Patrol detection system of intelligent substation circuit device

Abstract

The invention discloses a patrol detection system of an intelligent substation circuit device, and relates to the field of patrol detection of the intelligent substation circuit device, the patrol detection system comprises a patrol detection frame, the patrol detection frame comprises a detection unit, a patrol unit and an analysis unit, the detection unit is used for carrying out defect identification and voiceprint detection on an intelligent substation circuit, and the patrol unit is used for carrying out patrol detection; the detection unit comprises a voiceprint detection module and a voiceprint positioning module. The voiceprint detection module is used for obtaining voiceprint signals of the intelligent substation and the circuit in real time. Through the arrangement mode that the voiceprint detection module, the acousto-optic fusion module and the analysis unit are matched with one another and the voiceprint detection technology, the problems that abnormal voiceprint features of potential faults cannot be found in time when transformer substation equipment is manually inspected and heard are solved, meanwhile, fusion of multiple physical signals such as voiceprint, vibration and an electromagnetic field is facilitated, and the safety of transformer substation equipment is improved. And a cross-modal correlation model is constructed in combination with the graph neural network, so that compound faults can be distinguished accurately.

Description

Technical Field

[0001] The present invention relates to the field of patrol detection of circuit devices in intelligent substations, and in particular to a patrol detection system for circuit devices in intelligent substations. Background Art

[0002] Smart substations are usually a vital component of the power system. Smart substations are mainly responsible for the conversion of voltage and current, as well as the reception, distribution and transmission of electric energy. The internal equipment of smart substations is diverse, and the electrical equipment in smart substations is divided into primary equipment and secondary equipment.

[0003] The circuits of smart substations are usually very important. In order to increase the safety of the circuits in smart substations, staff usually conduct patrol inspections. Usually, staff are required to use handheld light and infrared equipment and experienced on-site listening to conduct fixed-point patrol inspections. This method is mainly used to judge obvious abnormalities, which will increase the labor intensity of staff and affect the accuracy of judgment. In addition, during the window period outside of periodic inspections, the abnormal voiceprint characteristics of potential faults cannot be discovered in time. At the same time, although the existing technology uses multi-source data collection, it only fuses features through simple splicing or arithmetic operations, and lacks cross-modal correlation modeling.

[0004] At the same time, before a serious accident occurs in the power equipment of a smart substation, there are often latent signals. Therefore, voiceprint collection coverage is used to remotely and safely collect and monitor the circuits of the smart substation to solve the problem that the abnormal voiceprint characteristics of potential faults cannot be discovered in time during manual inspections and listening of the circuits of the smart substation. At the same time, it is necessary to integrate multiple physical signals for cross-modal correlation. Summary of the Invention

[0005] The object of the present invention is to provide a patrol detection system for circuit devices of an intelligent substation to solve the problems raised in the above background technology.

[0006] To achieve the above-mentioned object, the present invention provides the following technical solution: a patrol detection system for a circuit device of an intelligent substation, comprising a patrol detection framework, wherein the patrol detection framework comprises:

[0007] A detection unit, which is used to identify defects in the smart substation circuit and perform voiceprint detection, and includes:

[0008] A voiceprint detection module, which is used to obtain voiceprint signals of smart substations and circuits in real time;

[0009] A voiceprint positioning module, which is used to determine the spatial location of the abnormal sound source;

[0010] An inspection unit, which is used to perform inspection management, autonomous inspection optimization, and dynamic risk assessment of smart substation circuits;

[0011] An analysis unit is used to fuse multiple physical signals, build a multi-dimensional state assessment model, and perform data analysis and storage.

[0012] Preferably, the voiceprint detection module deploys a non-contact voiceprint sensor array to cover the three-axis orientation of the substation. The voiceprint positioning module calculates the source coordinates by combining the phase difference of the voiceprint sensor array and the TDOA-based beamforming algorithm. The voiceprint detection module is connected to the voiceprint positioning module through data transmission. The abnormal sound sources determined by the voiceprint positioning module include local discharge points and gas leakage points.

[0013] Preferably, the formula of the TDOA-based beamforming algorithm is as follows:

[0014]

[0015] Among them, T ij It refers to the arrival time difference of the signals from microphones i and j, d i Refers to the straight-line distance from the sound source to microphone i, d j Refers to the straight-line distance from the sound source to microphone j, d i -d j is the sensor spacing, and c is the speed of sound.

[0016] Preferably, the detection unit further includes:

[0017] A defect recognition module, which is used to accurately classify multiple types of typical defects;

[0018] The acoustic-optical fusion module is used to combine the SIFT feature matching of the acoustic image and the visible light image to realize the AR superposition display of the sound field distribution and the real scene of the equipment, thereby visually displaying the sound source distribution and the real scene of the equipment. The voiceprint positioning module is connected to the defect recognition module through data transmission. The acoustic-optical fusion module is connected to the acoustic-optical fusion module through data transmission. The AR superposition display realizes the fusion display of the sound field and the real scene through the camera calibration matrix. The camera calibration matrix formula is as follows:

[0019] H=K[R|t]

[0020] Where H is the projection matrix, which is used to convert 3D world coordinates into 2D image pixel coordinates; K is the intrinsic parameter matrix, which is used to describe the internal optical properties of the camera; R is the rotation matrix, which is used to indicate the camera orientation; t is the translation vector, which is used to indicate the camera position; and [R|t] is the extrinsic parameter matrix, which is used to convert points in the world coordinate system into the camera coordinate system.

[0021] Preferably, the inspection unit includes:

[0022] Inspection management module, which is used to manage inspection data and records, and provide decision support by combining sound source location and fault alarm information;

[0023] An autonomous inspection optimization module, which is used to dynamically adjust sensor deployment strategies to optimize monitoring coverage and energy consumption balance;

[0024] A dynamic risk assessment module is used to predict the probability of failure and the remaining life based on the evolution law of the equipment voiceprint characteristics. The dynamic risk assessment module adopts a conditional variational autoencoder and a survival analysis model. The inspection management module is connected to the autonomous inspection optimization module through data interaction, and the dynamic risk assessment module is connected to the detection unit through data transmission.

[0025] Preferably, the autonomous inspection optimization module adopts deep deterministic policy gradient to define the state space, action space and reward function, which refer to the sensor position and environmental noise, movement step size and angle, coverage area and signal-to-noise ratio respectively. The Q value update formula included in the autonomous inspection optimization module is as follows:

[0026] Q(s,a)=E[r+rQ(s′,μ(s′))]

[0027] Among them, Q(s,a) refers to the action value function, which represents the expected cumulative reward after performing action a in state s, E refers to the expectation operator, which represents the statistical average of subsequent states and rewards, r refers to the immediate reward, γ is the discount factor, s′ refers to the state at the next moment, which is generated by the environment dynamics model or the experience replay pool, and μ(s′) refers to the target policy function, which determines the optimal action to be performed in state s′.

[0028] Preferably, the conditional variational autoencoder is used to extract the potential distribution of voiceprint features and generate a fault evolution path. The survival analysis model is used to combine historical fault data to quantify time-related risks. The risk function formula included in the dynamic risk assessment module is as follows:

[0029] λ(t|X)=λ0(t)exp(β T X)

[0030] Where X is the fingerprint feature vector, β is the weight parameter, which indicates the intensity of the influence of feature X on the risk, λ(t|X) is the risk function, which indicates the probability of instantaneous failure of the equipment under the condition of given feature X within time t, λ0(t) is the baseline risk function, which indicates the basic risk related only to practice, exp(β T X) refers to the risk ratio factor, which represents the scaling factor of the characteristic to the benchmark risk.

[0031] Preferably, the analysis unit comprises:

[0032] A multi-physics field coupling analysis module, which is used to fuse multiple physical signals and build a multi-dimensional equipment status assessment model;

[0033] An electromagnetic-acoustic coupling analysis module, which combines electromagnetic field simulation with acoustic signals to locate mechanical vibration anomalies caused by electromagnetic interference. The module uses a finite element-boundary element coupling algorithm to solve the coupling effect of Maxwell's equations and the acoustic wave equation;

[0034] The cloud data analysis and storage module is used to upload real-time monitoring data to the cloud, perform big data analysis, explore potential failure modes and trends, and provide data support for system optimization. The multi-physics field coupling analysis module and the electromagnetic-acoustic coupling analysis module are both connected to the cloud data analysis and storage module through data transmission.

[0035] Preferably, the multiple physical signals fused by the multi-physics coupling analysis module are voiceprint, infrared thermal imaging and vibration signals. The data fusion of the multi-physics coupling analysis module uses a graph neural network to map the acoustic signal, thermal imaging and vibration into nodes and edges in a graph structure to capture cross-modal correlations. The formula included in the data fusion of the multi-physics coupling analysis module is as follows:

[0036]

[0037] in, refers to the feature vector of node v in the l+1 layer, σ refers to the nonlinear activation function, which introduces the nonlinear expression ability of the model, N(v) refers to the set of neighbor nodes of node v, and W l refers to the learnable weight matrix of layer l, It refers to the feature vector of neighbor node u at layer l, and |N(v)| refers to the number of neighbors of node v, which is used to normalize the aggregation results.

[0038] Preferably, the electromagnetic-acoustic coupling analysis module uses the electromagnetic-acoustic coupling equation to calculate the vibration spectrum characteristics caused by eddy current damage to the iron core, and the formula is as follows:

[0039] ▽×(ν▽×Α)-ω 2 εΑ=J s +σΕ

[0040] Among them, ▽× refers to the curl operator, which is used to describe the rotation characteristics of the vector field, v refers to the magnetic resistivity, which characterizes the material's ability to block the magnetic field, Α refers to the magnetic vector potential, which is used to describe the magnetic potential distribution in the electromagnetic field, ω refers to the angular frequency, ε refers to the dielectric constant, which characterizes the polarization characteristics of the material in the electric field, and J s refers to the source current density, σ refers to the conductivity, which is used to describe the conductivity of the material, and E refers to the electric field strength.

[0041] The technical effects and advantages of the present invention are as follows:

[0042] (1) The present invention utilizes a voiceprint detection module, a voiceprint positioning module, a defect recognition module, an acoustic-optical fusion module, and an analysis unit in a coordinated manner, and adopts voiceprint detection technology to achieve voiceprint collection coverage of smart substation circuits, ensuring remote security patrol collection and monitoring of voiceprint data, thereby helping to solve the problem that abnormal voiceprint characteristics of potential faults cannot be discovered in time during manual patrols and listening of substation equipment. At the same time, it is conducive to the fusion of multiple physical signals such as voiceprints, vibrations, and electromagnetic fields, and the construction of a cross-modal association model in combination with a graph neural network, which is conducive to accurately distinguishing complex faults and improving the accuracy of diagnosis;

[0043] (2) The present invention uses a dynamic risk assessment module to predict the probability of failure and the remaining life based on the evolution law of the equipment voiceprint characteristics. At the same time, it uses a survival analysis model and a conditional variational autoencoder to quantify the equipment degradation trend, which is conducive to achieving failure probability prediction and reducing operation and maintenance costs;

[0044] (3) The present invention utilizes a configuration method in which the inspection management module and the autonomous inspection optimization module are coordinated to manage inspection data and records, and combines sound source positioning and fault alarm information to provide decision support to dynamically adjust sensor deployment, optimize monitoring coverage and energy consumption balance, thereby improving monitoring efficiency and facilitating the reduction of detection blind spots. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a framework diagram of the patrol detection framework of the present invention.

[0046] Figure 2 This is a block diagram of the detection unit of the present invention.

[0047] Figure 3 This is a block diagram of the analysis unit of the present invention.

[0048] Figure 4 This is a block diagram of the inspection unit of the present invention.

[0049] In the figure: 1. Patrol detection framework; 2. Detection unit; 201. Voiceprint detection module; 202. Voiceprint positioning module; 203. Defect recognition module; 204. Acoustic-optical fusion module; 3. Patrol inspection unit; 301. Patrol inspection management module; 302. Autonomous patrol inspection optimization module; 303. Dynamic risk assessment module; 4. Analysis unit; 401. Multi-physics field coupling analysis module; 402. Electromagnetic-acoustic coupling analysis module; 403. Cloud data analysis and storage module. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] The present invention provides Figure 1-4 A patrol detection system for a circuit device of an intelligent substation is shown, comprising a patrol detection framework 1, which comprises a detection unit 2, an inspection unit 3 and an analysis unit 4. The patrol detection framework 1 also comprises a visualization platform, which is used for aggregating station-level voiceprint data, real-time remote monitoring, intelligent analysis and sound visualization, and mastering the operating status of the equipment. The detection unit 2 is used for identifying defects in the circuit of the intelligent substation and performing voiceprint detection. The detection unit 2 comprises a voiceprint detection module 201 and a voiceprint positioning module 202. The voiceprint detection module 201 is used for obtaining voiceprint signals of the intelligent substation and circuit in real time. The voiceprint positioning module 202 is used to determine the spatial location of the abnormal sound source. The inspection unit 3 is used for patrol management, autonomous inspection optimization and dynamic risk assessment of the circuit of the intelligent substation. The analysis unit 4 is used to fuse multiple physical signals, construct a multi-dimensional state assessment model, and perform data analysis and storage.

[0052] Furthermore, the voiceprint detection module 201 deploys a non-contact voiceprint sensor array. The non-contact voiceprint sensor is used to collect the operating sound of high-voltage equipment and circuits in the smart substation, and has intelligent noise reduction capabilities to ensure audibility and high restoration. The non-contact voiceprint sensor array uses a high-precision microphone and an ultrasonic sensor in combination to cover the three-axis orientation of the substation. The voiceprint patrol detection method does not require personnel to listen to the sound on site, which is convenient for reducing manual labor intensity. At the same time, real-time online monitoring does not require professional personnel to invest, which can improve patrol efficiency. The voiceprint positioning module 202 calculates the source coordinates by combining the phase difference of the voiceprint sensor array and the beamforming algorithm based on TDOA. The voiceprint detection module 201 is connected to the voiceprint positioning module 202 through data transmission. The abnormal sound sources determined by the voiceprint positioning module 202 include local discharge points and gas leakage points. The formula of the beamforming algorithm based on TDOA is as follows:

[0053]

[0054] Among them, T ij It refers to the arrival time difference of the signals from microphones i and j, d i Refers to the straight-line distance from the sound source to microphone i, d j Refers to the straight-line distance from the sound source to microphone j, d i -d j Refers to the sensor spacing, and c refers to the speed of sound, which is beneficial for calculating the sound source coordinates by combining the microphone array phase difference to improve positioning accuracy.

[0055] Furthermore, the detection unit 2 also includes a defect recognition module 203 and an acoustic-optical fusion module 204. The defect recognition module 203 is used to accurately classify multiple types of typical defects, including DC bias magnetism, partial discharge, heavy overload, abnormal noise of the cooler, loose clamps and short-circuit impact. The acoustic-optical fusion module 204 is used to combine the SIFT feature matching of the acoustic image and the visible light image to realize the AR superposition display of the sound field distribution and the real scene of the equipment, and then visualize the sound source distribution and the real scene of the equipment. The voiceprint positioning module 202 is connected to the defect recognition module 203 by data transmission, and the acoustic-optical fusion module 204 is connected to the acoustic-optical fusion module 204 by data transmission. The AR superposition display realizes the fusion display of the sound field and the real scene through the camera calibration matrix. The camera calibration matrix formula is as follows:

[0056] H=K[R|t]

[0057] Among them, H refers to the projection matrix, which is used to convert three-dimensional world coordinates into two-dimensional image pixel coordinates, K refers to the intrinsic parameter matrix, which is used to describe the internal optical properties of the camera, R refers to the rotation matrix, which is used to indicate the camera orientation, t refers to the translation vector, which is used to indicate the camera position, [R|t] refers to the extrinsic parameter matrix, which is used to convert the point in the world coordinate system to the camera coordinate system, R is used to describe the rotation direction of the camera coordinate system relative to the world coordinate system, and the camera orientation is defined by Euler angles or rotation vectors, and t is used to describe the position of the origin of the camera coordinate system in the world coordinate system, and also indicates the spatial position offset of the camera.

[0058] Furthermore, the inspection unit 3 includes an inspection management module 301, an autonomous inspection optimization module 302 and a dynamic risk assessment module 303. The inspection management module 301 is used to manage inspection tasks and records, and provide decision support in combination with sound source positioning and fault alarm information. The autonomous inspection optimization module 302 is used to dynamically adjust the sensor deployment strategy and optimize the monitoring coverage and energy consumption balance. The dynamic risk assessment module 303 is used to predict the probability of failure and the remaining life based on the evolution law of the equipment voiceprint characteristics, which is conducive to realizing the failure probability prediction and reducing the operation and maintenance cost. The dynamic risk assessment module 303 adopts conditional variational autoencoder and survival analysis model. The inspection management module 301 is connected to the autonomous inspection optimization module 302 through data interaction, and the dynamic risk assessment module 303 is connected to the detection unit 2 through data transmission.

[0059] Specifically, the autonomous inspection optimization module 302 adopts deep deterministic policy gradient to define the state space, action space, and reward function, which refer to the sensor position and environmental noise, the moving step size and angle, the coverage area, and the signal-to-noise ratio, respectively. The Q value update formula included in the autonomous inspection optimization module 302 is as follows:

[0060] Q(s,a)=E[r+rQ(s′,μ(s′))]

[0061] Among them, Q(s,a) refers to the action value function, which represents the expected cumulative reward after performing action a in state s, E refers to the expectation operator, which represents the statistical average of subsequent states and rewards, r refers to the immediate reward, γ is the discount factor, s′ refers to the state at the next moment, which is generated by the environment dynamic model or the experience replay pool, and μ(s′) refers to the target policy function, which determines the optimal action to be performed in state s′. It is conducive to realizing autonomous obstacle avoidance and hot spot focusing of sensors through iterative optimization of the policy network μ.

[0062] Specifically, the conditional variational autoencoder is used to extract the potential distribution of voiceprint features and generate the fault evolution path. The survival analysis model is used to combine historical fault data and quantify time-related risks. The risk function formula included in the dynamic risk assessment module 303 is as follows:

[0063] λ(t|X)=λ0(t)exp(β T X)

[0064] Where X is the fingerprint feature vector, β is the weight parameter, which indicates the intensity of the influence of feature X on the risk, λ(t|X) is the risk function, which indicates the probability of instantaneous failure of the equipment under the condition of given feature X within time t, λ0(t) is the baseline risk function, which indicates the basic risk related only to practice, exp(β T X) refers to the risk ratio factor, which represents the scaling factor of the characteristic to the benchmark risk.

[0065] It should be noted that the analysis unit 4 includes a multi-physics field coupling analysis module 401, an electromagnetic-acoustic coupling analysis module 402 and a cloud data analysis and storage module 403. The multi-physics field coupling analysis module 401 is used to integrate multiple physical signals and construct a multi-dimensional status assessment model for the equipment. The electromagnetic-acoustic coupling analysis module 402 is used to combine electromagnetic field simulation and acoustic signals to locate mechanical vibration anomalies caused by electromagnetic interference. The electromagnetic-acoustic coupling analysis module 402 adopts a finite element-boundary element coupling algorithm to solve the coupling effect of Maxwell's equations and the acoustic wave equation. The cloud data analysis and storage module 403 is used to upload real-time monitoring data to the cloud, perform big data analysis, explore potential failure modes and trends, and provide data support for system optimization. The multi-physics field coupling analysis module 401 and the electromagnetic-acoustic coupling analysis module 402 are both connected to the cloud data analysis and storage module 403 through data transmission.

[0066] Specifically, the multiple physical signals fused by the multi-physics coupling analysis module 401 are voiceprint, infrared thermal imaging, and vibration signals. The data fusion of the multi-physics coupling analysis module 401 uses a graph neural network to map the acoustic signal, thermal imaging, and vibration into nodes and edges in a graph structure to capture cross-modal correlations. The formula included in the data fusion of the multi-physics coupling analysis module 401 is as follows:

[0067]

[0068] in, refers to the feature vector of node v in the l+1 layer, σ refers to the nonlinear activation function, which introduces the nonlinear expression ability of the model, N(v) refers to the set of neighbor nodes of node v, and W l refers to the learnable weight matrix of layer l, It refers to the feature vector of neighbor node u in layer l, and |N(v)| refers to the number of neighbors of node v, which is used to normalize the aggregation results. The above formula is conducive to identifying abnormal coupling patterns through neighborhood node feature aggregation, which is convenient for improving diagnostic accuracy and reducing the risk of single sensor misjudgment.

[0069] More specifically, the electromagnetic-acoustic coupling analysis module 402 uses the electromagnetic-acoustic coupling equation to calculate the vibration spectrum characteristics caused by eddy current damage to the core, which is conducive to locating electromagnetic faults with an accuracy of millimeter level. The formula is as follows:

[0070] ▽×(ν▽×Α)-ω 2 εΑ=J s +σΕ

[0071] Among them, ▽× refers to the curl operator, which is used to describe the rotation characteristics of the vector field, v refers to the magnetic resistivity, which characterizes the material's ability to block the magnetic field, Α refers to the magnetic vector potential, which is used to describe the magnetic potential distribution in the electromagnetic field, ω refers to the angular frequency, ε refers to the dielectric constant, which characterizes the polarization characteristics of the material in the electric field, and J s refers to the source current density, σ refers to the conductivity, which is used to describe the conductivity of the material, and E refers to the electric field strength.

[0072] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A patrol detection system for a circuit device of an intelligent substation, comprising a patrol detection framework (1), characterized in that: The patrol detection framework (1) includes: A detection unit (2), the detection unit (2) is used to identify defects in the smart substation circuit and perform voiceprint detection, the detection unit (2) comprising: A voiceprint detection module (201), the voiceprint detection module (201) is used to obtain smart substation and circuit voiceprint signals in real time; A voiceprint positioning module (202), the voiceprint positioning module (202) is used to determine the spatial location of an abnormal sound source; An inspection unit (3), the inspection unit (3) is used to perform inspection management, autonomous inspection optimization, and dynamic risk assessment on the circuits of the smart substation; An analysis unit (4) is used to fuse multiple physical signals, construct a multi-dimensional state assessment model, and perform data analysis and storage.

2. The patrol detection system for circuit devices of an intelligent substation according to claim 1, characterized in that: The voiceprint detection module (201) deploys a non-contact voiceprint sensor array to cover the three-axis orientation of the substation. The voiceprint positioning module (202) calculates the source coordinates by combining the phase difference of the voiceprint sensor array and a beamforming algorithm based on TDOA. The voiceprint detection module (201) is connected to the voiceprint positioning module (202) by data transmission. The abnormal sound sources determined by the voiceprint positioning module (202) include local discharge points and gas leakage points.

3. The patrol detection system for circuit devices of an intelligent substation according to claim 2, characterized in that: The formula of the TDOA-based beamforming algorithm is as follows: Among them, T ij It refers to the arrival time difference of the signals from microphones i and j, d i Refers to the straight-line distance from the sound source to microphone i, d j Refers to the straight-line distance from the sound source to microphone j, d i -d j is the sensor spacing, and c is the speed of sound.

4. The patrol detection system for circuit devices of an intelligent substation according to claim 1, characterized in that: The detection unit (2) further comprises: A defect recognition module (203), wherein the defect recognition module (203) is used to accurately classify multiple types of typical defects; The acoustic-optical fusion module (204) is used to combine the SIFT feature matching of the acoustic image and the visible light image to realize AR superposition display of the sound field distribution and the real scene of the equipment, thereby visually displaying the sound source distribution and the real scene of the equipment. The voiceprint positioning module (202) is connected to the defect recognition module (203) by data transmission. The acoustic-optical fusion module (204) is connected to the acoustic-optical fusion module (204) by data transmission. The AR superposition display realizes the fusion display of the sound field and the real scene through the camera calibration matrix. The camera calibration matrix formula is as follows: H=K[R|t] Where H is the projection matrix, which is used to convert 3D world coordinates into 2D image pixel coordinates; K is the intrinsic parameter matrix, which is used to describe the internal optical properties of the camera; R is the rotation matrix, which is used to indicate the camera orientation; t is the translation vector, which is used to indicate the camera position; and [R|t] is the extrinsic parameter matrix, which is used to convert points in the world coordinate system into the camera coordinate system.

5. The patrol detection system for circuit devices of an intelligent substation according to claim 1, characterized in that: The inspection unit (3) comprises: Inspection management module (301), the inspection management module (301) is used to manage inspection data and records, and provide decision support in combination with sound source location and fault alarm information; An autonomous inspection optimization module (302), the autonomous inspection optimization module (302) is used to dynamically adjust the sensor deployment strategy to optimize the monitoring coverage and energy consumption balance; A dynamic risk assessment module (303) is used to predict the probability of failure and the remaining life based on the evolution law of the device voiceprint characteristics. The dynamic risk assessment module (303) adopts a conditional variational autoencoder and a survival analysis model. The inspection management module (301) is connected to the autonomous inspection optimization module (302) through data interaction. The dynamic risk assessment module (303) is connected to the detection unit (2) through data transmission.

6. The patrol detection system for circuit devices of an intelligent substation according to claim 5, characterized in that: The autonomous inspection optimization module (302) adopts deep deterministic policy gradient to define the state space, action space and reward function, which refer to the sensor position and environmental noise, movement step size and angle, coverage area and signal-to-noise ratio respectively. The Q value update formula included in the autonomous inspection optimization module (302) is as follows: Q(s,a)=E[r+rQ(s′,μ(s′))] Among them, Q(s,a) refers to the action value function, which represents the expected cumulative reward after performing action a in state s, E refers to the expectation operator, which represents the statistical average of subsequent states and rewards, r refers to the immediate reward, γ is the discount factor, s′ refers to the state at the next moment, which is generated by the environment dynamics model or the experience replay pool, and μ(s′) refers to the target policy function, which determines the optimal action to be performed in state s′.

7. The patrol detection system for circuit devices of an intelligent substation according to claim 6, characterized in that: The conditional variational autoencoder is used to extract the potential distribution of voiceprint features and generate a fault evolution path. The survival analysis model is used to combine historical fault data and quantify time-related risks. The risk function formula included in the dynamic risk assessment module (303) is as follows: λ(t|X)=λ0(t)exp(β T X) Where X is the fingerprint feature vector, β is the weight parameter, which indicates the intensity of the influence of feature X on the risk, λ(t|X) is the risk function, which indicates the probability of instantaneous failure of the equipment under the condition of given feature X within time t, λ0(t) is the baseline risk function, which indicates the basic risk related only to practice, exp(β T X) refers to the risk ratio factor, which represents the scaling factor of the characteristic to the benchmark risk.

8. The patrol detection system for circuit devices of an intelligent substation according to claim 1, characterized in that: The analysis unit (4) comprises: A multi-physics field coupling analysis module (401), wherein the multi-physics field coupling analysis module (401) is used to fuse multiple physical signals and construct a multi-dimensional state assessment model for the equipment; An electromagnetic-acoustic coupling analysis module (402), the electromagnetic-acoustic coupling analysis module (402) being used to combine electromagnetic field simulation with acoustic signals to locate mechanical vibration anomalies caused by electromagnetic interference, the electromagnetic-acoustic coupling analysis module (402) using a finite element-boundary element coupling algorithm to solve the coupling effect of Maxwell's equations and acoustic wave equations; A cloud data analysis and storage module (403) is used to upload real-time monitoring data to the cloud, perform big data analysis, mine potential failure modes and trends, and provide data support for system optimization. The multi-physics field coupling analysis module (401) and the electromagnetic-acoustic coupling analysis module (402) are both connected to the cloud data analysis and storage module (403) by data transmission.

9. The patrol detection system for circuit devices of an intelligent substation according to claim 8, characterized in that: The multiple physical signals fused by the multi-physics coupling analysis module (401) are voiceprint, infrared thermal imaging and vibration signals. The data fusion of the multi-physics coupling analysis module (401) uses a graph neural network to map the acoustic signal, thermal imaging and vibration into nodes and edges in a graph structure to capture cross-modal correlation. The formula included in the data fusion of the multi-physics coupling analysis module (401) is as follows: in, refers to the feature vector of node v in the l+1 layer, σ refers to the nonlinear activation function, which introduces the nonlinear expression ability of the model, N(v) refers to the set of neighbor nodes of node v, and W l refers to the learnable weight matrix of layer l, It refers to the feature vector of neighbor node u at layer l, and |N(v)| refers to the number of neighbors of node v, which is used to normalize the aggregation results.

10. The patrol detection system for circuit devices of an intelligent substation according to claim 8, characterized in that: The electromagnetic-acoustic coupling analysis module (402) calculates the vibration spectrum characteristics caused by the eddy current damage of the iron core using the electromagnetic-acoustic coupling equation, and the formula is as follows: in, refers to the curl operator, which is used to describe the rotation characteristics of the vector field, v refers to the magnetic resistivity, which characterizes the material's ability to hinder the magnetic field, Α refers to the magnetic vector potential, which is used to describe the magnetic potential distribution in the electromagnetic field, ω refers to the angular frequency, ε refers to the dielectric constant, which characterizes the polarization characteristics of the material in the electric field, J s refers to the source current density, σ refers to the conductivity, which is used to describe the conductivity of the material, and E refers to the electric field strength.

Images