A cable inspection robot detection system and method based on multi-sensor data

Abstract

The application discloses a kind of cable inspection robot detection system and method based on multi-sensor data, including data acquisition module and detection module.Data acquisition module is used to collect multi-sensor data such as bearing weight, inclination, speed change and vibration signal in real time;Detection module includes processor unit and memory unit, and processor unit has the ability to classify sensor data, generate sub-training set, calculate corresponding single kernel function and weight, construct multi-kernel function and output detection results.The application realizes efficient, multi-dimensional abnormal identification and safety state judgment by constructing multi-kernel function, improves the intelligent level and fault detection accuracy of cable inspection robot, and is suitable for power inspection task under complex operating environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable inspection robot detection, and particularly relates to a cable inspection robot detection system and method based on multi-sensor data. BACKGROUND

[0002] The cable inspection robot is an intelligent mobile device for overhead cable inspection, mainly composed of a mobile carrier, a communication module and various detection devices, and has remote control operation and autonomous inspection capabilities. These robots are usually deployed on high-voltage transmission lines to perform infrared thermal imaging temperature measurement, environmental gas detection, visual recognition and other tasks to realize real-time monitoring and defect warning of the line operation state and ensure the stable operation of the power grid system.

[0003] However, most of the current cable inspection robots still have many deficiencies in actual operation. First, the overhead line environment is complex, the robot runs for a long time, and is easily affected by factors such as climate, gravity swing, mechanical fatigue, etc., resulting in safety problems such as device abnormalities such as falling, stalling, and even misjudgment. Second, the traditional fault detection method is mostly based on single sensor signal judgment, lacking fusion analysis of multi-dimensional state information such as bearing weight and vibration response, resulting in poor detection accuracy and stability, making it difficult to meet the long-term stable operation and adaptive safety detection requirements in complex environments.

[0004] In addition, the existing system generally lacks an effective data training mechanism, and a large amount of data collected from multiple sensors cannot be deeply processed and dynamically optimized, making it difficult to build a high-precision judgment model with multi-dimensional parameter weight adjustment capability. This makes the fault recognition and safety judgment of the inspection robot still rely on experience rules, lacking data-driven intelligent decision-making capability.

[0005] Therefore, there is an urgent need for a cable inspection robot detection system that can fuse multi-sensor data and has dynamic calculation and self-learning capabilities to realize more efficient, safer and more intelligent operation state judgment and early warning control. SUMMARY

[0006] In view of the deficiencies in the prior art, the purpose of the present application is to provide a cable inspection robot detection system and method based on multi-sensor data, which realizes efficient and multi-dimensional abnormality recognition and safety state judgment by constructing multi-kernel functions, to improve the fault detection accuracy of the cable inspection robot.

[0007] To achieve the above purpose, the present application provides the following technical solution: a cable inspection robot detection system based on multi-sensor data, comprising:

[0008] A data acquisition module is configured to acquire multi-sensor data detected by the cable inspection robot in real time, wherein the multi-sensor data includes bearing weight, inclination, speed change and vibration signal.

[0009] A detection module is connected to the data acquisition module and includes:

[0010] A processor unit includes:

[0011] A processing subunit is configured to classify each of the multi-sensor data and generate a plurality of sub-training sets corresponding to preset threshold intervals.

[0012] A calculation subunit is connected to the processing subunit and configured to assign a single kernel function to each of the sub-training sets and dynamically calculate a weight of the single kernel function according to data characteristics of each of the sub-training sets, wherein the data characteristics include storage amount, data quantization degree and detection error, the storage amount represents a total amount of data in each of the sub-training sets, and the number of the single kernel functions is equal to the number of the sub-training sets.

[0013] A construction subunit is connected to the calculation subunit and configured to construct a multi-kernel function according to the single kernel function, the weight of the single kernel function, the detection error and the number of the sub-training sets of each of the sub-training sets.

[0014] An execution subunit is connected to the construction subunit and configured to trigger a safety state judgment or an alarm instruction according to an output result of the multi-kernel function.

[0015] A memory unit is connected to the processor unit and configured to store the multi-kernel function after training.

[0016] Further, the processing subunit classifies each of the multi-sensor data into three levels to obtain a first sub-training set, a second sub-training set and a third sub-training set containing the bearing weight, the inclination, the speed change and the vibration signal.

[0017] The data in the first sub-training set satisfies the following conditions: the bearing weight is less than a first preset weight value, the inclination is less than a first preset inclination, the speed change is less than a first acceleration threshold value, and the vibration signal is less than a first amplitude threshold value.

[0018] The data in the second sub-training set satisfies the following conditions: the bearing weight is greater than or equal to the first preset weight value and less than or equal to a second preset weight value, the inclination is greater than or equal to the first preset inclination and less than or equal to a second preset inclination, the speed change is greater than or equal to the first acceleration threshold value and less than or equal to a second acceleration threshold value, and the vibration signal is greater than or equal to the first amplitude threshold value and less than or equal to a second amplitude threshold value.

[0019] The data in the third sub-training set satisfies:

[0020] The bearing weight is greater than the second preset weight value, the inclination is greater than the second preset inclination, the speed change is greater than the second acceleration threshold, and the vibration signal is greater than the second amplitude threshold.

[0021] Further, the calculation formula of the data quantization degree is configured as:

[0022] ;

[0023] wherein, the data quantization degree is represented by, the sum of all the bearing weights corresponding to the sub-training set is represented by; the sum of all the inclinations corresponding to the sub-training set is represented by; the sum of all the speed change values corresponding to the sub-training set is represented by; the sum of all the vibration signals corresponding to the sub-training set is represented by.

[0024] Further, the calculation formula of the detection error is configured as:

[0025] ;

[0026] wherein, the detection error is represented by, the bearing weight predicted by the single-core function is represented by; the bearing weight actually measured by the sensor is represented by; the inclination predicted by the single-core function is represented by; the inclination actually measured by the sensor is represented by; the speed change predicted by the single-core function is represented by; the speed change actually measured by the sensor is represented by; the vibration signal predicted by the single-core function is represented by; the vibration signal actually measured by the sensor is represented by.

[0027] Further, the calculation formula of the multi-core function is configured as:

[0028] ;

[0029] wherein, the input feature is represented by, the single-core function is represented by, the multi-core function is represented by, the single-core function weight is represented by, the number of sub-training sets is represented by, , represents a default initial weight, respectively represent preset first and second positive numbers, represents a preset adjustment parameter greater than zero, respectively represent a weight correction parameter corresponding to the storage amount, a weight correction parameter corresponding to the data quantization degree, and a weight correction parameter corresponding to the detection error, represents the detection error.

[0030] Further, the single kernel function configured by the first sub-training set is a Sigmoid kernel function, the single kernel function configured by the second sub-training set is a RBF kernel function, and the single kernel function configured by the third sub-training set is a Sigmoid kernel function or a Linear kernel function.

[0031] Further, the data acquisition module is further configured to acquire temperature, wind speed, and electromagnetic field intensity in an external environment in real time.

[0032] The processor unit further includes an optimization sub-unit connected to the construction sub-unit, and the optimization sub-unit is configured to process the temperature, the wind speed, and the electromagnetic field intensity to obtain a temperature interference factor, a strong wind interference factor, and an electromagnetic interference factor, and correct the single kernel function weight according to the temperature interference factor, the strong wind interference factor, and the electromagnetic interference factor to obtain a corrected single kernel function weight.

[0033] The construction sub-unit optimizes and adjusts the multi-kernel function according to the corrected single kernel function weight to adjust the output result of the multi-kernel function.

[0034] Further, the formula for calculating the corrected single kernel function weight is configured as:

[0035] ;

[0036] wherein, represents the corrected single kernel function weight, , , respectively represent a temperature interference weight coefficient, a wind speed interference weight coefficient, and an electromagnetic field interference weight coefficient, respectively represent the normalized temperature interference factor, the strong wind interference factor, and the electromagnetic interference factor.

[0037] Further, the calculation sub-unit switches the single kernel function corresponding to the first sub-training set, the second sub-training set, and the third sub-training set according to the temperature interference factor, the strong wind interference factor, and the electromagnetic interference factor, and the switching process includes:

[0038] The temperature interference factor, the strong wind interference factor and the electromagnetic interference factor are subtracted from preset first, second and third interference thresholds to obtain temperature interference difference, strong wind interference difference and electromagnetic interference difference respectively.

[0039] When the temperature interference difference is greater than the strong wind interference difference and the electromagnetic interference difference, the single kernel function corresponding to the first, second and third sub-training sets is switched to a Sigmoid kernel function.

[0040] When the strong wind interference difference is greater than the temperature interference difference and the electromagnetic interference difference, the single kernel function corresponding to the first, second and third sub-training sets is switched to a RBF kernel function.

[0041] When the electromagnetic interference difference is greater than the strong wind interference difference and the temperature interference difference, the single kernel function corresponding to the first, second and third sub-training sets is switched to a Linear kernel function.

[0042] A cable inspection robot detection method based on multi-sensor data, applied to the cable inspection robot detection system based on multi-sensor data, comprising:

[0043] Step S1, the data acquisition module acquires the multi-sensor data detected by the cable inspection robot in real time, and the multi-sensor data includes bearing weight, inclination, speed change and vibration signal;

[0044] Step S2, the processing subunit classifies and processes each multi-sensor data to generate a plurality of sub-training sets corresponding to preset threshold intervals;

[0045] Step S3, the calculation subunit assigns a single kernel function to each sub-training set, and dynamically calculates the weight of the single kernel function according to the data characteristics of each sub-training set, wherein the data characteristics include storage amount, data quantization degree and detection error, the storage amount represents the total amount of data in each sub-training set, and the number of single kernel functions is equal to the number of sub-training sets;

[0046] Step S4, the construction subunit constructs a multi-kernel function according to the single kernel function, the single kernel function weight, the detection error and the number of sub-training sets of each sub-training set, and the execution subunit triggers a safety state judgment or an alarm instruction according to the output result of the multi-kernel function;

[0047] Step S5, the memory unit stores the trained multi-kernel function.

[0048] The beneficial effects of the present application are:

[0049] The present application realizes accurate detection and fault prediction alarm of the running state of the cable inspection robot by introducing a multi-kernel function modeling method, and fusing multi-dimensional sensor data such as bearing weight, inclination, speed change and vibration signals.

[0050] Before generating the multi-kernel function, the present application classifies and processes the multi-sensor data, constructs multiple sub-training sets, and dynamically determines the weight of the corresponding single kernel function based on the data storage amount, data quantization degree and detection error of each sub-training set. This mechanism effectively improves the modeling accuracy and model adaptive ability, thereby significantly reducing the false alarm and missing rate, and enhancing the robustness and accuracy of the detection system in complex power inspection environment.

[0051] Meanwhile, the present application has online real-time monitoring and fault warning capability, can output alarm instructions before failure occurs, effectively avoids the problems of falling, stagnation and other problems caused by abnormal operation of the inspection robot, and guarantees the continuity, safety and efficiency of power inspection operation. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a structural schematic diagram of the cable inspection robot detection system based on multi-sensor data in the present application;

[0053] Figure 2 is a step flow chart of the cable inspection robot detection method based on multi-sensor data in the present application.

[0054] Reference signs: 1, data acquisition module; 2, detection module; 21, processor unit; 211, processing subunit; 212, calculation subunit; 213, construction subunit; 214, execution subunit; 215, optimization subunit; 22, memory unit. DETAILED DESCRIPTION

[0055] The present application will be further described in detail below in combination with the drawings and examples. Wherein the same parts are represented by the same reference signs. It should be noted that the words "front", "back", "left", "right", "up" and "down" used in the following description refer to the directions in the drawings, and the words "bottom surface" and "top surface", "inner" and "outer" refer to the directions towards or away from the geometric center of a particular part.

[0056] Example 1, refer to Figure 1 As the first embodiment of the present application, the embodiment provides a cable inspection robot detection system based on multi-sensor data, which can realize accurate detection and fault prediction alarm of the running state of the cable inspection robot, comprising:

[0057] The data acquisition module 1 is configured to acquire the multi-sensor data detected by the cable inspection robot in real time, and the multi-sensor data includes the bearing weight, the inclination, the speed change and the vibration signal.

[0058] The detection module 2 is connected to the data acquisition module 1 and includes:

[0059] The processor unit 21 includes:

[0060] The processing sub-unit 211 is configured to classify and process the multi-sensor data to generate a plurality of sub-training sets corresponding to preset threshold intervals.

[0061] The calculation sub-unit 212 is connected to the processing sub-unit 211 and is configured to assign a single kernel function to each sub-training set and dynamically calculate the weight of the single kernel function according to the data characteristics of each sub-training set, wherein the data characteristics include the storage amount, the data quantization degree and the detection error, the storage amount represents the total amount of data in each sub-training set, and the number of single kernel functions is equal to the number of sub-training sets.

[0062] The construction sub-unit 213 is connected to the calculation sub-unit 212 and is configured to construct a multi-kernel function according to the single kernel function, the weight of the single kernel function, the detection error and the number of sub-training sets of each sub-training set.

[0063] The execution sub-unit 214 is connected to the construction sub-unit 213 and is configured to trigger a safety state judgment or an alarm instruction according to the output result of the multi-kernel function.

[0064] The memory unit 22 is connected to the processor unit 21 and is configured to store the trained multi-kernel function.

[0065] The working principle of the embodiment 1 is as follows:

[0066] In this embodiment, a set of sensor units are included, which are respectively configured to acquire the bearing weight, the inclination, the speed change and the vibration signal during the operation of the cable inspection robot. Each sensor can be installed at a key position of the robot, such as a main frame, a walking wheel support, an image acquisition module and a driving motor device, so as to obtain the above multi-factor data in real time.

[0067] In the data processing process, the multi-factor data collected by each sensor is first input into the data acquisition module 1 and then transmitted to the connected processor unit 21. The processor unit 21 internally includes the following sub-modules:

[0068] The processing subunit 211 preliminarily classifies the bearing weight, inclination, speed change and vibration signal data according to attributes, and constructs a plurality of sub-training sets respectively. For example, the bearing weight can cover the robot's own weight change, image acquisition device load fluctuation, snow adhesion caused weight change, etc., and the inclination, speed and vibration change can correspond to the robot rack structure instability, drive device failure or walking wheel abnormality and the like.

[0069] The computing subunit 212 assigns a single kernel function to each sub-training set, and dynamically calculates the weight value of each single kernel function according to the data memory amount (i.e. the total number of this type of data) of the corresponding sub-training set, the data quantization degree (data accuracy and distribution characteristics), and the detection accuracy (the ability to distinguish the actual fault of this type of signal in the historical detection).

[0070] The constructing subunit 213 constructs the above-mentioned single kernel functions and their weights in a linear combination manner to form a multi-kernel function model, which has the ability to fuse multiple factors and is suitable for multi-dimensional fault judgment in complex operating environments.

[0071] The executing subunit 214, when the system is in the running mode, inputs the new sensor data as input into the trained multi-kernel function, and the system judges the safety state according to the function output result. If the detection result exceeds the preset threshold range, an alarm instruction is triggered, prompting that there may be a running fault such as falling, instability or equipment abnormality.

[0072] The memory unit 22 is used to store the multi-kernel function model and related training set information constructed in the training stage, and supports subsequent calling and model iteration update.

[0073] The embodiment realizes accurate monitoring and fault prediction of the inspection robot state by classifying and modeling multi-source sensor data, multi-kernel fusion and weight adaptive allocation. Compared with the traditional method of relying on single signal threshold judgment, the system significantly improves the false alarm rate control ability and real-time responsiveness, and is particularly suitable for long-term continuous operation scenes such as power overhead cables.

[0074] Preferably, the processing subunit 211 classifies the multi-sensor data into three levels to obtain a first sub-training set, a second sub-training set and a third sub-training set containing bearing weight, inclination, speed change and vibration signal;

[0075] The data in the first sub-training set satisfies: the bearing weight is less than a first preset weight value, the inclination is less than a first preset inclination, the speed change is less than a first acceleration threshold, and the vibration signal is less than a first amplitude threshold;

[0076] The data in the second sub-training set satisfy: the bearing weight is greater than or equal to a first preset weight value and less than or equal to a second preset weight value, the inclination is greater than or equal to a first preset inclination and less than or equal to a second preset inclination, the speed change is greater than or equal to a first acceleration threshold and less than or equal to a second acceleration threshold, and the vibration signal is greater than or equal to a first amplitude threshold and less than or equal to a second amplitude threshold.

[0077] The data in the third sub-training set satisfy:

[0078] The bearing weight is greater than the second preset weight value, the inclination is greater than the second preset inclination, the speed change is greater than the second acceleration threshold, and the vibration signal is greater than the second amplitude threshold.

[0079] Specifically, in the embodiment, the processing manner of the processing subunit 211 to the multi-sensor data is further optimized, a three-level classification strategy is adopted to refine and classify the bearing weight, the inclination, the speed change and the vibration signal, thereby constructing the first sub-training set, the second sub-training set and the third sub-training set, so as to improve the training precision and the prediction accuracy of the subsequent multi-kernel function.

[0080] To realize reasonable hierarchical classification processing, the following typical thresholds are selected in the embodiment (which can be adjusted according to the application environment):

[0081] The first preset weight value is 50 kg, and the second preset weight value is 100 kg; the first preset inclination is 16.7°, and the second preset inclination is 30°.

[0082] The first acceleration threshold and the second acceleration threshold correspond to the lower limit and the upper limit of the speed change respectively, and the unit is m / s².

[0083] The first amplitude threshold and the second amplitude threshold correspond to the vibration amplitude boundary of the robot in the normal and abnormal running states respectively, and the unit is mm / s².

[0084] After the data acquisition module 1 collects the multi-factor signals of the inspection robot in the running process in real time, the processing subunit 211 classifies and processes various data according to the following rules:

[0085] The first sub-training set contains the working condition data of the stable and safe running state, and satisfies the following conditions: the bearing weight < 50 kg; the inclination < 16.7°; the speed change < the first acceleration threshold; and the vibration signal < the first amplitude threshold.

[0086] The second sub-training set contains the running state data in the tolerable fluctuation range, and satisfies the following conditions: 50 kg ≤ the bearing weight ≤ 100 kg; 16.7° ≤ the inclination ≤ 30°; the first acceleration threshold ≤ the speed change ≤ the second acceleration threshold; and the first amplitude threshold ≤ the vibration signal ≤ the second amplitude threshold.

[0087] Third sub-training set: containing the working condition data that may have running risks or equipment abnormalities, meeting the following conditions: bearing weight > 100 kg; inclination > 30°; speed change > second acceleration threshold; vibration signal > second amplitude threshold.

[0088] Through the three-level classification method described in this embodiment, the discrimination and label accuracy of the training data can be effectively improved, which helps the adaptability and generalization ability of the single kernel function in the training process. Further, when constructing the multi-kernel function, the system can perform fine modeling based on the characteristics of each sub-training set, thereby improving the accuracy, stability and response speed of fault prediction.

[0089] The classification method is particularly suitable for inspection scenes in complex or high-risk operating environments, such as high wind, snow accumulation or high load inspection conditions. By dynamically identifying "edge state" and "risk state" data, safety warnings can be triggered in time, effectively ensuring the operation reliability of the cable inspection robot.

[0090] Preferably, the calculation formula of the data quantization degree is configured as:

[0091] ;

[0092] Wherein, represents the data quantization degree, represents the sum of all bearing weights in the corresponding sub-training set; represents the sum of all inclinations in the corresponding sub-training set; represents the sum of all speed changes in the corresponding sub-training set; represents the sum of all vibration signals in the corresponding sub-training set.

[0093] The data quantization degree reflects the overall activity and expression ability of the data signals in the sub-training set, which serves as the initial "weight basis" for each single kernel function when combined in a multi-kernel.

[0094] Preferably, the calculation formula of the detection error is configured as:

[0095] ;

[0096] Wherein, represents the detection error, represents the bearing weight predicted by the single kernel function; represents the bearing weight actually measured by the sensor; represents the inclination predicted by the single kernel function; represents the inclination actually measured by the sensor; represents the speed change predicted by the single kernel function; represents the speed change actually measured by the sensor; a vibration signal representing a single kernel function prediction; a vibration signal representing a sensor actual measurement.

[0097] Specifically, in the embodiment, the kernel function parameters are optimized by detecting errors, and the weight distribution of the single kernel function in the final multi-kernel function is adjusted accordingly, so that the high-precision sub-model has higher influence.

[0098] By introducing the above two calculation formulas, the system in the embodiment can achieve:

[0099] The multi-source data intensity under different training subsets is effectively characterized, avoiding the interference of the "weak information group" on model training; the single kernel function is dynamically optimized through error feedback, enhancing the model prediction ability; more reasonable multi-kernel function weight configuration is realized, improving the overall judgment accuracy of the system and reducing the false alarm and missed alarm probability.

[0100] Actual tests show that under high dynamic and complex working conditions, the strategy significantly improves the state recognition sensitivity and alarm triggering timeliness of the inspection robot, ensuring long-term reliable operation of the equipment.

[0101] Preferably, the calculation formula of the multi-kernel function is configured as:

[0102] ;

[0103] wherein, represents an input feature, represents a single kernel function, represents a multi-kernel function, represents a single kernel function weight, represents the number of sub-training sets, , represents a default initial weight, respectively represent a preset first positive number and a second positive number, represents a preset adjustment parameter greater than zero, respectively represent a weight correction parameter corresponding to the storage amount, a weight correction parameter corresponding to the data quantization degree, and a weight correction parameter corresponding to the detection error, represents a detection error.

[0104] Specifically, in the embodiment, the system initially allocates a default initial weight to each single kernel function, and the value of the initial weight is 1 by default, indicating that each single kernel function contributes equally to the detection result. During the training or detection process, the single kernel function weight is dynamically corrected according to the detection error, storage amount, and data quantization degree of the single kernel function, and is dynamically corrected through the following weight adjustment formula, thereby changing the output of the multi-kernel function.

[0105] The output of the multi-core function can be used for one of the following two determination methods:

[0106] Method one: if the multi-core function is a numerical output, compare it with the preset alarm threshold, and if it is greater than the preset alarm threshold, trigger the alarm;

[0107] Method two: if the multi-core function output is a label type (such as "safe / unsafe"), directly decide whether to trigger the alarm instruction according to it.

[0108] The weight adjustment mechanism introduced in this embodiment can make the detection model dynamically adapt to changes in the operating environment according to the prediction performance of each type of data. Especially in the case of sudden interference, structural loss or complex load fluctuation, this method can effectively reduce the false alarm rate caused by a single signal misjudgment, and improve the response accuracy and robustness of the overall system. For example, in a rainy and snowy environment, the load change caused by the increased weight of the robot due to snow may affect the output of some single-core functions. At this time, the system adjusts the relevant weights in real time to ensure that the overall judgment still has high stability.

[0109] This embodiment further verifies the effectiveness of the multi-core function detection method based on the adjustable weight mechanism in actual inspection tasks, and designs and implements a comparative experiment to compare the performance differences of the fixed weight strategy and the adjustable weight strategy in false alarm control and detection accuracy.

[0110] I. Experimental design:

[0111] Purpose of the experiment: to verify the effect of the adjustable weight mechanism of the single-core function on the improvement of the detection accuracy and the false alarm rate.

[0112] Experimental object: an intelligent cable inspection robot deployed on a certain ultra-high voltage transmission line.

[0113] Experimental scenario: set as a regular inspection task of key lines in a substation, including common abnormal scenarios such as load fluctuation, weather change and equipment aging.

[0114] II. Experimental group setting:

[0115] Control group (fixed weight): the weight of each single-core function is fixedly set to =1, i.e. equal weight, and each single-core function satisfies the constraint .

[0116] Experimental group (adjustable weight): the weight is dynamically adjusted according to the characteristics of the sub-training set, supporting error-driven and memory-sensitive adjustment mechanisms, i.e. .

[0117] The experimental group uses to realize memory-sensitive adjustment (where e=0.2 is the best), and also designs an error compensation mechanism: uses The single-core function detection error is increased by 1%, and the weight is automatically corrected by 0.15 The attenuation factor prevents error accumulation from causing overall judgment deviation.

[0118] III. Result comparison and effect verification

[0119] In a continuous running test period of one month, the fixed weight group and the adjustable weight group are run under the same input data conditions, and the experimental results are shown in Table 1:

[0120] Table 1

[0121]

[0122] Through the comparison of the data in Table 1, it is found that the adjustable weight strategy in the embodiment can reduce the false positive rate by about 34%, and performs more acutely in the response ability to sudden abnormalities, and is suitable for deployment in the ultra-high voltage line inspection task with extremely high real-time and reliability requirements.

[0123] Preferably, the single-core function configured by the first sub-training set is a Sigmoid kernel function, the single-core function configured by the second sub-training set is an RBF kernel function, and the single-core function configured by the third sub-training set is a Sigmoid kernel function or a Linear kernel function.

[0124] Embodiment 2, which is the second embodiment of the present application, is different from the previous embodiment in that the embodiment provides an optimization subunit 215, which can realize real-time optimization of the single-core function weight by introducing external environmental parameters, thereby realizing adaptive adjustment of the multi-core function model and improving the detection accuracy and stability of the system in complex natural environments. The data acquisition module 1 is also used to collect the temperature, wind speed and electromagnetic field intensity in the external environment in real time;

[0125] The processor unit 21 also includes an optimization subunit 215 connected to the construction subunit 213, and the optimization subunit 215 is used to process the temperature, wind speed and electromagnetic field intensity to obtain a temperature interference factor, a strong wind interference factor and an electromagnetic interference factor, and correct the single-core function weight according to the temperature interference factor, the strong wind interference factor and the electromagnetic interference factor to obtain a corrected single-core function weight;

[0126] The construction subunit 213 optimizes and adjusts the multi-core function according to the corrected single-core function weight to adjust the output result of the multi-core function.

[0127] Working principle of embodiment 2:

[0128] On the basis of the existing detection system based on multi-sensor data, the embodiment adds an environmental interference factor identification and dynamic correction mechanism, and realizes real-time optimization of the single-core function weight by introducing external environmental parameters, thereby realizing adaptive adjustment of the multi-core function model.

[0129] In this embodiment, the data acquisition module 1 is extended to integrate three types of environmental sensors for collecting:

[0130] Temperature: The environmental temperature sensor is installed outside the robot body;

[0131] Wind speed: The airflow changes in the robot operating area are detected in real time by the wind speed measurement module;

[0132] Electromagnetic field intensity: A small electromagnetic induction probe is used to detect the electromagnetic interference level under the surrounding high-voltage line.

[0133] These environmental parameters are closely related to the robot operating state. Under the conditions of excessive wind speed, abnormal electromagnetic interference, or extreme temperature changes, the sensor signals and detection judgments are prone to deviation.

[0134] The newly added optimization sub-unit 215 function in this embodiment is:

[0135] Calculate the environmental interference factor:

[0136] The optimization sub-unit 215 compares the collected environmental parameters with the preset safety range to obtain three interference factors:

[0137] Temperature interference factor: calculated according to the deviation of the current temperature from the normal operating temperature;

[0138] Strong wind interference factor: calculated according to the degree of wind speed exceeding the safe operating threshold;

[0139] Electromagnetic interference factor: calculated according to the degree of abnormality of the current electromagnetic field intensity.

[0140] The original weight of each single-core function will be dynamically adjusted based on the interference factor. The corrected single-core function weight calculation formula is configured as:

[0141] ;

[0142] where, represents the corrected single-core function weight, , , temperature interference weight coefficient, wind speed interference weight coefficient, and electromagnetic field interference weight coefficient (obtained according to experience or training), respectively represent the normalized temperature interference factor, strong wind interference factor, and electromagnetic interference factor, with a value range of (0, 1). When the environmental interference increases, the weight of the corresponding core function will be appropriately reduced to prevent the training data affected by noise from interfering with the final judgment.

[0143] The construction subunit reconstructs the final multi-kernel function output after obtaining the corrected single-kernel function weight:

[0144] ;

[0145] The optimized output result is used for finally determining the robot running state, and can effectively improve the system robustness and false alarm suppression capability in the case of extreme environment or severe fluctuation.

[0146] The comparative test under the simulation of the summer high temperature + strong wind + electromagnetic interference comprehensive scene shows that:

[0147] The false alarm rate of the uncorrected model in the interference scene is 15.6%;

[0148] The false alarm rate of the optimized model in embodiment 2 is reduced to 8.9%, and the error is reduced by about 43%;

[0149] The response time of the key abnormality (such as off-track, abnormal load) is advanced by 1.2 seconds.

[0150] The above results prove that the embodiment significantly enhances the adaptability, accuracy and real-time performance of the detection system under the condition of multi-source interference through the environment perception and weight correction mechanism, and is particularly suitable for deployment in high-voltage power transmission inspection tasks with complex climate and severe electromagnetic environment.

[0151] Preferably, the calculation subunit 212 switches the single-kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set according to the temperature interference factor, the strong wind interference factor and the electromagnetic interference factor, and the switching process includes:

[0152] The temperature interference difference, the strong wind interference difference and the electromagnetic interference difference are obtained by subtracting the preset first interference threshold, the second interference threshold and the third interference threshold from the temperature interference factor, the strong wind interference factor and the electromagnetic interference factor, respectively;

[0153] When the temperature interference difference is greater than the strong wind interference difference and the electromagnetic interference difference, the single-kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to the Sigmoid kernel function;

[0154] When the strong wind interference difference is greater than the temperature interference difference and the electromagnetic interference difference, the single-kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to the RBF kernel function;

[0155] When the electromagnetic interference difference is greater than the strong wind interference difference and the temperature interference difference, the single-kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to the Linear kernel function.

[0156] Specifically, in this embodiment, to further improve the model expression ability and fault tolerance of the detection system under complex environmental changes, a single-core function adaptive switching mechanism is added in the detection system based on multi-sensor data. This mechanism dynamically switches the single-core function type corresponding to each sub-training set according to the environmental interference factors by the calculation sub-unit 212, realizing more efficient and environment-aware multi-core function model construction.

[0157] The environment-aware sensors integrated in the system collect temperature, wind speed, and electromagnetic field strength information in real time, generating three interference factors: temperature interference factor, strong wind interference factor, and electromagnetic interference factor. Each interference factor is respectively calculated by subtracting its corresponding interference threshold value, and three interference differences are obtained: ;

[0158] Among them, are the preset upper threshold values of temperature, wind speed, and electromagnetic interference, respectively.

[0159] Single-core function adaptive switching mechanism:

[0160] According to the size relationship of the three interference differences, the single-core function type corresponding to all sub-training sets (first, second, and third sub-training sets) is automatically switched:

[0161] When the temperature interference is dominant ( largest):

[0162] Sigmoid kernel function is enabled: suitable for processing temperature-related nonlinear disturbances (such as structural changes caused by component thermal expansion, signal drift caused by thermal drift).

[0163] When strong wind interference is dominant ( largest):

[0164] RBF (Radial Basis Function) kernel function is enabled: suitable for processing high-speed dynamic disturbances, and can effectively deal with wind-induced oscillation, acceleration change, and tilt anomaly.

[0165] When electromagnetic interference is dominant ( largest):

[0166] Linear kernel function is enabled: to deal with stable deviation changes caused by electromagnetic induction, and to maintain stability of electromagnetic drift using linear modeling.

[0167] Function switching example:

[0168] ;

[0169] Based on the above data calculation, we can get: , , . Since Therefore, it is determined that temperature disturbance is the main influencing factor, and the single-core function bound to the three types of sub-training sets is immediately switched to a Sigmoid kernel function to enhance the model's recognition and adaptation ability to temperature disturbance.

[0170] The actual deployment in a certain ultra-high voltage line in a plateau area with a severe environmental change inspection robot system shows that under different dominant interference conditions (such as high temperature + breeze, electromagnetic disturbance + low temperature, etc.), the kernel function switching can effectively reduce the false matching; compared with the fixed kernel function model, the detection false alarm rate of the dynamic kernel function switching mechanism is reduced by about 26%, and the key fault recognition rate is increased by more than 17%.

[0171] A cable inspection robot detection method based on multi-sensor data is applied to the above-mentioned cable inspection robot detection system based on multi-sensor data, referring to Figure 2 , comprising:

[0172] Step S1, the data acquisition module 1 acquires the multi-sensor data detected by the cable inspection robot in real time, and the multi-sensor data includes bearing weight, inclination, speed change and vibration signal;

[0173] Step S2, the processing subunit 211 classifies and processes each of the multi-sensor data to generate a plurality of sub-training sets corresponding to a preset threshold interval;

[0174] Step S3, the calculation subunit 212 assigns a single-core function to each of the sub-training sets, and dynamically calculates the weight of the single-core function according to the data characteristics of each of the sub-training sets, the data characteristics including storage capacity, data quantization degree and detection error, the storage capacity representing the total amount of data in each of the sub-training sets, and the number of single-core functions being equal to the number of sub-training sets;

[0175] Step S4, the construction subunit 213 constructs a multi-core function according to the single-core function, the single-core function weight, the detection error and the number of sub-training sets of each of the sub-training sets, and the execution subunit 214 triggers a safety state judgment or an alarm instruction according to the output result of the multi-core function;

[0176] Step S5, the memory unit 22 stores the trained multi-core function.

[0177] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments, and any technical solution falling within the scope of the present application shall be considered as falling within the protection scope of the present application. It should be noted that for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application are also considered as the protection scope of the present application.

Claims

1. A multi-sensor data based cable inspection robot detection system, comprising: The application relates to a cable inspection robot, which comprises a data acquisition module (1) for acquiring multi-sensor data detected by the cable inspection robot in real time, wherein the multi-sensor data comprises bearing weight, inclination, speed change and vibration signal; a detection module (2) connected to the data acquisition module (1) and comprising a processor unit (21) comprising a processing subunit (211) for classifying each multi-sensor data and generating a plurality of sub-training sets corresponding to preset threshold intervals; a calculation subunit (212) connected to the processing subunit (211) and used for assigning a single kernel function to each sub-training set and dynamically calculating a single kernel function weight according to data characteristics of each sub-training set, wherein the data characteristics comprise storage amount, data quantization degree and detection error, the storage amount represents the total amount of data in each sub-training set, and the number of single kernel functions is equal to the number of sub-training sets; a calculation formula of the data quantization degree is configured as: a construction subunit (213) connected to the calculation subunit (212) and used for constructing a multi-kernel function according to the single kernel function, the single kernel function weight, the detection error and the number of sub-training sets of each sub-training set; a calculation formula of the multi-kernel function is configured as: an execution subunit (214) connected to the construction subunit (213) and used for triggering a safety state judgment or an alarm instruction according to an output result of the multi-kernel function; and a memory unit (22) connected to the processor unit (21) and used for storing the trained multi-kernel function. The processing subunit (211) classifies each multi-sensor data into three levels to obtain a first sub-training set, a second sub-training set and a third sub-training set containing the bearing weight, the inclination, the speed change and the vibration signal; the data in the first sub-training set satisfies the conditions that the bearing weight is less than a first preset weight value, the inclination is less than a first preset inclination, the speed change is less than a first acceleration threshold value and the vibration signal is less than a first amplitude threshold value; the data in the second sub-training set satisfies the conditions that the bearing weight is greater than or equal to the first preset weight value and less than or equal to a second preset weight value, the inclination is greater than or equal to the first preset inclination and less than or equal to a second preset inclination, the speed change is greater than or equal to the first acceleration threshold value and less than or equal to a second acceleration threshold value, and the vibration signal is greater than or equal to the first amplitude threshold value and less than or equal to a second amplitude threshold value; and the data in the third sub-training set satisfies the conditions that the bearing weight is greater than the second preset weight value, the inclination is greater than the second preset inclination, the speed change is greater than the second acceleration threshold value and the vibration signal is greater than the second amplitude threshold value. The calculation formula of the detection error is configured as: the single kernel function configured for the first sub-training set is a Sigmoid kernel function, the single kernel function configured for the second sub-training set is a RBF kernel function, and the single kernel function configured for the third sub-training set is a Sigmoid kernel function or a Linear kernel function. ​ ​ ​ ​ ； wherein, represents the data quantification degree, represents the sum of all the bearing weights corresponding to the sub-training set; represents the sum of all the inclinations corresponding to the sub-training set; represents the sum of all the speed change values corresponding to the sub-training set; represents the sum of all the vibration signals corresponding to the sub-training set; ​ ​ ； wherein, denotes an input feature, denotes the single-core function, denotes the multi-core function, denotes the single-core function weight, denotes the number of the sub-training sets, , denotes a default initial weight, respectively denote a preset first positive number and a second positive number, denotes a preset adjustment parameter greater than zero, respectively denote a weight correction parameter corresponding to the storage amount, a weight correction parameter corresponding to the data quantization degree, and a weight correction parameter corresponding to the detection error, denotes the detection error; ​ ​ 2. The multi-sensor data based cable inspection robot detection system of claim 1, wherein: ​ ​ ​ ​ ​ 3. The multi-sensor data based cable inspection robot detection system of claim 1, wherein: ​ ； wherein, represents the load weight predicted by the single core function; represents the load weight actually measured by the sensor; represents the inclination predicted by the single core function; represents the inclination actually measured by the sensor; represents the speed change predicted by the single core function; represents the speed change actually measured by the sensor; represents the vibration signal predicted by the single core function; represents the vibration signal actually measured by the sensor.

4. The multi-sensor data based cable inspection robot detection system of claim 2, wherein: ​ 5. The multi-sensor data based cable inspection robot detection system of claim 1, wherein: The data acquisition module (1) is also used for collecting temperature, wind speed and electromagnetic field intensity in an external environment in real time; The processor unit (21) further comprises an optimization sub-unit (215) connected with the construction sub-unit (213), and the optimization sub-unit (215) is used for processing temperature interference factors, strong wind interference factors and electromagnetic interference factors according to the temperature, the wind speed and the electromagnetic field intensity, and correcting the single kernel function weight to obtain a corrected single kernel function weight according to the temperature interference factors, the strong wind interference factors and the electromagnetic interference factors; The construction sub-unit (213) optimizes and adjusts the multi-kernel function according to the corrected single kernel function weight, so as to adjust the output result of the multi-kernel function.

6. The multi-sensor data based cable inspection robot detection system of claim 5, wherein: The formula of the corrected single kernel function weight is configured as: ； wherein, represents the corrected single core function weight, , , respectively represent a temperature interference weight coefficient, a wind speed interference weight coefficient, and an electromagnetic field interference weight coefficient, respectively represent the normalized temperature interference factor, the strong wind interference factor, and the electromagnetic interference factor.

7. The multi-sensor data based cable inspection robot detection system of claim 5, wherein: The calculation sub-unit (212) switches the single kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set according to the temperature interference factors, the strong wind interference factors and the electromagnetic interference factors, and the switching process comprises: The temperature interference difference value, the strong wind interference difference value and the electromagnetic interference difference value are obtained by subtracting preset first interference threshold values, second interference threshold values and third interference threshold values from the temperature interference factors, the strong wind interference factors and the electromagnetic interference factors respectively; When the temperature interference difference value is greater than the strong wind interference difference value and the electromagnetic interference difference value, the single kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to a Sigmoid kernel function; When the strong wind interference difference value is greater than the temperature interference difference value and the electromagnetic interference difference value, the single kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to a RBF kernel function; When the electromagnetic interference difference value is greater than the strong wind interference difference value and the temperature interference difference value, the single kernel function corresponding to the first sub-training set, the second sub-training set and the third sub-training set is switched to a Linear kernel function. 8.A method for detecting a cable inspection robot based on multi-sensor data, applied to the system for detecting the cable inspection robot based on multi-sensor data in any one of claims 1-7, characterized in that, Comprise: Step S1, the data acquisition module (1) collects the multi-sensor data detected by the cable inspection robot in real time, and the multi-sensor data comprises bearing weight, inclination, speed change and vibration signal; Step S2, the processing sub-unit (211) classifies and processes each multi-sensor data to generate a plurality of sub-training sets corresponding to preset threshold intervals; Step S3, the calculation sub-unit (212) allocates a single kernel function to each sub-training set, and dynamically calculates a single kernel function weight according to the data characteristics of each sub-training set, wherein the data characteristics comprise storage amount, data quantization degree and detection error, the storage amount represents the total amount of data in each sub-training set, and the number of single kernel functions is equal to the number of sub-training sets; Step S4, the construction sub-unit (213) constructs a multi-kernel function according to the single kernel function, the single kernel function weight, the detection error and the number of sub-training sets of each sub-training set, and the execution sub-unit (214) triggers a safety state judgment or an alarm instruction according to the output result of the multi-kernel function. Step S5, the memory unit (22) stores the trained multi-kernel function.

Images