemporal fusion accuracy test: This invention uses edge-end nanosecond-level timestamp calibration and three-dimensional spatial registration technology to achieve a multi-source heterogeneous data time synchronization error of ≤0.8ms and a spatial coordinate registration error of ≤2cm. Traditional solutions lack a unified spatiotemporal reference, resulting in a data time synchronization error of ≥15ms and a spatial matching error of ≥30cm. This invention improves the spatiotemporal alignment accuracy by more than 90%.

[0047] 2. Mapping and localization performance test: The improved RBPF-SLAM algorithm of this invention can recognize all types of obstacles, with a 100% recognition rate for non-planar obstacles. The robot inspection and localization error is ≤3cm, and the map outline is complete and without distortion. The traditional single laser SLAM scheme cannot recognize non-planar obstacles, with an inspection and localization error of 15-20cm, serious loop drift during long-distance inspection, and extremely poor mapping accuracy.

[0048] 3. Fault location accuracy test: Based on the multi-feature cross correction algorithm, the fault location error of various types of cable overheating, partial discharge and structural faults is ≤1m; the fault location error of traditional single sensor solution is 3-5m, which cannot accurately locate the fault point and makes maintenance and troubleshooting difficult.

[0049] 4. Hazard warning performance test: The four-level hierarchical early warning system of this invention combines multi-feature coupling analysis and achieves an early warning accuracy rate of 99.2%, with no missed judgments and no false judgments, and can predict potential hazards in advance; the early warning accuracy rate of traditional solutions is only 68.5%, and the hazard missed judgment rate is 21.3%, which cannot achieve early warning.

[0050] 5. Health assessment and decision-making performance test: The CHI quantitative assessment model of this invention has a health status assessment accuracy of 95.6% and an operation and maintenance decision matching rate of 98%, and can accurately predict the aging trend of equipment; traditional solutions do not have quantitative health assessment capabilities and can only rely on human experience to judge, which is highly subjective and has low accuracy.

[0051] 6. Engineering operation and maintenance efficiency test: Compared with the traditional planned maintenance mode, the condition-based maintenance mode of this invention can reduce the operation and maintenance labor cost by 45%, reduce the incidence of sudden cable failures by 62%, and significantly improve the safety and economy of underground cable trench operation and maintenance.

[0052] The quantitative data above show that the performance of each core technology module of this invention is significantly better than that of existing traditional technologies, and the technical solution is reasonable, feasible, and highly applicable to engineering.

[0053] This integrated intelligent sensing platform system for underground cable trenches, based on multi-sensor fusion, adopts a multi-source heterogeneous sensing system for cable trench scenarios. It constructs an edge-cloud collaborative spatiotemporal fusion sensing architecture, integrating three types of heterogeneous sensing devices: fixed multi-parameter sensors, distributed optical fibers, and mobile robot visual thermal imaging. This breaks down the barriers of traditional single-module independent monitoring. Through edge-cloud collaborative mode of unified spatiotemporal reference at edge nodes and deep fusion computing in the cloud, it achieves integrated integration of sensor data from multiple sources, different frequencies, and different physical mechanisms. This improves the integrity, comprehensiveness, and data consistency of cable trench scenario sensing, providing full-dimensional data support for subsequent fault analysis and status assessment. This invention constructs a sub-millisecond-level spatiotemporal synchronization fusion mechanism for three types of data: electrical quantities, environmental quantities, and structural quantities. This mechanism achieves precise alignment of multi-source data, solidifies the data foundation for intelligent sensing, and specifically adapts to the acquisition characteristics of the three types of heterogeneous sensor data. Through timestamp anchoring technology and three-dimensional spatial interpolation algorithms, it achieves precise correlation of all monitoring data under a unified spatiotemporal benchmark, eliminating data analysis distortion and fault misjudgment caused by spatiotemporal deviations. This improves the reliability and accuracy of multi-source data fusion analysis and provides precise physical data support for subsequent feature extraction, fault diagnosis, and health assessment. This invention employs an improved RBPF-SLAM multi-sensor collaborative mapping algorithm to enhance the localization and mapping accuracy in complex cable trench environments. It integrates multiple sensor information from LiDAR, depth camera, ultrasonic sensors, encoders, and gyroscopes, and adds ORB feature extraction functionality from the depth camera. This enables precise identification of non-planar obstacles such as suspended cables and loose stones within the cable trench, solving the problems of single 2D LiDAR's inability to detect non-planar obstacles and excessive loopback feedback positioning errors. Through four-step iterative calculations, the robot's pose and environmental map are optimized in real time, adapting to the special scenarios of narrow, complex cable trenches with numerous irregular obstacles, ensuring the stability and accuracy of mobile sensing equipment operations. This invention constructs a multi-feature cross-validation fault location mechanism and a four-level hierarchical early warning system to improve fault location accuracy and the standardization of operation and maintenance response. It integrates three core fault features: distributed optical fiber temperature field data, vibration fault impact time difference, and partial discharge signal amplitude attenuation. Through time-domain differential correction and spatial interpolation algorithms, it achieves cross-correction of fault points, strictly controlling the fault location error within 1 meter. At the same time, it establishes a standardized four-level hidden danger hierarchical early warning mechanism, which accurately classifies the early warning level based on the characteristics of hidden dangers, parameter deviation, and development trend, and automatically matches the operation and maintenance handling process. This realizes standardized management of the entire process of hidden danger from discovery, investigation, handling to emergency response, and improves the timeliness and professionalism of cable trench fault operation and maintenance.This invention constructs a comprehensive health assessment model for underground cable trenches (CHI), enabling quantitative analysis of equipment health status and promoting the intelligent transformation of operation and maintenance (O&M) models. Sensor weights are determined through joint training with an expert system and historical fault data. A 0-100 point quantitative health assessment system is constructed using a normalized mapping function. Combined with time-series trend analysis, three-level O&M decision-making suggestions and a visualized heatmap are generated. This breaks away from the traditional O&M model of non-quantitative assessment and passive fault repair, accurately predicting the aging and potential development trends of cable equipment. It supports the transformation of power O&M from traditional planned maintenance to precise condition-based maintenance, reducing cable fault rates and O&M costs.

[0054] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A comprehensive intelligent sensing platform system for underground cable trenches based on multi-sensor fusion, comprising a heterogeneous sensing and acquisition module, edge computing nodes, and a cloud service platform, characterized in that: The heterogeneous sensing acquisition module includes a fixed multi-parameter sensor, a distributed fiber optic sensor, and a mobile robot visual thermal imaging sensing unit, which are used to simultaneously acquire three types of heterogeneous monitoring data: electrical quantities, environmental quantities, and structural quantities of the cable trench. The edge computing node is used to perform nanosecond-level timestamp annotation on the multi-source heterogeneous sensing data, register the spatial coordinate system based on the cable trench topology map, and generate a sensing data stream with a unified spatiotemporal label. The cloud service platform incorporates an improved RBPF-SLAM collaborative mapping module, a fault diagnosis and location module, a four-level hierarchical early warning module, and a CHI health assessment module. These modules are used to integrate multiple algorithms to achieve environmental mapping, fault location, hazard early warning, and quantitative assessment of health status. Based on the edge-cloud collaborative architecture, the platform enables intelligent perception and operation and maintenance decision-making for the entire cable trench scenario.

2. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion as described in claim 1, characterized in that: The CHI health assessment module incorporates a cable health index (CHI) assessment model. The calculation formula for the CHI assessment model is as follows: ； The parameters are explained as follows: CHI is the cable health index, with a value range of [0,100]. The larger the value, the better the health status of the cable equipment; n is the total number of sensor types participating in the health assessment; i is the sensor type number, with values ​​of 1, 2, 3...n; wi is the weight coefficient corresponding to the i-th type of sensor, which is obtained by iterative training of the expert system combined with historical cable fault data, and the sum of the weight coefficients is 1. This is the normalization mapping function for the monitoring data of the i-th type of sensor, used to convert the original monitoring data of different dimensions and different numerical ranges into standardized values ​​in the [0,1] interval; This refers to the raw monitoring data collected by the i-th type of sensor.

3. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The improved RBPF-SLAM collaborative mapping module incorporates an improved RBPF-SLAM algorithm, which includes a formula for calculating the joint distribution of observations, used for multi-sensor data fusion modeling. The formula is as follows: ； in, Let be the joint probability distribution of the mobile robot pose and the environment map at time k; represents the continuous pose state variables of the robot from time 1 to k; m represents the constructed grid map model of the cable trench environment. The environmental observation data from multiple sensors at times 1 to k; The odometer control data output by the robot encoder and gyroscope from time 1 to k; Estimating the probability of a map given its pose and observation data; This represents the pose estimation probability based on known observation and control data.

4. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 3, characterized in that: The iterative process of the improved RBPF-SLAM algorithm includes four steps: sampling, weight calculation, resampling, and map estimation. It also integrates the ORB feature extraction function of the depth camera. The ORB feature extraction is used to collect feature vectors of special regions in the image inside the cable trench, identify non-planar obstacles such as suspended cables and loose stones, make up for the detection defects of a single LiDAR, and suppress the positioning error caused by loop feedback.

5. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The fault diagnosis and location module is equipped with a time-domain differential correction algorithm to eliminate time-series deviations in multi-source fault data. The algorithm formula is as follows: ； in, T0 is the time-domain correction deviation value of a single set of fault data; T0 is the original timestamp of the fault sensing data of the current frame; k is the total number of historical data frames involved in the correction calculation; t is the sequence number of the historical data frame; Tt is the timestamp of the historical fault data of the t-th frame.

6. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 5, characterized in that: The fault diagnosis and location module is also equipped with a fault location cross-correction calculation formula for accurately locating cable fault points. The formula is as follows: ； Where L represents the final corrected precise location of the cable fault; L1 represents the fault location calculated based on distributed optical fiber temperature field data; L2 represents the fault location calculated based on the time difference of the fault impact from the vibration sensor; L3 represents the fault location calculated based on the amplitude attenuation law of the partial discharge signal; α, β, and γ are the location correction weights corresponding to the three types of fault characteristics, and satisfy the following conditions: The weights are derived from training and optimization using historical fault location data.

7. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The four-level hierarchical early warning module has a built-in four-level hierarchical early warning system, including prompt level, early warning level, alarm level, and emergency level; the prompt level is when the parameters show a trend deviation and there are no clear fault characteristics, pushing operation and maintenance attention information; The early warning level indicates the presence of potential hazards and automatically generates an inspection work order; the alarm level confirms the existence of potential faults and pushes an on-site handling notification; the emergency level automatically triggers the emergency response process in the event of emergencies such as open flames or water levels exceeding limits.

8. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The cloud service platform generates three-level operation and maintenance decision suggestions—early warning, alarm, and maintenance—based on the CHI health index and time series trend analysis. It also outputs a full-area health heat map of the cable trench, enabling visualized monitoring of cable health status and supporting the transformation of power operation and maintenance from a planned maintenance mode to a condition-based maintenance mode.

9. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The edge computing node is equipped with a three-dimensional spatial interpolation algorithm for accurate registration of spatial coordinates of multi-source sensor data. The algorithm formula is as follows: ； The parameters are defined as follows: S(x,y,z) is the interpolated sensor data at the three-dimensional coordinates (x,y,z) of the cable trench; x, y, and z are the horizontal, vertical, and longitudinal coordinates of the three-dimensional space of the cable trench, respectively; m is the number of effective peripheral sensors involved in the interpolation calculation; j is the serial number of the peripheral sensor. The spatial interpolation weight of the j-th sensor is calculated inversely proportional to the spatial distance between the sensor and the target point. Install coordinates on the j-th sensor. The raw data collected at the location.

10. The integrated intelligent sensing platform system for underground cable trenches based on multi-sensor fusion according to claim 1, characterized in that: The heterogeneous sensing acquisition module collects three types of monitoring data: electrical quantities, environmental quantities, and structural quantities. The electrical quantities include high-frequency partial discharge pulses, sheath circulating current, and cable joint temperatures, with an acquisition frequency of 10kHz-50kHz. The environmental quantities include photoelectric point-type water immersion level, MEMS electrochemical harmful gas concentration, and digital temperature and humidity, with an acquisition frequency of 1kHz-5kHz. The structural quantities include MEMS accelerometer micro-vibration data and fiber optic grating displacement and deformation data, with an acquisition frequency of 2kHz-10kHz. The edge computing node uses timestamp anchoring technology to control the time synchronization error of the three types of heterogeneous frequency data to the sub-millisecond level.

Images