Low-voltage feeder cabinet automatic test system and method based on artificial intelligence

Abstract

The invention discloses a low-voltage feeder cabinet automatic test system and method based on artificial intelligence, and relates to the technical field of power system intelligent detection, and the method comprises the steps: extracting equipment parameters in an initial knowledge graph through a neuromorphic calculation engine, and dynamically generating a antagonism test waveform through a driving pulse neural network; injecting the antagonism test waveform into a corresponding drawer output terminal of the to-be-tested cabinet body through a dynamic load simulator, and synchronously acquiring response signals of a current waveform, temperature distribution and a circuit breaker action time sequence; setting a fault judgment threshold according to the historical sample and the real-time operation state, and generating a fault diagnosis sub-report and a health state based on a comparison result of the fault node score and the fault judgment threshold; the equipment parameters are extracted by using the neuromorphic calculation engine and the antagonism test waveform is dynamically generated, so that the intelligent level, the diagnosis accuracy and the operation safety of the low-voltage feeder cabinet test are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent testing technology for power systems, and in particular to an automatic testing system and method for low-voltage feeder cabinets based on artificial intelligence. Background Technology

[0002] In the process of intelligent transformation of power systems, low-voltage feeder cabinets, as key terminal control and protection equipment in distribution networks, directly affect the stability of the entire system due to their operational reliability. With the development of Industry 4.0 and digital twin technology, traditional low-voltage feeder cabinet testing methods that rely on manual inspection and static testing are gradually being replaced by automated and intelligent solutions. In recent years, data-driven intelligent diagnostic technology has made some progress in this field. Typical examples include using sensor networks to collect electrical parameters in real time and combining them with machine learning algorithms to classify and identify circuit breaker states; some studies have introduced preliminary digital twin models to simulate equipment behavior and predict potential faults.

[0003] Although existing technologies have improved the testing efficiency and accuracy of low-voltage feeder cabinets to some extent, they still have significant limitations in dealing with nonlinear responses and early fault identification under complex operating conditions. Current mainstream methods usually rely on preset test waveforms for excitation, which is difficult to cover extreme or adversarial operating scenarios that may occur in actual operation, resulting in insufficient test coverage. Most systems only focus on single parameter threshold judgment and lack a fusion analysis mechanism for multi-source heterogeneous signals (such as current waveforms, temperature distribution and timing actions), thus limiting the sensitivity to progressive faults such as insulation aging. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an automatic testing method for low-voltage feeder cabinets based on artificial intelligence to solve the problems of insufficient test coverage and poor adaptability to complex operating conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an automatic testing method for low-voltage feeder cabinets based on artificial intelligence, comprising: scanning the QR code of the cabinet under test to obtain design drawing information, constructing a cognitive digital twin and generating an initial knowledge graph; extracting equipment parameters from the initial knowledge graph using a neuromorphic computing engine, and dynamically generating adversarial test waveforms by driving a pulse neural network; injecting the adversarial test waveforms into the corresponding drawer output terminals of the cabinet under test through a dynamic load simulator, and simultaneously acquiring response signals of current waveforms, temperature distribution, and circuit breaker action timing; inputting the response signals into the cognitive digital twin to extract spatial temperature distribution data, and activating the insulation aging branch of the initial knowledge graph when the local temperature rise rate is detected to exceed the temperature rise threshold, extracting temperature rise anomaly indicators, current disturbance indicators, and action time difference indicators, and calculating fault node scores; setting fault judgment thresholds based on historical samples and real-time operating status, and generating fault diagnosis sub-reports and health status based on the comparison results of fault node scores and fault judgment thresholds; integrating the fault diagnosis sub-reports and health statuses of all drawers, calculating the health index of each drawer and the overall health index, and generating a test report.

[0008] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the steps of scanning the QR code of the cabinet under test to obtain design drawing information, constructing a cognitive digital twin, and generating an initial knowledge graph are as follows:

[0009] Scan the QR code of the cabinet to be tested to obtain design drawing information and equipment parameters. Based on the equipment parameters, drive the neural symbolic joint engine, extract visual features of the drawings through visual neural network, and combine first-order logic rules to infer equipment type and connection relationship to generate a preliminary knowledge graph.

[0010] The device topology relationships in the preliminary knowledge graph are input into the quantum annealing optimizer to obtain the optimal topology structure by solving for the device connection strength and importance weights, and output the optimized electrical connection graph.

[0011] When a conflict in device parameters is detected, the multi-agent debate engine is activated. By integrating the confidence decisions of agents based on electrical specifications, historical faults, and physical constraints, the conflicting parameters are automatically corrected and a correction log is generated.

[0012] The optimized electrical connection graph and correction logs are imported into the Hypergraph database to construct a cognitive digital twin. The device nodes, corrected parameters, and topological relationships in the electrical connection graph are encoded into a structured knowledge network to form an initial knowledge graph.

[0013] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the steps of extracting equipment parameters from an initial knowledge graph using a neuromorphic computing engine and dynamically generating adversarial test waveforms by driving a spiking neural network are as follows:

[0014] The circuit breaker operating time constant and transformer characteristic frequency are extracted from the initial knowledge graph to construct the equipment feature vector. The equipment feature vector is input into the neuromorphic computing engine to drive the spiking neural network to generate the fundamental wave test waveform.

[0015] An adversarial test waveform is generated by injecting true random noise into the fundamental test waveform using a quantum random number generator.

[0016] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the steps include: injecting adversarial test waveforms into the corresponding drawer output terminals of the cabinet under test using a dynamic load simulator, and simultaneously acquiring response signals such as current waveforms, temperature distribution, and circuit breaker operating timing. The specific steps are as follows:

[0017] Based on the cabinet topology parameters in the cognitive digital twin, the output impedance of the load simulator is dynamically adjusted to match the terminals of the drawer under test;

[0018] The adversarial test waveform is injected into the drawer terminal after impedance matching, and the three-channel sensor array is triggered by the quantum synchronization clock to synchronously collect the current waveform signal, infrared thermal imaging temperature distribution signal and fiber optic sensing circuit breaker action signal.

[0019] It integrates current waveform signals, infrared thermal imaging temperature distribution signals, and fiber optic circuit breaker action signals to output a response signal.

[0020] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the specific steps for calculating the fault node score are as follows:

[0021] The response signal is input into the cognitive digital twin, the spatial temperature distribution data of the thermal_map field is extracted, the temperature rise rate at each coordinate point is calculated, and the temperature rise threshold is set according to the thermal collapse critical point of the insulation material.

[0022] When the temperature rise rate exceeds the temperature rise threshold, the insulation aging branch in the initial knowledge graph is activated, and the fault node is obtained by matching the nodes in the initial knowledge graph through heat conduction path tracing.

[0023] When injecting the fundamental test waveform into the dynamic load simulator, the current waveform and circuit breaker action timing data are collected, and the abnormal temperature rise index, current disturbance index and action time difference index are calculated. The fault node score is obtained by combining the results.

[0024] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the generation of fault diagnosis sub-reports and health status is based on dynamically setting fault judgment thresholds according to the historical fault occurrence probability and real-time equipment status. When the fault node score is greater than the fault judgment threshold, a fault diagnosis sub-report is generated; when the fault node score is less than the fault judgment threshold, it is marked as a healthy state.

[0025] As a preferred embodiment of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets described in this invention, the steps of integrating the fault diagnosis sub-reports and health status of all drawers, calculating the health index of each drawer and the overall health index, and generating a test report are as follows:

[0026] Extract the fault diagnosis sub-reports and health status of each drawer, and use the health index calculation formula to convert the fault node scores into drawer health indices;

[0027] The drawer health index is weighted and summarized to obtain the overall health index, and the health levels are divided. A test report is generated by integrating the fault diagnosis sub-reports of each drawer with the health levels.

[0028] Secondly, this invention provides an automatic testing system for low-voltage feeder cabinets based on artificial intelligence, comprising a knowledge graph module, a waveform module, a signal module, a weighting module, a state evaluation module, and an integration module. The knowledge graph module is used to scan the QR code of the cabinet under test to obtain design drawing information, construct a cognitive digital twin, and generate an initial knowledge graph. The waveform module is used to extract equipment parameters from the initial knowledge graph using a neuromorphic computing engine and dynamically generate adversarial test waveforms by driving a pulse neural network. The signal module is used to inject the adversarial test waveforms into the corresponding drawer output terminals of the cabinet under test through a dynamic load simulator, and simultaneously acquire current waveforms, temperature distribution, and circuit breakers. The system includes: a response signal for the device's action timing; a weighting module, which inputs the response signal into the cognitive digital twin to extract spatial temperature distribution data; when the local temperature rise rate exceeds the temperature rise threshold, it activates the insulation aging branch of the initial knowledge graph to extract abnormal temperature rise indicators, current disturbance indicators, and action time difference indicators, and calculates the fault node score; a status assessment module, which sets the fault judgment threshold based on historical samples and real-time operating status, and generates a fault diagnosis sub-report and health status based on the comparison result of the fault node score and the fault judgment threshold; and an integration module, which integrates the fault diagnosis sub-reports and health status of all drawers, calculates the health index of each drawer and the overall health index, and generates a test report.

[0029] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in the first aspect of the present invention.

[0030] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based automatic testing method for low-voltage feeder cabinets as described in the first aspect of the present invention.

[0031] The beneficial effects of this invention are as follows: By utilizing a neuromorphic computing engine to extract equipment parameters and dynamically generate adversarial test waveforms, it can accurately simulate complex electrical environments under real working conditions, significantly improving the targeting of tests and the fault detection rate; by inputting response signals into a cognitive digital twin to extract spatial temperature distribution data, and activating the insulation aging branch in the knowledge graph when the temperature rise is abnormal to calculate the fault node score, it realizes the early identification and quantitative assessment of potential insulation defects, greatly improving the intelligence level, diagnostic accuracy and operational safety of low-voltage feeder cabinet testing. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of an automated testing method for low-voltage feeder cabinets based on artificial intelligence.

[0034] Figure 2 This is a schematic diagram of an AI-based automated testing system for low-voltage feeder cabinets.

[0035] Figure 3 A flowchart for constructing a cognitive digital twin.

[0036] Figure 4 A flowchart for generating adversarial test waveforms. Detailed Implementation

[0037] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0038] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0040] Reference Figures 1-4 This is one embodiment of the present invention, which provides an automatic testing method for low-voltage feeder cabinets based on artificial intelligence, including the following steps:

[0041] S1. Scan the QR code of the cabinet to be tested to obtain design drawing information, construct a cognitive digital twin and generate an initial knowledge graph.

[0042] Scan the QR code of the cabinet to be tested to obtain design drawing information and equipment parameters. Based on the equipment parameters, drive the neural symbolic joint engine, extract visual features from the drawings through visual neural network, and combine them with first-order logic rules to infer the equipment type and connection relationship, generating a preliminary knowledge graph.

[0043] Specifically, the QR code on the cabinet under test is scanned using an industrial QR code scanner to obtain the encrypted design drawing data stream. The data is then decrypted using the national cryptographic SM4 dynamic key to generate structured equipment parameters. Based on the equipment parameters, a neural-symbolic joint engine is driven to perform dual-channel processing: the visual neural network channel uses VisionTransformer to parse the drawing bitmap and output a visual feature tensor, while the symbolic logic channel uses a Prolog first-order logic engine to perform equipment connection rule reasoning. The visual feature tensor and the symbolic logic reasoning results are merged to generate a preliminary knowledge graph containing equipment entity nodes and topological connection relationships.

[0044] The device topology relationships in the preliminary knowledge graph are input into the quantum annealing optimizer to obtain the optimal topology structure by solving for the device connection strength and importance weights, and output the optimized electrical connection graph.

[0045] Specifically, circuit breaker terminals, busbar connection points, and copper busbar connection paths are extracted from the preliminary knowledge graph. All nodes are sorted alphabetically to generate an adjacency matrix. The matrix is ​​then input into the quantum annealing optimizer to set the Ising model Hamiltonian equation: the connection strength between devices is defined by the cable cross-sectional area, and the device importance weight is a fixed value for the circuit breaker. The D-Wave quantum processor performs annealing operations for an example of 2000 iterations to solve for the lowest energy state and output the device connection state vector. The connection state vector is then reconstructed to generate an optimized electrical connection graph after deleting redundant connection paths.

[0046] It should be noted that an adjacency matrix is ​​generated by arranging all entity nodes in alphabetical order from the initial knowledge graph: node pairs connected by copper busbars are marked as 1 in the matrix (undirected connections require bidirectional assignment), independent terminals retain all rows of 0, and the diagonal is always 0, thus obtaining the device topology relationship in the initial knowledge graph.

[0047] When a conflict in device parameters is detected, the multi-agent debate engine is activated. By integrating the confidence decisions of agents based on electrical specifications, historical faults, and physical constraints, the conflicting parameters are automatically corrected, and a correction log is generated.

[0048] Specifically, when a conflict in equipment parameters is detected, the multi-agent debate engine is activated. Based on the power equipment parameter self-correction technology of multi-agent debate, a correction scheme and confidence level are obtained. The historical fault agent searches the State Grid case library and outputs parameter adjustment suggestions and confidence levels. The physical constraint agent verifies the feasibility of the scheme based on the thermal expansion coefficient of the material and outputs the confidence level. The confidence level decision is obtained by calculating through a weighted formula. The scheme with the highest confidence level is selected to perform parameter coverage and generate a correction log.

[0049] The optimized electrical connection graph and correction logs are imported into the Hypergraph database to construct a cognitive digital twin. The device nodes, corrected parameters, and topological relationships in the electrical connection graph are encoded into a structured knowledge network to form an initial knowledge graph.

[0050] Specifically, the optimized electrical connection graph output by the quantum annealing optimizer includes circuit breaker terminals, busbar connection point entities, and copper busbar connection paths after removing redundant connection paths. The copper busbar connection path strength is obtained, and a hyperedge structure with device entities as nodes is created. The copper busbar connection path strength is stored as a hyperedge attribute. Timeline event nodes are constructed based on correction logs and associated with the corresponding device hyperedges. Thermodynamic equations are embedded in the hyperedge attributes using pre-compiled Java. UDF enables real-time solving, establishing a device parameter version chain based on the correction log time series to obtain a cognitive digital twin. The correction logs generated by the multi-agent debate engine are jointly imported into the HyperGraphDB storage engine. In the HyperGraphDB hypergraph database, device entity vertices, topological hyperedges, and correction parameters are organized into a knowledge network with electrical connection semantics according to spatial topological relationships, based on the HyperGraphDB hypergraph data structure. Data rows with device identifier field as transformer number and parameter name field as ratio are selected from the correction log, and the param_value corresponding to the timestamp is extracted as the transformer parameter ratio value. The transformer parameter ratio value in the correction log is overwritten with the corresponding vertex attribute field to generate an initial knowledge graph containing device entity vertices, topological hyperedges, and correction parameters.

[0051] S2. Utilize a neuromorphic computing engine to extract device parameters from the initial knowledge graph, and dynamically generate adversarial test waveforms by driving a spiking neural network.

[0052] The circuit breaker operating time constant and transformer characteristic frequency are extracted from the initial knowledge graph to construct the equipment feature vector. The equipment feature vector is then input into the neuromorphic computing engine to drive the spiking neural network to generate the fundamental wave test waveform.

[0053] Specifically, the circuit breaker operating time constant is read from the circuit breaker entity attribute field of the initial knowledge graph, and the transformer characteristic frequency is read from the transformer entity attribute field; the circuit breaker operating time constant and the transformer characteristic frequency are combined to form a two-dimensional device feature vector; the two-dimensional device feature vector is input into the driving spiking neural network deployed on the neuromorphic chip, and the fundamental wave test waveform with an exponentially decaying envelope is generated by the firing of spiking neurons.

[0054] It should be noted that the driving spiking neural network was trained based on 50,000 sets of measured waveform samples from a power distribution equipment fault waveform database, including three typical scenarios: transient process of vacuum circuit breaker tripping, abnormal magnetic saturation of transformers, and arcing fault of contactors. The spiking neural network was constructed using the Intel Lava framework. The network structure includes a layer of 128 leaky integral ignition neurons with example values ​​and an adaptive threshold adjustment layer. The input layer receives a two-dimensional feature vector, and the output layer fits the envelope value of the target waveform. The synaptic weights of neurons were adjusted using a pulse time-dependent plasticity learning rule. The optimizer used the neuromorphic hardware-compatible SLAYER algorithm. The loss function was defined as the root mean square error between the output waveform and the measured waveform in the key time window. Training terminated when the error of the validation set was lower than the specified value of 0.05 volts for 20 consecutive cycles, resulting in the trained driving spiking neural network.

[0055] An adversarial test waveform is generated by injecting true random noise into the fundamental test waveform using a quantum random number generator.

[0056] Specifically, the fundamental test waveform generated by the driving spiking neural network is input into the ID Quantique QuantisQRNG (quantum random number generator). The quantum random number generator outputs a truly random binary sequence within a 1 nanosecond time window based on the quantum tunneling effect. The binary sequence is converted into an analog noise voltage signal by a DAC digital-to-analog converter at the same 100 kHz sampling rate as the fundamental test waveform. The fundamental test waveform and the analog noise voltage signal are then added algebraically in the real time in the signal synthesis circuit to output an adversarial test waveform injected with truly random noise.

[0057] S3. The adversarial test waveform is injected into the corresponding drawer output terminal of the cabinet under test through the dynamic load simulator, and the response signals of current waveform, temperature distribution and circuit breaker action sequence are collected simultaneously.

[0058] Based on the cabinet topology parameters in the cognitive digital twin, the output impedance of the load simulator is dynamically adjusted to match the terminals of the drawer under test.

[0059] Specifically, the cabinet topology parameters in the cognitive digital twin originate from the initial knowledge graph stored in HyperGraphDB. The initial knowledge graph includes the vertex attributes of circuit breaker entities storing the spatial coordinates of circuit breaker terminals, the vertex attributes of busbar segment connection points storing the spatial coordinates of busbar segment connection points, and the hyperedge attributes of copper busbar connection paths storing the connection type and geometric parameters. Based on the geometric parameters of the connection paths, the target impedance is calculated using a formula. The load simulator automatically adjusts the output stage inductor-capacitor network according to the target impedance to achieve matching with the terminals of the drawer under test.

[0060] It should be noted that the formula for calculating the target impedance amplitude is:

[0061] ;

[0062] in, express Frequency of The target impedance amplitude of the connection path, Indicates the first Line inductance of the connection path, Indicates the first Line capacitance of a connecting path, Indicates the first The line resistance of the connection path, Indicates frequency, Indicates the connection path.

[0063] The adversarial test waveform is injected into the impedance-matched drawer terminal, and the three-channel sensor array is triggered by the quantum synchronization clock to synchronously acquire the current waveform signal, the infrared thermal imaging temperature distribution signal, and the fiber optic sensing circuit breaker action signal.

[0064] Specifically, the adversarial test waveform is applied to the impedance-matched terminals of the test drawer via a high-voltage injection probe; the ID Quantique Clavis3 quantum key distribution generates entangled photon pairs to initiate a quantum clock synchronization protocol, transmitting clock pulse signals to a three-channel sensor array via optical fiber; the Rogowski coil sensor array responds to the clock pulse by acquiring current waveform signals at an example sampling rate of 200kHz; the FLIR A700 infrared thermal imager array responds to the clock pulse by acquiring 128×128 pixel temperature distribution signals and marking spatial coordinates at an example frame rate of 120Hz; and the fiber Bragg grating sensor responds to the clock pulse by sampling the circuit breaker action signal at an example value of 10MHz and calculating the action delay.

[0065] It integrates current waveform signals, infrared thermal imaging temperature distribution signals, and fiber optic circuit breaker action signals to output a response signal.

[0066] Specifically, the current waveform signal acquired by the Rogowski coil sensor array is analyzed based on the quantum synchronous clock timestamp, and the temperature distribution signal acquired by the FLIR A700 infrared thermal imager array is synchronously aligned. The timestamp of the circuit breaker action event recorded by the fiber Bragg grating sensor is associated with it. The time series of the current waveform signal and the spatial coordinates of the temperature distribution signal are mapped to a unified coordinate system through a three-dimensional spatiotemporal transformation matrix, and the time label of the circuit breaker action signal is marked as a key event node. Within the action time node window, the abrupt change features of the current waveform signal and the spatial gradient change of the temperature distribution signal are extracted to generate an action response feature vector and associate it with the spatial coordinate point. The current signal sample point is matched to the spatial position of the thermal imaging frame corresponding to the timestamp, and a structured response signal is output.

[0067] S4. Input the response signal into the cognitive digital twin to extract spatial temperature distribution data. When the local temperature rise rate is detected to exceed the temperature rise threshold, activate the insulation aging branch of the initial knowledge graph, extract the temperature rise anomaly index, current disturbance index and action time difference index, and calculate the fault node score.

[0068] The response signal is input into the cognitive digital twin, the spatial temperature distribution data of the thermal_map field is extracted, the temperature rise rate at each coordinate point is calculated, and the temperature rise threshold is set according to the thermal collapse critical point of the insulation material.

[0069] Specifically, the collected response signals are input into the cognitive digital twin. The thermal_map field of the response signal object is directly accessed via Python. The temperature matrix and the absolute coordinate system of the device are parsed. Each spatial coordinate point is iterated using row and column indexes to extract the spatial temperature distribution data of each coordinate (including the temperature value sequence of three-dimensional coordinate points; for example, the spatial coordinate point (1250, 780, 30) is 56.7℃ at t=0ms and 62.1℃ at t=100ms). The temperature rise rate of each coordinate within a fixed time window of 100ms is calculated for each spatial coordinate point. By gradually increasing the sample temperature and monitoring the mass change, heat absorption or release, the temperature point at which the insulation material begins to undergo significant thermal degradation is taken as the thermal collapse critical point of the insulation material, and a global temperature rise threshold is set based on the thermal collapse critical point of the insulation material.

[0070] It should be noted that the Leica AT960 laser tracker was used to locate the equipment. The installation flange center was used as the coordinate origin (0,0,0). The axial directions were defined as follows: the X-axis was horizontal and parallel to the direction of the busbar, the Y-axis was horizontal and perpendicular to the busbar, and the Z-axis was vertical and upward. The structural feature points of the equipment were scanned with a laser. The distance from the flange center to the connection point at the beginning of the busbar was used as the benchmark for verification. The origin coordinates (0,0,0), the axis direction vectors, and the error compensation parameters were written into the spatial calibration database to obtain the absolute coordinate system of the equipment.

[0071] It should be noted that the formula for calculating the rate of temperature rise is:

[0072] ;

[0073] in, Indicates the rate of temperature rise. express Temperature 100 milliseconds later. express The temperature of a moment.

[0074] When the temperature rise rate exceeds the temperature rise threshold, the insulation aging branch in the initial knowledge graph is activated, and the fault node is obtained by matching the nodes in the initial knowledge graph through heat conduction path tracing.

[0075] Specifically, when the temperature rise rate at a spatial coordinate point exceeds the temperature threshold, the insulation aging branch of the insulation material entity in the initial knowledge graph of the cognitive digital twin is activated; a three-dimensional heat conduction path tracing based on the Fourier heat conduction equation is performed, tracing from the high-temperature point along the temperature gradient descent direction to the electrical connection point; temperature distribution signals are collected using a FLIR A700 infrared thermal imager array, and the abnormal hot zone of the temperature rise rate (e.g., starting point 102.3, 58.7, 0 mm units) is located along the heat diffusion gradient descent direction to obtain the endpoint coordinates of the insulation aging branch; the equipment point cloud data is acquired using a FARO Focus S350 laser scanner, and the contour vertices of the circuit breaker terminals and busbar connection points are extracted to obtain the spatial coordinates of the equipment entity vertices in the initial knowledge graph; the endpoint coordinates of the insulation aging branch are matched with the spatial coordinates of the equipment entity vertices in the initial knowledge graph; the matched equipment entity vertex is the fault node.

[0076] During the fundamental wave test waveform injection in the dynamic load simulator, current waveform data and circuit breaker action timing data are collected. Combined with the activated insulation aging branches and fault nodes, abnormal temperature rise index, current disturbance index and action time difference index are extracted, and fault node score is calculated based on weight coefficients.

[0077] Specifically, during the fundamental wave test waveform injection from the dynamic load simulator, the Rogowski coil sensor collects current waveform data. The current change rate time series is obtained through differential calculation. The Rogowski coil sensor continuously collects current waveform data of the branch where the fault node is located, obtaining a discrete current sampling sequence. After denoising and smoothing the discrete current sampling sequence, the current change rate at each sampling point is calculated according to the ratio of the current difference between adjacent sampling times to the sampling time interval, thus obtaining the corresponding current change rate time series. The actual tripping time is obtained based on the circuit breaker auxiliary contact signal and control circuit time scale. The maximum temperature rise rate of the fault node within the detection window, the root mean square of the current change rate at the fault node, and the absolute value of the deviation of the actual tripping time from the rated tripping time are used as three types of original characteristic quantities. The upper limit of temperature rise rate, the upper limit of current change rate, and the upper limit of allowable action time difference from historical test samples of the same type of feeder cabinet are used as normalization benchmarks to obtain temperature rise anomaly index, current disturbance index, and action time difference index. These are then weighted and summed according to weighting coefficients to obtain the fault node score.

[0078] The weighting coefficients were obtained by calibrating historical test samples of the same type of feeder cabinet. During the calibration process, the temperature rise rate, current change rate and action time deviation were standardized, and the insulation aging fault was used as a label for fitting. The three weighting coefficients were then normalized so that the sum of the three weighting coefficients was 1.

[0079] S5. Establish fault judgment thresholds based on historical samples and make online corrections in conjunction with real-time operating status. Generate fault diagnosis sub-reports and health status based on the comparison results between fault node scores and fault judgment thresholds.

[0080] The fault determination threshold is dynamically set based on the historical failure probability of the equipment and the real-time equipment status. When the fault node score is greater than the fault determination threshold, a fault diagnosis sub-report is generated. When the fault node score is less than the fault determination threshold, it is marked as healthy.

[0081] Specifically, historical test samples are categorized according to whether insulation aging faults have occurred. The fault node score range corresponding to samples with insulation aging faults is statistically analyzed, and a basic fault judgment threshold is obtained by combining ambient temperature, ambient humidity, load rate, and service life. During online detection, real-time ambient temperature, ambient humidity, and load rate are collected. When the ambient temperature or humidity increases, the fault sensitivity is increased, and when the load rate increases, the fault tolerance range is reduced, thereby correcting the basic fault judgment threshold and obtaining the fault judgment threshold. When the fault node score exceeds the fault judgment threshold, a fault diagnosis sub-report is generated. When the fault node score does not exceed the current fault judgment threshold, a health status marker is written into the initial knowledge graph.

[0082] S6. Integrate the fault diagnosis sub-reports and health status of all drawers, calculate the health index of each drawer and the overall health index, and generate a test report.

[0083] Extract the fault diagnosis sub-reports and health status of each drawer, and use the health index calculation formula to convert the fault node scores into drawer health indices.

[0084] Specifically, the fault node score of each drawer is read, and the drawer health index is calculated using a unified health index mapping function; then, the drawer importance weight is obtained according to the rated current level, load level or circuit importance of each drawer, and the overall health index is calculated using a weighted average method.

[0085] It should be noted that the expression for calculating the drawer health index is:

[0086] ;

[0087] in, Indicates the first The health index of each drawer Indicates the first Each drawer corresponds to a fault node's fault node score. Represents the drawer index variable. This represents the upper limit benchmark value of the health index.

[0088] The health status of each drawer is divided into different levels based on its health index value. The fault diagnosis sub-reports of all drawers are combined with the health status of each level to generate a test report.

[0089] Specifically, based on the five-level classification standard defined by the health index values ​​of all drawers (for example, a health index value ≥ 0.95 is "healthy", [0.90, 0.95) is "sub-healthy", [0.85, 0.90) is "attention", [0.75, 0.85) is "warning", and < 0.75 is "dangerous"), the key fields of the fault diagnosis sub-report (faulty drawers include fault type "insulation aging" and location coordinates), and healthy drawers retain the health index value and health status text description are bound together; according to the equipment report architecture, a three-level structure is created in the JSON-LD format test report: report header fields, drawer status table (including health index values ​​and status levels), and fault details fields (only faulty drawers output specific diagnostic data), and a test report is generated.

[0090] This embodiment also provides an artificial intelligence-based automatic testing system for low-voltage feeder cabinets, including: a knowledge graph module, a waveform module, a signal module, a weighting module, a state evaluation module, and an integration module. The knowledge graph module is used to scan the QR code of the cabinet under test to obtain design drawing information, construct a cognitive digital twin, and generate an initial knowledge graph. The waveform module is used to extract equipment parameters from the initial knowledge graph using a neuromorphic computing engine and dynamically generate adversarial test waveforms by driving a pulse neural network. The signal module is used to inject the adversarial test waveforms into the corresponding drawer output terminals of the cabinet under test through a dynamic load simulator, and simultaneously collect current waveforms, temperature distribution, and circuit breaker dynamics. The system consists of a timing response signal module; a weighting module, which inputs the response signal into the cognitive digital twin to extract spatial temperature distribution data; an insulation aging branch of the initial knowledge graph is activated when the local temperature rise rate exceeds the temperature rise threshold, extracting abnormal temperature rise indicators, current disturbance indicators, and action time difference indicators, and calculating the fault node score; a status assessment module, which sets the fault judgment threshold based on historical samples and real-time operating status, and generates a fault diagnosis sub-report and health status based on the comparison result of the fault node score and the fault judgment threshold; and an integration module, which integrates the fault diagnosis sub-reports and health status of all drawers, calculates the health index of each drawer and the overall health index, and generates a test report.

[0091] This embodiment also provides a computer device applicable to the automatic testing method for low-voltage feeder cabinets based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as proposed in the above embodiment.

[0092] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0093] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In summary, this invention utilizes a neuromorphic computing engine to extract equipment parameters and dynamically generate adversarial test waveforms, enabling precise simulation of complex electrical environments under real-world operating conditions, significantly improving the relevance of testing and the fault detection rate. Furthermore, by inputting response signals into a cognitive digital twin to extract spatial temperature distribution data and activating the insulation aging branch in the knowledge graph when temperature rises abnormally to calculate fault node scores, it achieves early identification and quantitative assessment of potential insulation defects, greatly improving the intelligence level, diagnostic accuracy, and operational safety of low-voltage feeder cabinet testing.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automatic testing method for low-voltage feeder cabinets based on artificial intelligence, characterized in that: include, Scan the QR code of the cabinet to be tested to obtain design drawing information, construct a cognitive digital twin and generate an initial knowledge graph; The device parameters in the initial knowledge graph are extracted using a neuromorphic computing engine, and adversarial test waveforms are dynamically generated by driving a spiking neural network. The adversarial test waveform is injected into the corresponding drawer output terminal of the cabinet under test through a dynamic load simulator, and the response signals of current waveform, temperature distribution and circuit breaker action sequence are collected simultaneously. The response signal is input into the cognitive digital twin to extract spatial temperature distribution data. When the local temperature rise rate is detected to exceed the temperature rise threshold, the insulation aging branch of the initial knowledge graph is activated to extract temperature rise anomaly index, current disturbance index and action time difference index, and calculate the fault node score. Based on historical samples and real-time operating status, a fault judgment threshold is set, and a fault diagnosis sub-report and health status are generated based on the comparison results between the fault node score and the fault judgment threshold. Integrate the fault diagnosis sub-reports and health status of all drawers, calculate the health index of each drawer and the overall health index, and generate a test report.

2. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 1, characterized in that: The steps for scanning the QR code of the cabinet to be tested to obtain design drawing information, constructing a cognitive digital twin, and generating an initial knowledge graph are as follows. Scan the QR code of the cabinet to be tested to obtain design drawing information and equipment parameters. Based on the equipment parameters, drive the neural symbolic joint engine, extract visual features of the drawings through visual neural network, and combine first-order logic rules to infer equipment type and connection relationship to generate a preliminary knowledge graph. The device topology relationships in the preliminary knowledge graph are input into the quantum annealing optimizer to obtain the optimal topology structure by solving for the device connection strength and importance weights, and output the optimized electrical connection graph. When a conflict in device parameters is detected, the multi-agent debate engine is activated. By integrating the confidence decisions of agents based on electrical specifications, historical faults, and physical constraints, the conflicting parameters are automatically corrected and a correction log is generated. The optimized electrical connection graph and correction logs are imported into the Hypergraph database to construct a cognitive digital twin. The device nodes, corrected parameters, and topological relationships in the electrical connection graph are encoded into a structured knowledge network to form an initial knowledge graph.

3. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 2, characterized in that: The process involves extracting device parameters from the initial knowledge graph using a neuromorphic computing engine and dynamically generating adversarial test waveforms by driving a spiking neural network. The specific steps are as follows: The circuit breaker operating time constant and transformer characteristic frequency are extracted from the initial knowledge graph to construct the equipment feature vector. The equipment feature vector is input into the neuromorphic computing engine to drive the spiking neural network to generate the fundamental wave test waveform. An adversarial test waveform is generated by injecting true random noise into the fundamental test waveform using a quantum random number generator.

4. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 3, characterized in that: The process involves injecting the adversarial test waveform into the corresponding drawer output terminal of the cabinet under test using a dynamic load simulator, and simultaneously acquiring response signals such as current waveform, temperature distribution, and circuit breaker operating sequence. The specific steps are as follows. Based on the cabinet topology parameters in the cognitive digital twin, the output impedance of the load simulator is dynamically adjusted to match the terminals of the drawer under test; The adversarial test waveform is injected into the drawer terminal after impedance matching, and the three-channel sensor array is triggered by the quantum synchronization clock to synchronously collect the current waveform signal, infrared thermal imaging temperature distribution signal and fiber optic sensing circuit breaker action signal. It integrates current waveform signals, infrared thermal imaging temperature distribution signals, and fiber optic circuit breaker action signals to output a response signal.

5. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 1, characterized in that: The specific steps for calculating the score of the faulty node are as follows: The response signal is input into the cognitive digital twin, the spatial temperature distribution data of the thermal_map field is extracted, the temperature rise rate at each coordinate point is calculated, and the temperature rise threshold is set according to the thermal collapse critical point of the insulation material. When the temperature rise rate exceeds the temperature rise threshold, the insulation aging branch in the initial knowledge graph is activated, and the fault node is obtained by matching the nodes in the initial knowledge graph through heat conduction path tracing. When injecting the fundamental test waveform into the dynamic load simulator, the current waveform and circuit breaker action timing data are collected, and the abnormal temperature rise index, current disturbance index and action time difference index are calculated. The fault node score is obtained by combining the results.

6. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 1, characterized in that: The generation of fault diagnosis sub-reports and health status is based on the dynamic setting of fault judgment thresholds according to the historical fault occurrence probability and real-time equipment status. When the fault node score is greater than the fault judgment threshold, a fault diagnosis sub-report is generated; when the fault node score is less than the fault judgment threshold, it is marked as a healthy state.

7. The automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in claim 6, characterized in that: The process involves integrating fault diagnosis sub-reports and health status data from all drawers, calculating the health index of each drawer and the overall health index, and generating a test report. The specific steps are as follows: Extract the fault diagnosis sub-reports and health status of each drawer, and use the health index calculation formula to convert the fault node scores into drawer health indices; The drawer health index is weighted and summarized to obtain the overall health index, and the health levels are divided. A test report is generated by integrating the fault diagnosis sub-reports of each drawer with the health levels.

8. An automatic testing system for low-voltage feeder cabinets based on artificial intelligence, based on the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in any one of claims 1 to 7, characterized in that: It includes a knowledge graph module, a waveform module, a signal module, a weighting module, a state evaluation module, and an integration module. The knowledge graph module is used to scan the QR code of the cabinet under test to obtain design drawing information, construct a cognitive digital twin, and generate an initial knowledge graph. The waveform module is used to extract device parameters from the initial knowledge graph using a neuromorphic computing engine and dynamically generate adversarial test waveforms by driving a spiking neural network. The signal module is used to inject the adversarial test waveform into the corresponding drawer output terminal of the cabinet under test through the dynamic load simulator, and simultaneously collect the response signals of current waveform, temperature distribution and circuit breaker action sequence. The weighting module is used to input the response signal into the cognitive digital twin to extract spatial temperature distribution data. When the local temperature rise rate is detected to exceed the temperature rise threshold, the insulation aging branch of the initial knowledge graph is activated to extract temperature rise anomaly index, current disturbance index and action time difference index, and calculate the fault node score. The status assessment module is used to set fault judgment thresholds based on historical samples and real-time operating status, and generate fault diagnosis sub-reports and health status based on the comparison results of fault node scores and fault judgment thresholds. The integration module is used to integrate the fault diagnosis sub-reports and health status of all drawers, calculate the health index of each drawer and the overall health index, and generate a test report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the automatic testing method for low-voltage feeder cabinets based on artificial intelligence as described in any one of claims 1 to 7.

Images