Switch cabinet partial discharge mode identification and diagnosis system based on artificial intelligence

Abstract

The invention discloses a switch cabinet partial discharge mode identification and diagnosis system based on artificial intelligence, and relates to the field of switch cabinets, and the system comprises an acquisition module which is used for collecting electromagnetic radiation, temperature strain and sound wave vibration signals generated during the operation of a switch cabinet, and converting the original signals into a digital sequence; the enhancement processing module is used for carrying out noise separation and characteristic signal amplification on the digital sequence and outputting a multi-channel time domain signal set subjected to standardization processing; according to the invention, through multi-dimensional signal acquisition and intelligent processing, noise can be accurately separated and effective features can be amplified, and accurate determination of partial discharge types and positions can be realized in combination with space-time correlation analysis.

Description

Technical Field

[0001] This invention relates to the field of switchgear technology, specifically to an artificial intelligence-based system for partial discharge pattern recognition and diagnosis in switchgear. Background Technology

[0002] Partial discharge in switchgear refers to localized breakdown or ionization caused by excessively high local electric field strength in the internal insulation. Prolonged presence of this phenomenon can accelerate insulation aging, potentially leading to short circuits, equipment damage, and even accidents. Timely detection through testing is crucial to ensure safe equipment operation.

[0003] Patent application No. 202410160777.8 discloses a method and system for multi-source partial discharge pattern recognition of switchgear based on multi-model fusion, comprising: acquiring multi-source partial discharge data of power switchgear equipment, wherein the multi-source partial discharge data includes at least one multi-source partial discharge signal and a discharge phase distribution map of each discharge phase in the at least one multi-source partial discharge signal; identifying the discharge phase distribution map according to a preset image classification model to obtain a first classification label corresponding to the discharge phase distribution map; clustering and segmenting the at least one multi-source partial discharge signal according to a preset multi-source partial discharge clustering model to obtain individual single-source partial discharge signals corresponding to the at least one multi-source partial discharge signal; and processing the individual single-source partial discharge signals respectively. Feature extraction is performed to obtain feature information corresponding to the single-source partial discharge signal. The feature information includes time-domain feature information, frequency-domain feature information, and phase feature information. This application aims to solve the problem that "in practical applications, there may be multiple partial discharge sources discharging simultaneously in switch cabinets, and the above methods are subject to severe electrical interference within the switch cabinet during partial discharge detection, resulting in weak partial discharge signals, low signal-to-noise ratio, and difficulty in accurately identifying discharge characteristics and discharge modes. In addition, since some defects have small partial discharge amounts, they are easily submerged in noise and are not easily visible in the image. Using image recognition alone can lead to missed detections and make it difficult to accurately identify discharge characteristics and discharge modes. Therefore, the accuracy of current detection technologies in signal processing and pattern recognition still needs to be improved."

[0004] However, for high-precision monitoring of partial discharge in switchgear, the more monitoring methods the better.

[0005] To address this, an artificial intelligence-based partial discharge pattern recognition and diagnosis system for switchgear was proposed. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an artificial intelligence-based switch cabinet partial discharge pattern recognition and diagnosis system, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions;

[0008] This invention discloses an artificial intelligence-based partial discharge pattern recognition and diagnosis system for switchgear, comprising:

[0009] The system comprises the following modules: Acquisition module, which collects electromagnetic radiation, temperature strain, and acoustic vibration signals generated during switchgear operation and converts the raw signals into digital sequences; Enhancement module, which performs noise separation and feature signal amplification on the digital sequences and outputs a standardized multi-channel time-domain signal set; Extraction module, which performs spatial correlation modeling on transient features in the time-domain signal set, synchronously analyzes the temporal evolution of signals, and generates a feature encoding set that integrates spatiotemporal dimensions; Diagnosis module, which constructs a partial discharge feature-fault mapping rule base, inputs the feature encoding set into a multi-path inference model for pattern matching, and outputs partial discharge type determination results and spatial positioning parameters; Association module, which receives partial discharge type determination results and spatial positioning parameters, calls a preset historical operating condition database of the switchgear, extracts feature baseline data of the corresponding location under the same operating conditions, and generates a difference map between real-time features and baseline features; and Inference module, which receives the difference map, analyzes the dynamic change of feature difference values ​​over time, calculates the time change rate of partial discharge features, and generates a diagnostic prediction report containing the development trend of partial discharge within a preset future time period.

[0010] Furthermore, the acquisition module includes at least three distributed sensor arrays, and the electromagnetic radiation signal is acquired through an ultra-wideband antenna array with a sampling frequency of [missing information]. satisfy , This represents the highest frequency bandwidth of electromagnetic radiation signals.

[0011] Temperature strain signals are acquired through fiber Bragg grating sensors, with sampling intervals of... According to the rated current of the switchgear Adaptive adjustment, that is:

[0012] In the formula This represents a proportionality constant, with a value ranging from 0.01 to 0.1. ;

[0013] The acoustic vibration signal is acquired through a piezoelectric sensor, and all sensors are synchronized through a timestamp mechanism to ensure that the acquisition time deviation does not exceed 10µs.

[0014] The distributed sensor array consists of an ultra-wideband antenna array, fiber optic grating sensors, and piezoelectric sensors.

[0015] Furthermore, the noise separation process of the enhancement processing module includes:

[0016] A noise threshold function is constructed based on the signal frequency domain characteristics to perform adaptive denoising on the digital sequence. The formula for calculation is:

[0017] ;

[0018] In the formula: This is the adjustment coefficient; The signal variance of the digital sequence; The main frequency of the signal; This is the dominant noise frequency;

[0019] Among them, the adjustment coefficient Based on signal-to-noise ratio Dynamic adjustment to meet =1.5- , ∈[0.01,1], when When it increases Linear decrease, adjustment coefficient Dynamically adjusted based on the ratio of the noise dominant frequency to the signal dominant frequency to satisfy... , ∈(0,1], when the noise frequency is close to the signal frequency, Nonlinear increasing gain; during the amplification of the characteristic signal, the gain of the denoised signal is adjusted, and the gain coefficient... .

[0020] Furthermore, the spatial association modeling operation in the extraction module is as follows:

[0021] Construct a spatial neighborhood matrix for transient features, where the matrix elements Let represent the correlation degree of transient features between the i-th and j-th monitoring points, and ;

[0022] This represents the physical distance between monitoring points i and j. Indicates the spatial attenuation coefficient;

[0023] When simultaneously analyzing the temporal evolution patterns, the nonlinear evolution exponent of the temporal characteristics is calculated. :

[0024] ;

[0025] In the formula: Let be the transient eigenvalue at time t; The length of the time series;

[0026] in, Spatial attenuation coefficient used to quantify the intensity of nonlinear change of features over time. Spatial scale coefficient The value is dynamically adjusted based on the internal structure type of the switchgear; for gas-insulated switchgear, ∈[0.8,1.2]; for solid-insulated switchgear ∈[1.2,1.8]; For mixed-insulation switchgear, ∈[0.5,2.0], and its value is linearly related to the proportion of the insulating medium, that is, for every 10% increase in the proportion of gas in the insulating medium, the value increases. Lowered by 0.1 The area of ​​the monitoring zone inside the switchgear. This is the frequency correction factor. ∈[0.1,0.5], The reference frequency for the partial discharge signal is , and The value is controlled within the range of [0.3, 3.0]. This represents the propagation speed of the partial discharge signal in the medium within the switch cabinet.

[0027] Furthermore, the multi-path reasoning model in the diagnostic module includes three parallel reasoning paths:

[0028] The fusion weights of the output results from each path, based on rule-based reasoning, case-based reasoning, and neural network-based reasoning. satisfy: ;

[0029] in, This represents the inference confidence of the k-th path. , This represents the failure probability output by the k-th path. This represents the historical average failure probability, and the fusion result is the weighted sum of the outputs of each path, i.e. When P ≥ 0.8, it is determined to be a definite partial discharge type;

[0030] The fused spatial positioning parameters are three-dimensional coordinates (x, y, z). , The spatial coordinate components output for the k-th path.

[0031] Furthermore, when the association module extracts feature baseline data, it employs a sliding window filtering process, with the window length L dynamically set according to the switchgear's operating time t. ceil indicates rounding up, t is in hours, and L is in the number of samples;

[0032] The baseline data must meet the stability verification conditions:

[0033] Baseline fluctuation values ​​over 5 consecutive windows , This indicates the initial baseline value; otherwise, baseline recalibration is triggered.

[0034] Furthermore, when the deduction module calculates the time rate of change v of the partial discharge characteristics, it considers the change of the second derivative of the characteristic value, and the calculation formula is as follows:

[0035] ;

[0036] In the formula: Let represent the rate of change of the partial discharge characteristic at time t′; As a regulating factor, ∈[0.3,0.7], when the nonlinearity of the characteristic signal increases The value increases; The time step is calculated and kept consistent with the acquisition interval of the acquisition module; This represents the partial discharge eigenvalue at time t′. Similarly;

[0037] Then the future preset time period The formula for calculating the predicted value of partial discharge characteristics within the range is:

[0038] , express The temporal rate of change of partial discharge characteristics at time t;

[0039] Among them, when When the absolute value is greater than the preset threshold, the integration step size is automatically reduced to Δt′ / 2. The preset threshold ranges from [0.1, 0.5], and the unit is: feature value / second.

[0040] Furthermore, the inference module is internally equipped with optimization logic:

[0041] Based on the historical accuracy feedback from the diagnostic prediction report, the feature encoding dimension of the extraction module and the mapping rule base of the diagnostic module are dynamically adjusted. The adjustment formula for the feature encoding dimension d is as follows:

[0042] d'=d×(1+η×(1-A)), where d' is the adjusted feature encoding dimension, η is the adjustment coefficient, A is the historical diagnostic accuracy, and A∈[0,1];

[0043] The adjustments to the mapping rule base include: setting dynamic weights for each mapping rule. When rule r is correctly activated in historical diagnostics, = ×[1+θ×(1-A)], For the adjusted dynamic weights, when rule r is incorrectly activated, = ×[1-θ×(1-A)], where θ is the rule adjustment coefficient, θ∈[0.1,0.3], and the sum of the weights of all rules remains 1 after adjustment;

[0044] Meanwhile, if the weight of any rule is still ≤0.01 after 3 consecutive adjustments, it will be automatically removed from the rule base and a new rule will be generated based on the newly collected data in the past 3 months to supplement the rule base;

[0045] Specifically, when A≥0.9, θ=0.1; when 0.6≤A<0.9, θ=0.2; and when A<0.6, θ=0.3.

[0046] Furthermore, the diagnostic prediction report in the simulation module includes: partial discharge type determination result, spatial positioning parameters, partial discharge development trend within a preset future time period, and partial discharge risk level;

[0047] The predicted trend of partial discharge development within the future preset time period is the result of the prediction module analyzing the dynamic change law of the feature difference value over time and calculating the time change rate of the partial discharge feature, which is the prediction result of the evolution direction and degree of the partial discharge feature of the switch cabinet in the future set time period.

[0048] The local discharge risk level is:

[0049] ;

[0050] In the formula: These are the weighting coefficients; This represents the maximum difference between the real-time feature and the baseline feature. The time-varying rate of change of the partial discharge characteristic; It is a nonlinear evolution index representing temporal characteristics;

[0051] in, The values ​​are user-defined on the system side, and all are within the range of [0.2, 0.5], and their sum is 1. ∈[0,0.3) is considered low risk. ∈[0.3,0.7) represents medium risk. A value ≥0.7 indicates high risk. By calculating the absolute difference between each feature component in the real-time feature encoding set and the corresponding component of the baseline feature, the maximum value is taken as... The value of .

[0052] Furthermore, the acquisition module is interconnected with an enhancement processing module and an extraction module via a wireless network. The extraction module is interconnected with a diagnostic module via a wireless network. The diagnostic module is interconnected with an association module via a wireless network. The association module is interconnected with a deduction module via a wireless network.

[0053] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0054] This invention provides an AI-based partial discharge pattern recognition and diagnosis system for switchgear. During operation, the system accurately separates noise and amplifies effective features through multi-dimensional signal acquisition and intelligent processing. Combined with spatiotemporal correlation analysis, it accurately determines the type and location of partial discharge. Simultaneously, it generates feature difference maps by associating historical operating data and calculates the time change rate by analyzing dynamic change patterns to accurately predict future partial discharge development trends. Its multi-path reasoning and dynamic optimization mechanism can adaptively adjust according to the diagnostic accuracy, adapting to different types of switchgear and effectively improving the accuracy and timeliness of partial discharge identification. Furthermore, through risk level assessment, it can provide early warning of potential faults, reduce equipment operation risks, and provide comprehensive protection for the safe and stable operation of switchgear, effectively improving the intelligence level and reliability of power equipment operation and maintenance. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0056] Figure 1 This is a schematic diagram of an AI-based switchgear partial discharge pattern recognition and diagnosis system. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0058] The present invention will be further described below with reference to embodiments.

[0059] Example:

[0060] This embodiment presents an AI-based switchgear partial discharge pattern recognition and diagnosis system, such as... Figure 1 As shown, it includes:

[0061] The acquisition module is used to acquire electromagnetic radiation, temperature strain and acoustic vibration signals generated during the operation of the switchgear, and convert the raw signals into digital sequences.

[0062] The acquisition module includes at least three distributed sensor arrays. Electromagnetic radiation signals are acquired through an ultra-wideband antenna array, with a sampling frequency of [missing information]. satisfy , This represents the highest frequency bandwidth of electromagnetic radiation signals.

[0063] Temperature strain signals are acquired through fiber Bragg grating sensors, with sampling intervals of... According to the rated current of the switchgear Adaptive adjustment, that is:

[0064] In the formula This represents a proportionality constant, with a value ranging from 0.01 to 0.1. ;

[0065] The acoustic vibration signal is acquired through a piezoelectric sensor, and all sensors are synchronized through a timestamp mechanism to ensure that the acquisition time deviation does not exceed 10µs.

[0066] The distributed sensor array consists of an ultra-wideband antenna array, fiber optic grating sensors, and piezoelectric sensors.

[0067] The enhancement processing module is used to separate noise and amplify characteristic signals from digital sequences, and output a standardized multi-channel time-domain signal set.

[0068] The noise separation process of the enhancement processing module includes:

[0069] A noise threshold function is constructed based on the frequency domain characteristics of the signal to perform adaptive denoising on digital sequences. The formula for calculation is:

[0070] ;

[0071] In the formula: This is the adjustment coefficient; The signal variance of the digital sequence; The main frequency of the signal; This is the dominant noise frequency;

[0072] Among them, the adjustment coefficient Based on signal-to-noise ratio Dynamic adjustment to meet =1.5- , ∈[0.01,1], when When it increases Linear decrease, adjustment coefficient Dynamically adjusted based on the ratio of the noise dominant frequency to the signal dominant frequency to satisfy... , ∈(0,1], when the noise frequency is close to the signal frequency, Nonlinear increasing gain; during the amplification of the characteristic signal, the gain of the denoised signal is adjusted, and the gain coefficient... ;

[0073] The above formula incorporates the adjustment coefficient. Signal variance The signal and noise main frequencies are used to construct an adaptive noise reduction threshold. Its core lies in dynamically adjusting the threshold by combining the multi-dimensional characteristics of the signal and noise. Compared with a fixed threshold, it can more accurately distinguish between effective signals and noise. In complex electromagnetic environments, it can avoid the effective signals being mistakenly deleted and can fully suppress noise interference.

[0074] The extraction module is used to perform spatial correlation modeling on transient features in the time-domain signal set, simultaneously analyze the temporal evolution law of the signal, and generate a feature encoding set that integrates spatiotemporal dimensions;

[0075] The spatial association modeling operation in the extraction module is as follows:

[0076] Construct a spatial neighborhood matrix for transient features, where the matrix elements Let represent the correlation degree of transient features between the i-th and j-th monitoring points, and ;

[0077] This represents the physical distance between monitoring points i and j. Indicates the spatial attenuation coefficient;

[0078] When simultaneously analyzing the temporal evolution patterns, the nonlinear evolution exponent of the temporal characteristics is calculated. :

[0079] ;

[0080] In the formula: Let be the transient eigenvalue at time t; The length of the time series;

[0081] in, Spatial attenuation coefficient used to quantify the intensity of nonlinear change of features over time. Spatial scale coefficient The value is dynamically adjusted based on the internal structure type of the switchgear; for gas-insulated switchgear, ∈[0.8,1.2]; for solid-insulated switchgear ∈[1.2,1.8]; For mixed-insulation switchgear, ∈[0.5,2.0], and its value is linearly related to the proportion of the insulating medium, that is, for every 10% increase in the proportion of gas in the insulating medium, the value increases. Lowered by 0.1 The area of ​​the monitoring zone inside the switchgear. This is the frequency correction factor. ∈[0.1,0.5], The reference frequency for the partial discharge signal is , and The value is controlled within the range of [0.3, 3.0]. This refers to the propagation speed of the partial discharge signal in the medium within the switchgear.

[0082] The above formula calculates the intensity of nonlinear changes through time series analysis, which can capture the complex evolution of partial discharge characteristics over time. Compared with linear analysis, this formula can effectively identify nonlinear behaviors such as abrupt changes and accelerated changes in characteristics, more accurately reflect the dynamic trend of partial discharge development, and provide richer time series information for subsequent diagnosis.

[0083] The diagnostic module is used to build a partial discharge feature-fault mapping rule base, input the feature encoding set into the multi-path reasoning model for pattern matching, and output the partial discharge type determination result and spatial positioning parameters.

[0084] It should be noted that:

[0085] The relationship between the pattern matching operations of the "partial discharge feature-fault mapping rule base" and the "multi-path inference model":

[0086] The rule base is the foundation for pattern matching. It stores the association rules between different partial discharge feature codes and corresponding fault types and positioning parameters, providing an initial judgment standard for inference. The multi-path inference model is the core carrier for performing the matching operation. After receiving the feature code set output by the spatiotemporal feature fusion extraction module, it calls the association rules in the rule base. Through multi-path parallel inference, it compares the feature codes with the feature items in the rule base layer by layer, calculates weights and performs cross-validation. Finally, it outputs the partial discharge type determination result and spatial positioning parameters with the highest rule matching degree.

[0087] The multi-path inference model in the diagnostic module includes three parallel inference paths:

[0088] The fusion weights of the output results from each path, based on rule-based reasoning, case-based reasoning, and neural network-based reasoning. satisfy: ;

[0089] in, This represents the inference confidence of the k-th path. , This represents the failure probability output by the k-th path. This represents the historical average failure probability, and the fusion result is the weighted sum of the outputs of each path, i.e. When P ≥ 0.8, it is determined to be a definite partial discharge type;

[0090] It should be noted that:

[0091] "Defining the type of partial discharge" refers to the specific type of partial discharge in the switchgear that can be clearly distinguished after pattern matching is performed by the multi-path inference model of the diagnostic module.

[0092] Partial discharge is a localized discharge phenomenon caused by insulation defects inside switchgear. Different insulation defects (such as surface contamination, internal air gaps, and metal tip burrs) will produce electromagnetic radiation, temperature strain, and acoustic vibration signals with different characteristics. The diagnostic module constructs a partial discharge feature-fault mapping rule base and performs matching analysis on the feature encoding set fused with spatiotemporal dimensions generated by the extraction module. The specific discharge type (such as corona discharge, surface discharge, internal discharge, etc.) is finally determined as the "clear partial discharge type," which is one of the core outputs of the system for accurate diagnosis of partial discharge faults in switchgear.

[0093] The fused spatial positioning parameters are three-dimensional coordinates (x, y, z). , The spatial coordinate components output for the k-th path;

[0094] The association module is used to receive the partial discharge type determination result and spatial positioning parameters, call the preset switchgear historical operating condition database, extract the characteristic baseline data of the corresponding location under the same operating condition, and generate a difference map between real-time features and baseline features.

[0095] When the correlation module extracts feature baseline data, it uses a sliding window filter. The window length L is dynamically set according to the operating time t of the switchgear. ceil indicates rounding up, t is in hours, and L is in the number of samples;

[0096] Baseline data must meet stability verification conditions:

[0097] Baseline fluctuation values ​​over 5 consecutive windows , This indicates the initial baseline value; otherwise, baseline recalibration is triggered.

[0098] The extrapolation module is used to receive the difference map, analyze the dynamic change of the feature difference value over time, calculate the time change rate of the partial discharge feature, and generate a diagnostic prediction report containing the development trend of partial discharge within a preset future period.

[0099] When the derivation module calculates the time rate of change v of the partial discharge characteristic, it considers the change of the second derivative of the characteristic value, and the calculation formula is as follows:

[0100] ;

[0101] In the formula: Let represent the rate of change of the partial discharge characteristic at time t′; As a regulating factor, ∈[0.3,0.7], when the nonlinearity of the characteristic signal increases The value increases; The time step is calculated and kept consistent with the acquisition interval of the acquisition module; This represents the partial discharge eigenvalue at time t′. Similarly;

[0102] The above formula introduces a second derivative and an adjustment factor. The adjustment factor increases as the degree of nonlinearity of the characteristic increases, enhancing the sensitivity to the rate of change. Compared to using only the first derivative, this formula can capture the acceleration of characteristic changes, more accurately reflecting the speed of partial discharge development and providing more reliable rate of change data for subsequent trend prediction.

[0103] Then the future preset time period The formula for calculating the predicted value of partial discharge characteristics within the range is:

[0104] , express The temporal rate of change of partial discharge characteristics at time t;

[0105] Among them, when When the absolute value is greater than a preset threshold, the integration step size is automatically reduced to Δt′ / 2. The preset threshold ranges from [0.1, 0.5], and the unit is: feature value / second.

[0106] The above formula is obtained by integrating the rate of change over time. Calculate, and in When the threshold is exceeded, the integration step size is reduced. This design improves efficiency with a large step size when the feature changes slowly and improves accuracy with a small step size when it changes rapidly, balancing prediction efficiency and accuracy. It can effectively reduce prediction errors, especially in the rapid development stage of partial discharge.

[0107] The deduction module has built-in optimization logic:

[0108] Based on the historical accuracy feedback from the diagnostic prediction report, the feature encoding dimension of the extraction module and the mapping rule base of the diagnostic module are dynamically adjusted. The adjustment formula for the feature encoding dimension d is as follows:

[0109] d'=d×(1+η×(1-A)), where d' is the adjusted feature encoding dimension, η is the adjustment coefficient, A is the historical diagnostic accuracy, and A∈[0,1];

[0110] The above formula is dynamically adjusted based on historical diagnostic accuracy. When the accuracy is low, the dimension is increased to retain more features, and when the accuracy is high, the dimension is reduced to simplify the model. This mechanism realizes the adaptive optimization of the model, which can ensure accuracy through high dimension under complex conditions and improve efficiency through low dimension under stable conditions.

[0111] The adjustments to the mapping rule base include: setting dynamic weights for each mapping rule. When rule r is correctly activated in historical diagnostics, = ×[1+θ×(1-A)], For the adjusted dynamic weights, when rule r is incorrectly activated, = ×[1-θ×(1-A)], where θ is the rule adjustment coefficient, θ∈[0.1,0.3], and the sum of the weights of all rules remains 1 after adjustment;

[0112] The above formula increases the weight when a rule is correctly activated and decreases the weight when it is incorrectly activated. Furthermore, θ is adjusted according to the accuracy level. This design allows the rule base to update itself, strengthens effective rules, weakens or even eliminates invalid rules, and enhances the system's adaptability to new failure modes.

[0113] Meanwhile, if the weight of any rule is still ≤0.01 after 3 consecutive adjustments, it will be automatically removed from the rule base and a new rule will be generated based on the newly collected data in the past 3 months to supplement the rule base;

[0114] Specifically, when A ≥ 0.9, θ = 0.1; when 0.6 ≤ A < 0.9, θ = 0.2; and when A < 0.6, θ = 0.3.

[0115] The diagnostic prediction report in the simulation module includes: partial discharge type determination results, spatial positioning parameters, partial discharge development trend within the preset future time period, and partial discharge risk level;

[0116] The future partial discharge development trend within the preset time period is the prediction result generated by the simulation module after analyzing the dynamic change law of the characteristic difference value over time and calculating the time change rate of the partial discharge characteristics, which is the prediction result of the evolution direction and degree of the partial discharge characteristics of the switchgear in the future set time period.

[0117] The partial discharge risk level is:

[0118] ;

[0119] In the formula: These are the weighting coefficients; This represents the maximum difference between the real-time feature and the baseline feature. The time-varying rate of change of the partial discharge characteristic; It is a nonlinear evolution index representing temporal characteristics;

[0120] in, The values ​​are user-defined on the system side, and all are within the range of [0.2, 0.5], and their sum is 1. ∈[0,0.3) is considered low risk. ∈[0.3,0.7) represents medium risk. A value ≥0.7 indicates high risk. By calculating the absolute difference between each feature component in the real-time feature encoding set and the corresponding component of the baseline feature, the maximum value is taken as... The possible values ​​of ;

[0121] The above formula integrates the maximum difference between real-time and baseline features, the rate of change over time, and the nonlinear evolution index, and calculates the risk value by combining custom weights. This formula assesses risk from multiple dimensions, including the degree of feature difference, the rate of change, and the complexity of evolution. Compared with a single indicator, it can more comprehensively reflect the actual degree of harm of partial discharge and provide users with an intuitive basis for risk judgment.

[0122] The acquisition module interacts with the enhancement processing module and the extraction module via a wireless network. The extraction module interacts with the diagnostic module via a wireless network. The diagnostic module interacts with the correlation module via a wireless network. The correlation module interacts with the inference module via a wireless network.

[0123] In this embodiment, the acquisition module collects electromagnetic radiation, temperature strain, and acoustic vibration signals generated during the operation of the switchgear and converts the raw signals into digital sequences. The enhancement processing module then performs noise separation and feature signal amplification on the digital sequences, outputting a standardized multi-channel time-domain signal set. The extraction module further performs spatial correlation modeling on the transient features in the time-domain signal set, synchronously analyzes the temporal evolution of the signals, and generates a feature encoding set that integrates spatiotemporal dimensions. The diagnosis module then constructs a partial discharge feature-fault mapping rule library, inputs the feature encoding set into a multi-path inference model for pattern matching, and outputs the partial discharge type determination result and spatial positioning parameters. The correlation module receives the partial discharge type determination result and spatial positioning parameters, calls the preset historical operating condition database of the switchgear, extracts the feature baseline data of the corresponding location under the same operating conditions, and generates a difference map between the real-time features and the baseline features. Finally, the inference module receives the difference map, analyzes the dynamic change law of the feature difference value over time, calculates the time change rate of the partial discharge feature, and generates a diagnostic prediction report containing the development trend of partial discharge in the future preset time period.

[0124] The system in the above embodiments can accurately capture electromagnetic, temperature and vibration signals during switchgear operation, effectively separate noise and enhance key features, accurately identify the type and location of partial discharge by combining spatiotemporal features, clearly present real-time differences by comparing historical operating condition baseline data, predict the development trend of partial discharge, assess the risk level, dynamically optimize diagnostic logic, greatly improve diagnostic accuracy and foresight, help identify hidden dangers in advance, reduce downtime due to failure, and ensure the safe and stable operation of equipment.

[0125] The following is an application example of the system described in the above embodiments:

[0126] In a 220kV substation, the 10kV section of the hybrid insulation switchgear has been in operation for 5 years. In order to detect potential partial discharge faults in a timely manner, the operation and maintenance team deployed an artificial intelligence-based switchgear partial discharge pattern recognition and diagnosis system to monitor and diagnose the switchgear in real time.

[0127] I. Signal Acquisition Stage

[0128] The system acquisition module activates three sets of distributed sensor arrays to simultaneously acquire three types of signals:

[0129] Ultra-wideband antenna arrays acquire electromagnetic radiation signals. Based on the highest frequency bandwidth of the signal, the sampling frequency is set to twice that frequency to ensure complete capture of signal characteristics.

[0130] The fiber optic grating sensor collects temperature strain signals. The rated current of this switchgear is 630A, the proportionality constant is 0.05, and the sampling interval is calculated based on the ratio of the proportionality constant to the rated current, which is approximately 2.86 seconds.

[0131] The piezoelectric sensor collects acoustic vibration signals. Through a timestamp synchronization mechanism, the acquisition time deviation of the three types of sensors is controlled within 8µs, which meets the requirement of not exceeding 10µs.

[0132] II. Signal Enhancement Processing Stage

[0133] The enhancement processing module processes the acquired digital sequences:

[0134] During noise separation, a noise threshold function is constructed based on the signal frequency domain characteristics. The signal-to-noise ratio (SNR) is calculated to be 0.5, and the adjustment coefficient is set to 1.0 (calculated by subtracting the SNR from 1.5). The ratio of the noise's dominant frequency to the signal's dominant frequency is 0.8, and the adjustment coefficient increases non-linearly to 0.8 as the noise frequency approaches the signal frequency. Combining the signal variance and dominant frequency, the noise threshold is finally determined, completing adaptive denoising.

[0135] During the characteristic signal amplification stage, the gain of the denoised signal is adjusted, and the gain coefficient is set to 1.2 to effectively amplify the partial discharge characteristic signal.

[0136] After processing, a standardized multi-channel time-domain signal set is output, laying the foundation for subsequent analysis.

[0137] III. Feature Extraction Stage

[0138] The extraction module processes the time-domain signal set:

[0139] In spatial correlation modeling, a spatial neighborhood matrix of transient features is constructed. Two monitoring points i and j are selected, with a physical distance of 2 meters. Since the switchgear is a hybrid insulation type with a gas content of 30% in the insulation medium, the spatial scale coefficient is set to 1.2 (the initial value is adjusted down by 0.1 for every 10% increase in gas content). The monitoring area is 10 square meters, the frequency correction coefficient is set to 0.3, the reference frequency of the partial discharge signal is 50 MHz, and the propagation speed is 0.2 m / µs. The calculated spatial attenuation coefficient is 1.5. Finally, the transient feature correlation degree between the two monitoring points is 0.05 (calculated using an exponential function, i.e., the result of exp(-1.5×2)).

[0140] The evolution of the time series was analyzed synchronously. The time series length was 1000. By calculating the change of transient characteristic values ​​at time t and the adjacent time, the nonlinear evolution index of the time series characteristics was found to be 0.8, which quantified the intensity of the nonlinear change of the characteristics over time.

[0141] Finally, a feature encoding set that integrates spatiotemporal dimensions is generated.

[0142] IV. Partial Radiation Diagnosis Stage

[0143] The diagnostic module calls the partial discharge feature-fault mapping rule base and starts the multi-path inference model:

[0144] The confidence levels of the three parallel reasoning paths (rule-based reasoning, case-based reasoning, and neural network-based reasoning) are 0.3, 0.2, and 0.5, respectively, and the corresponding fusion weights are also 0.3, 0.2, and 0.5.

[0145] The failure probabilities of each path output are 0.9, 0.8, and 0.85, respectively. After weighted summation, the fusion result is 0.855. Since it is greater than 0.8, it is determined to be a clear partial discharge type - floating discharge.

[0146] The spatial positioning parameters are fused and calculated to obtain three-dimensional coordinates (1.2 meters, 0.8 meters, 0.5 meters), which accurately locates the partial discharge position.

[0147] V. Feature Correlation Analysis Stage

[0148] The associated module calls the historical operating condition database of the switchgear to extract the characteristic baseline data of the corresponding positioning point under the same operating conditions:

[0149] The switchgear operates for 5 years (approximately 43,800 hours). The length of the sliding window is calculated by taking one sample every 1,000 hours of operation, resulting in 44 samples of window length.

[0150] The baseline data was validated for stability. The baseline fluctuation value was less than 5% of the initial baseline value for five consecutive windows, and the validation was passed.

[0151] A difference map between real-time features and baseline features is generated, showing that the maximum difference between the two is 0.3.

[0152] VI. Trend Analysis and Risk Assessment Phase

[0153] The inference module performs analysis based on the difference map:

[0154] When calculating the time change rate of partial discharge characteristics, since the nonlinearity of the characteristic signal is moderate, the adjustment factor is set to 0.5, and the time step is consistent with the acquisition interval (1 second). Combining the partial discharge characteristic values ​​at time t′ and the subsequent 2 seconds (2, 3, and 5 respectively), the time change rate v is calculated to be 1.2 characteristic values / second.

[0155] When predicting the development trend of partial discharge in the next 24 hours, since the absolute value of v (1.2) is greater than the preset threshold of 0.3, the integration step size is automatically reduced from 1 second to 0.5 seconds; after integration calculation, it is predicted that the partial discharge characteristic value will reach 8 after 24 hours.

[0156] In the risk level calculation, the weighting coefficients are 0.3, 0.4, and 0.3 respectively (totaling 1). Combined with the maximum difference value of 0.3, the time change rate of 1.2, and the nonlinear evolution index of 0.8, the risk value R is 0.81, which is judged as high risk.

[0157] VII. System Optimization and Diagnostic Report

[0158] The system inference module is optimized based on the historical diagnostic accuracy (0.75):

[0159] The feature encoding dimension is adjusted to 1.05 times the original dimension (calculated by adding 0.2 to 1 and multiplying by the accuracy deviation).

[0160] A certain mapping rule was correctly activated due to historical diagnosis, and its weight was adjusted to 1.05 times the original weight (calculated by adding 0.2 to the accuracy deviation), while the sum of the weights of all rules remained at 1.

[0161] The final diagnostic prediction report includes: the partial discharge type is suspended discharge, the spatial location is (1.2 m, 0.8 m, 0.5 m), the partial discharge characteristic value will rise to 8 in the next 24 hours, the risk level is high risk, and it is recommended to arrange maintenance immediately.

[0162] In summary, during operation, the system in the above embodiments can accurately separate noise and amplify effective features through multi-dimensional signal acquisition and intelligent processing. Combined with spatiotemporal correlation analysis, it can accurately determine the type and location of partial discharge. At the same time, it generates feature difference maps by associating historical operating data, calculates the time change rate by analyzing dynamic change patterns, and accurately predicts the future development trend of partial discharge. Its multi-path reasoning and dynamic optimization mechanism can adaptively adjust with the diagnostic accuracy, adapting to different types of switchgear and effectively improving the accuracy and timeliness of partial discharge identification. In addition, through risk level assessment, it can provide early warning of potential faults, reduce equipment operation risks, provide comprehensive protection for the safe and stable operation of switchgear, and effectively improve the intelligence level and reliability of power equipment operation and maintenance.

[0163] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI-based system for partial discharge pattern recognition and diagnosis in switchgear, characterized in that, include: The acquisition module is used to acquire electromagnetic radiation, temperature strain and acoustic vibration signals generated during the operation of the switchgear, and convert the raw signals into digital sequences. The enhancement processing module is used to separate noise and amplify characteristic signals from digital sequences, and output a standardized multi-channel time-domain signal set. The extraction module is used to perform spatial correlation modeling on transient features in the time-domain signal set, simultaneously analyze the temporal evolution law of the signal, and generate a feature encoding set that integrates spatiotemporal dimensions; The diagnostic module is used to build a partial discharge feature-fault mapping rule base, input the feature encoding set into the multi-path reasoning model for pattern matching, and output the partial discharge type determination result and spatial positioning parameters. The association module is used to receive the partial discharge type determination result and spatial positioning parameters, call the preset historical operating condition database of switchgear, extract the characteristic baseline data of the corresponding location under the same operating condition, and generate a difference map between real-time features and baseline features. The extrapolation module receives the difference map, analyzes the dynamic change of the feature difference values ​​over time, calculates the time change rate of the partial discharge characteristics, and generates a diagnostic prediction report containing the development trend of partial discharge within a preset future time period.

2. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The acquisition module includes at least three distributed sensor arrays, and the electromagnetic radiation signal is acquired through an ultra-wideband antenna array with a sampling frequency of [missing information]. satisfy , This represents the highest frequency bandwidth of electromagnetic radiation signals. Temperature strain signals are acquired through fiber Bragg grating sensors, with sampling intervals of... According to the rated current of the switchgear Adaptive adjustment, that is: In the formula This represents a proportionality constant, with a value ranging from 0.01 to 0.

1. ; The acoustic vibration signal is acquired through a piezoelectric sensor, and all sensors are synchronized through a timestamp mechanism to ensure that the acquisition time deviation does not exceed 10µs. The distributed sensor array consists of an ultra-wideband antenna array, fiber optic grating sensors, and piezoelectric sensors.

3. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The noise separation process of the enhancement processing module includes: A noise threshold function is constructed based on the signal frequency domain characteristics to perform adaptive denoising on the digital sequence. The formula for calculation is: ； In the formula: This is the adjustment coefficient; The signal variance of the digital sequence; The main frequency of the signal; This is the dominant noise frequency; Among them, the adjustment coefficient Based on signal-to-noise ratio Dynamic adjustment to meet =1.5- , ∈[0.01,1], when When it increases Linear decrease, adjustment coefficient Dynamically adjusted based on the ratio of the noise dominant frequency to the signal dominant frequency to satisfy... , ∈(0,1], when the noise frequency is close to the signal frequency, Nonlinear increasing gain; during the amplification of the characteristic signal, the gain of the denoised signal is adjusted, and the gain coefficient... .

4. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The spatial association modeling operation in the extraction module is as follows: Construct a spatial neighborhood matrix for transient features, where the matrix elements Let represent the correlation degree of transient features between the i-th and j-th monitoring points, and ; This represents the physical distance between monitoring points i and j. Indicates the spatial attenuation coefficient; When simultaneously analyzing the temporal evolution patterns, the nonlinear evolution exponent of the temporal characteristics is calculated. : ； In the formula: Let be the transient eigenvalue at time t; The length of the time series; in, Spatial attenuation coefficient used to quantify the intensity of nonlinear change of features over time. Spatial scale coefficient The value is dynamically adjusted based on the internal structure type of the switchgear; for gas-insulated switchgear, ∈[0.8,1.2]; For solid-insulated switchgear ∈[1.2,1.8]; For mixed-insulation switchgear, ∈[0.5,2.0], and its value is linearly related to the proportion of the insulating medium, that is, for every 10% increase in the proportion of gas in the insulating medium, the value increases. Lowered by 0.1 The area of ​​the monitoring zone inside the switchgear. This is the frequency correction factor. ∈[0.1,0.5], The reference frequency for the partial discharge signal is , and The value is controlled within the range of [0.3, 3.0]. This represents the propagation speed of the partial discharge signal in the medium within the switch cabinet.

5. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The diagnostic module includes a multi-path inference model comprising three parallel inference paths: The fusion weights of the output results from each path, based on rule-based reasoning, case-based reasoning, and neural network-based reasoning. satisfy: ; in, This represents the inference confidence of the k-th path. , This represents the failure probability output by the k-th path. This represents the historical average failure probability, and the fusion result is the weighted sum of the outputs of each path, i.e. When P ≥ 0.8, it is determined to be a definite partial discharge type; The fused spatial positioning parameters are three-dimensional coordinates (x, y, z). , The spatial coordinate components output for the k-th path.

6. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, When the correlation module extracts feature baseline data, it uses a sliding window filter. The window length L is dynamically set according to the operating time t of the switchgear. ceil indicates rounding up, t is in hours, and L is in the number of samples; The baseline data must meet the stability verification conditions: Baseline fluctuation values ​​over 5 consecutive windows , This indicates the initial baseline value; otherwise, baseline recalibration is triggered.

7. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, When the deduction module calculates the time-varying rate v of the partial discharge characteristic, it considers the change of the second derivative of the characteristic value, and the calculation formula is as follows: ； In the formula: Let represent the rate of change of the partial discharge characteristic at time t′; As a regulating factor, ∈[0.3,0.7], when the nonlinearity of the characteristic signal increases The value increases; The time step is calculated and kept consistent with the acquisition interval of the acquisition module; This represents the partial discharge eigenvalue at time t′. Similarly; Then the future preset time period The formula for calculating the predicted value of partial discharge characteristics within the range is: , express The temporal rate of change of partial discharge characteristics at time t; Among them, when When the absolute value is greater than the preset threshold, the integration step size is automatically reduced to Δt′ / 2. The preset threshold ranges from [0.1, 0.5], and the unit is: feature value / second.

8. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The inference module has internal optimization logic: Based on the historical accuracy feedback from the diagnostic prediction report, the feature encoding dimension of the extraction module and the mapping rule base of the diagnostic module are dynamically adjusted. The adjustment formula for the feature encoding dimension d is as follows: d'=d×(1+η×(1-A)), where d' is the adjusted feature encoding dimension, η is the adjustment coefficient, A is the historical diagnostic accuracy, and A∈[0,1]; The adjustments to the mapping rule base include: setting dynamic weights for each mapping rule. When rule r is correctly activated in historical diagnostics, = ×[1+θ×(1-A)], For the adjusted dynamic weights, when rule r is incorrectly activated, = ×[1-θ×(1-A)], where θ is the rule adjustment coefficient, θ∈[0.1,0.3], and the sum of the weights of all rules remains 1 after adjustment; Meanwhile, if the weight of any rule is still ≤0.01 after 3 consecutive adjustments, it will be automatically removed from the rule base and a new rule will be generated based on the newly collected data in the past 3 months to supplement the rule base; Specifically, when A≥0.9, θ=0.1; when 0.6≤A<0.9, θ=0.2; and when A<0.6, θ=0.

3.

9. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The diagnostic prediction report in the simulation module includes: partial discharge type determination result, spatial positioning parameters, partial discharge development trend in the future preset time period, and partial discharge risk level; The predicted trend of partial discharge development within the future preset time period is the result of the prediction module analyzing the dynamic change law of the feature difference value over time and calculating the time change rate of the partial discharge feature, which is the prediction result of the evolution direction and degree of the partial discharge feature of the switch cabinet in the future set time period. The risk level of the partial discharge is: ； In the formula: These are the weighting coefficients; This represents the maximum difference between the real-time feature and the baseline feature. The time-varying rate of change of the partial discharge characteristic; It is a nonlinear evolution index representing temporal characteristics; in, The values ​​are user-defined on the system side, and all are within the range of [0.2, 0.5], and their sum is 1. ∈[0,0.3) is considered low risk. ∈[0.3,0.7) represents medium risk. A value ≥0.7 indicates high risk. By calculating the absolute difference between each feature component in the real-time feature encoding set and the corresponding component of the baseline feature, the maximum value is taken as... The value of .

10. The AI-based switchgear partial discharge pattern recognition and diagnosis system according to claim 1, characterized in that, The acquisition module is interconnected with the enhancement processing module and the extraction module via a wireless network. The extraction module is interconnected with the diagnostic module via a wireless network. The diagnostic module is interconnected with the correlation module via a wireless network. The correlation module is interconnected with the inference module via a wireless network.

Images