Method and device for partial discharge fault identification for pole-mounted circuit breakers

Abstract

This invention relates to the field of fault diagnosis technology and discloses a method for identifying partial discharge faults in pole-mounted circuit breakers. The method includes: acquiring radio frequency (RF) and ultrasonic signals collected when partial discharge occurs in the pole-mounted circuit breaker; constructing a partial discharge phase distribution (PRPD) map based on the RF and ultrasonic signals; constructing a PRPD map feature sample dataset using the image features of the PRPD grayscale image corresponding to the PRPD map; and inputting the PRPD map feature sample dataset into a pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker. By organically combining the collaborative sensing of RF and ultrasonic signals with map-based intelligent identification, the accuracy of partial discharge fault identification in pole-mounted circuit breakers is improved. This invention also discloses a device for identifying partial discharge faults in pole-mounted circuit breakers.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and for example to a method and apparatus for identifying partial discharge faults in pole-mounted circuit breakers. Background Technology

[0002] The safety of the power grid distribution network is a crucial foundation for the stable operation of the power system. As core equipment in the power grid distribution network system, pole-mounted circuit breakers are exposed to the complex outdoor environment for extended periods. Affected by factors such as temperature, humidity, pollution, and mechanical stress, their internal insulation is prone to aging and defects such as air gaps, which can lead to partial discharge. If partial discharge faults are not identified and addressed in a timely manner, they will gradually evolve into insulation breakdown, causing the pole-mounted circuit breaker to malfunction, fail to operate, or even explode, seriously threatening the safe operation of the distribution network. Therefore, timely and accurate identification of internal partial discharge faults in pole-mounted circuit breakers is a vital prerequisite for ensuring the reliable operation of the distribution network.

[0003] In related technologies, fault diagnosis of pole-mounted circuit breakers is performed using electrical modal data (such as CT three-phase current, opening and closing coil current, and auxiliary contact voltage signals) and mechanical modal data (such as triaxial vibration signals of the operating mechanism box and spindle angular displacement signals).

[0004] In the process of implementing the embodiments of this disclosure, at least the following problems were found in the related art: Using electrical and mechanical modal data makes it difficult to effectively detect and identify early partial discharge defects inside the high-voltage pole of the pole-mounted circuit breaker, resulting in low accuracy in identifying partial discharge faults in the pole-mounted circuit breaker.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To address the deficiencies in the existing technology, the present invention provides a method and apparatus for identifying partial discharge faults in pole-mounted circuit breakers, thereby improving the accuracy of partial discharge fault identification in pole-mounted circuit breakers.

[0007] According to one aspect of the present invention, the present invention provides a method for partial discharge fault identification of a pole-mounted circuit breaker, comprising: acquiring radio frequency signals and ultrasonic signals collected when partial discharge occurs in the pole-mounted circuit breaker; constructing a partial discharge phase distribution (PRPD) map based on the radio frequency signals and ultrasonic signals; constructing a PRPD map feature sample dataset using image features of the PRPD grayscale image corresponding to the PRPD map; and inputting the PRPD map feature sample dataset into a pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker.

[0008] Optionally, based on radio frequency (RF) signals and ultrasonic signals, a partial discharge phase distribution (PRPD) map is constructed, including: determining the power frequency phase range and discharge amplitude range of each discharge pulse according to the time-domain data of the RF and ultrasonic signals, dividing the power frequency phase range into multiple first intervals, and dividing the discharge amplitude range into multiple second intervals; statistically analyzing the frequency of each discharge pulse in different combinations of the first and second intervals for each RF and ultrasonic signal, generating an RF PRPD map and an ultrasonic PRPD map with power frequency phase as the first dimension, discharge amplitude as the second dimension, and frequency as the statistical value; and performing map layer fusion of the RF PRPD map and the ultrasonic PRPD map to obtain the PRPD map.

[0009] Optionally, the image features of the PRPD grayscale image include: Hu invariant moments used to characterize the morphological features of the PRPD grayscale image; and the mean and standard deviation used to characterize the statistical features of the PRPD grayscale image.

[0010] Optionally, the PRPD image feature sample dataset is input into a pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker. This includes: inputting the PRPD image feature sample dataset into a pre-trained first partial discharge fault identification model, which performs classification and identification based on image features and outputs preliminary identification results; inputting the preliminary identification results into a second partial discharge fault identification model, which performs decision fusion on the preliminary identification results, calculates the basic probability assignment (BPA) value for different partial discharge faults, performs evidence fusion, and outputs the final identification result based on the confidence level after evidence fusion.

[0011] Optionally, the first partial discharge fault identification model is a classification model that combines a one-dimensional convolutional neural network with a support vector machine classifier; the second partial discharge fault identification model is a KNN-DS multi-information fusion model based on the K-nearest neighbor algorithm and DS evidence theory.

[0012] Optionally, the first partial discharge fault identification model is trained as follows: obtain PRPD map feature sample data containing known partial discharge fault type labels; wherein, the PRPD map feature sample dataset includes a training set and a validation set; use the training set to perform supervised training on the initial classification model, and use the validation set to validate the trained initial classification model, thereby obtaining the trained first partial discharge fault identification model.

[0013] Optionally, the second partial discharge fault identification model outputs the final identification result as follows: Based on Euclidean distance, it searches for the K nearest neighbor samples of the sample to be identified in the PRPD map feature sample dataset, and calculates the support of the sample to be identified for each partial discharge fault type based on the distribution of the K nearest neighbor samples, and generates a BPA value after normalization; it performs evidence fusion on multiple BPA values ​​based on the Dempster combination rule to obtain a joint BPA value; based on the joint BPA value, it selects the partial discharge fault type with the highest confidence as the final identification result according to the maximum confidence decision criterion.

[0014] Optionally, the multiple BPA values ​​include a first BPA value generated based on radio frequency signal characteristics and a second BPA value generated based on ultrasonic signal characteristics; evidence fusion of the multiple BPA values ​​is performed based on Dempster's combination rule to obtain a joint BPA value, including: calculating a conflict coefficient K based on the first BPA value and the second BPA value; wherein, the conflict coefficient K characterizes the degree of conflict between the first BPA value and the second BPA value for the support of each partial discharge fault type; when the conflict coefficient K is less than 1, the first BPA value and the second BPA value are orthogonally summed based on Dempster's combination rule, and the joint BPA value is obtained after normalization.

[0015] Optionally, the training dataset for the pre-trained partial discharge fault identification model is obtained as follows: Based on the simulated experimental platform for partial discharge faults inside the pole-mounted circuit breaker, different types of partial discharge fault defects are set in the high-voltage pole of the pole-mounted circuit breaker, and partial discharge is generated after the operating voltage is applied; corresponding radio frequency signals and ultrasonic signals are collected through radio frequency sensors and ultrasonic sensors; a PRPD map is constructed based on the collected radio frequency signals and ultrasonic signals, the image features of the PRPD grayscale image corresponding to the PRPD map are extracted, and the image features are used to construct a PRPD map feature sample dataset with partial discharge fault type labels as the training dataset.

[0016] According to another aspect of the present invention, the present invention also provides a partial discharge fault identification device for a pole-mounted circuit breaker, including a processor and a memory storing program instructions, the processor being configured to execute the aforementioned partial discharge fault identification method for a pole-mounted circuit breaker when executing the program instructions.

[0017] The method and apparatus for identifying partial discharge faults in pole-mounted circuit breakers provided by this invention have at least the following beneficial technical effects: In the signal acquisition stage, both radio frequency (RF) and ultrasonic signals are simultaneously acquired when partial discharge occurs in the pole-mounted circuit breaker. Leveraging the inherent advantages of RF signals' high sensitivity to electromagnetic transients during discharge and the immunity of ultrasonic signals to strong electromagnetic interference, complementary synergy between the two signal types at the information level is achieved, improving the completeness and anti-interference capability of partial discharge fault information from the source. Simultaneously, in the identification stage, a partial discharge fault identification (PDF) map is constructed based on the RF and ultrasonic signals, and image features are extracted from the grayscale images. This transforms PDF fault identification into a mature image pattern recognition problem. Utilizing the deep feature mining capabilities of a pre-trained PDF fault identification model, the image features contained in the map are automatically extracted and judged, achieving intelligent perception and accurate classification of the partial discharge defect state. Thus, through the organic combination of collaborative perception of RF and ultrasonic signals and map-based intelligent identification, the accuracy of PDF fault identification in pole-mounted circuit breakers is improved. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a partial discharge fault identification method for pole-mounted circuit breakers provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another method for identifying partial discharge faults in pole-mounted circuit breakers provided in an embodiment of the present invention; Figure 3 A flowchart illustrating another method for identifying partial discharge faults in pole-mounted circuit breakers provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating another method for identifying partial discharge faults in pole-mounted circuit breakers provided by an embodiment of the present invention. Figure 5 A flowchart illustrating another method for identifying partial discharge faults in pole-mounted circuit breakers provided in an embodiment of the present invention; Figure 6 This is a training loss function graph of PRPD map features of an ultrasound signal provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a partial discharge fault identification device for a pole-mounted circuit breaker provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a partial discharge fault simulation experimental platform for pole-mounted circuit breakers provided in an embodiment of the present invention. Detailed Implementation

[0019] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely 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.

[0020] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0021] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0022] In this embodiment of the invention, the pole-mounted circuit breaker is a high-voltage switching device installed on the pole of an overhead power line in a power distribution network. When a fault such as a short circuit or overload occurs on the line, the pole-mounted circuit breaker can quickly and reliably disconnect the fault current, isolate the faulty section, and protect the normal operation of the rest of the power grid.

[0023] Partial discharge faults in pole-mounted circuit breakers refer to insulation partial discharge defects occurring inside the high-voltage poles of pole-mounted circuit breakers. The essence of this type of defect is an early deterioration state in which the insulation system undergoes partial breakdown under the action of a strong electric field, but the overall insulation has not yet been penetrated. If it is not identified and intervened in time, it will gradually develop into an insulation breakdown fault.

[0024] For example, partial discharge faults in pole-mounted circuit breakers include air gap discharge defects, tip discharge defects, and surface discharge defects. Air gap discharge defects refer to tiny bubbles or voids formed inside insulating materials (such as epoxy resin castings) due to manufacturing defects or material aging. Under the action of a high-voltage electric field, the gas inside the bubbles experiences a higher field strength due to its lower dielectric constant, resulting in partial discharge defects. Tip discharge defects refer to protrusions with extremely small radii of curvature formed on the surface of the conductor due to processing burrs, metal particle adhesion, or mechanical damage. Under the action of a high-voltage electric field, the electric field at the tip of the protrusion is highly concentrated, causing partial discharge defects to occur in the surrounding insulating medium. Surface discharge defects refer to conductive channels formed on the surface of the insulating material due to moisture, dirt deposition, or creepage carbonization. These channels form along the interface between the insulating medium and the surrounding environment, and under the action of a high-voltage electric field, they cause partial discharge defects along this interface.

[0025] In this embodiment of the invention, combined with Figure 8As shown, an experimental platform for simulating partial discharge faults inside a pole-mounted circuit breaker was constructed. This platform includes a high-voltage AC power supply, a pole-mounted circuit breaker, a radio frequency (RF) sensor, an ultrasonic sensor, and a data acquisition instrument. Specifically: the high-voltage AC power supply is electrically connected to the pole-mounted circuit breaker to provide high-voltage AC power; different types of partial discharge faults are installed within the high-voltage pole of the pole-mounted circuit breaker; the RF sensor is positioned horizontally at a predetermined distance (e.g., 1.5 meters) from the fault point of the pole-mounted circuit breaker, and the ultrasonic sensor is installed on the high-voltage pole side of the pole-mounted circuit breaker; the data acquisition instrument is electrically connected to both the RF and ultrasonic sensors. In practical applications, the high-voltage AC power supply applies operating voltage to the pole-mounted circuit breaker, causing partial discharge within the high-voltage pole, generating RF and ultrasonic signals. The RF and ultrasonic sensors respectively sense these signals, and the data acquisition instrument synchronously acquires and records them to obtain time-domain data of the RF and ultrasonic signals corresponding to different partial discharge fault types. For example, the partial discharge faults include air gap discharge defects, tip discharge defects, and surface discharge defects.

[0026] Combination Figure 1 As shown, this embodiment of the invention provides a method for identifying partial discharge faults in pole-mounted circuit breakers, comprising the following steps: S10, acquire the radio frequency signal and ultrasonic signal collected when partial discharge occurs in the pole-mounted circuit breaker.

[0027] Here, radio frequency (RF) signals refer to the electromagnetic wave signals excited and radiated by the discharge pulse during partial discharge, with frequencies within the radio frequency band. The rapid charge transfer during partial discharge generates steep pulse currents, which radiate electromagnetic energy into the surrounding space, forming RF signals that can be detected by RF sensors.

[0028] Ultrasonic signals refer to the mechanical vibration signals generated by the surrounding medium due to physical effects such as thermal expansion and stress release during partial discharge. Their frequency range is within the ultrasonic band. The energy released instantaneously by partial discharge causes a slight mechanical deformation in the surrounding insulating medium. This deformation propagates outward in the form of ultrasonic waves, forming an ultrasonic signal that can be detected by ultrasonic sensors.

[0029] S20, based on radio frequency signals and ultrasonic signals, constructs a partial discharge phase distribution (PRPD) map.

[0030] S30. Construct a PRPD map feature sample dataset using the image features of the PRPD grayscale images corresponding to the PRPD map.

[0031] In practical applications, after obtaining the PRPD map, the PRPD map is converted into a 64×64 PRPD grayscale image, and the color transition from white to black is represented by 0-255 levels.

[0032] S40. Input the PRPD map feature sample dataset into the pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker.

[0033] The partial discharge fault identification method for pole-mounted circuit breakers provided in this invention simultaneously acquires radio frequency (RF) and ultrasonic signals when partial discharge occurs in the pole-mounted circuit breaker during the signal acquisition stage. Leveraging the inherent advantages of RF signals' high sensitivity to electromagnetic transients during discharge and the immunity of ultrasonic signals to strong electromagnetic interference, the two types of signals achieve complementary synergy at the information level, improving the completeness and anti-interference capability of partial discharge fault information from the source. Simultaneously, in the identification stage, a partial discharge fault identification (PDF) map is constructed based on the RF and ultrasonic signals, and image features of the grayscale image are extracted. This transforms the PDF identification into a mature image pattern recognition problem. Utilizing the deep feature mining capabilities of a pre-trained PDF identification model, the image features contained in the map are automatically extracted and judged, achieving intelligent perception and accurate classification of the partial discharge defect state. Thus, through the organic combination of collaborative perception of RF and ultrasonic signals and map-based intelligent identification, the accuracy of partial discharge fault identification in pole-mounted circuit breakers is improved.

[0034] Optionally, combined Figure 2 As shown, in step S20, a partial discharge phase distribution (PRPD) map is constructed based on radio frequency signals and ultrasonic signals, including: S201, based on the time-domain data of the radio frequency signal and the ultrasonic signal, determine the power frequency phase range and discharge amplitude range of each discharge pulse respectively, and divide the power frequency phase range into multiple first intervals and the discharge amplitude range into multiple second intervals.

[0035] In some possible implementations, the power frequency phase is calculated using the following formula:

[0036] in, For power frequency phase, The operating voltage frequency (e.g., 50Hz). The time when the discharge pulse occurs. This refers to the zero-crossing time of the operating voltage. The range of values ​​is ,For example, .

[0037] S202, respectively count the frequency of each discharge pulse of the radio frequency signal and the ultrasonic signal in different combinations of the first interval and the second interval, and generate radio frequency PRPD map and ultrasonic PRPD map with power frequency phase as the first dimension, discharge amplitude as the second dimension and frequency as the statistical value.

[0038] S203, the radio frequency PRPD map and the ultrasound PRPD map are fused at the map layer to obtain the PRPD map.

[0039] In some possible implementations, the formula for constructing the PRPD map is as follows:

[0040] Where H(m,n) represents the frequency of partial discharge faults occurring within the power frequency phase and discharge amplitude range, and N is the total number of discharge pulses. For power frequency phase, The discharge amplitude, This is the m-th interval of the limiting axis. This is the nth interval of the amplitude axis.

[0041] The following is a specific example illustrating the PRPD map construction process.

[0042] Assuming a tip discharge defect exists within the high-voltage pole of the pole-mounted circuit breaker, time-domain data of radio frequency and ultrasonic signals were simultaneously acquired for 0.04 seconds (two power frequency cycles) using a data acquisition instrument. First, pulse identification was performed on the time-domain waveforms to extract the occurrence time of each discharge pulse. and peak discharge Taking radio frequency signals as an example, a total of 6 discharge pulses were identified, namely pulse 1. Pulse 2 Pulse 3 Pulse 4 Pulse 5 Pulse 6 Calculate the power frequency phase corresponding to each pulse. For example, pulse 1: Pulse 4: .

[0043] The power frequency phase from 0° to 360° is divided into 12 first intervals (each interval is 30°), and the discharge amplitude from 0 to 100mV is divided into 5 second intervals (each interval is 20mV). The combinations of intervals that each pulse falls into are counted to generate a two-dimensional histogram. For example, pulses 1 and 5 both fall into the phase interval 30°-60° and the amplitude interval 20-40mV, so the frequency count for this grid is 2; pulse 2 falls into the phase interval 120°-150° and the amplitude interval 80-100mV, so the count is 1.

[0044] The time-domain data of the ultrasound signal were processed in the same way to generate radio frequency PRPD maps and ultrasound PRPD maps respectively. The frequency matrices of the two maps were then superimposed on corresponding grids or the maximum value was taken to complete the map layer fusion, resulting in the final PRPD map.

[0045] Radio frequency (RF) signals are highly sensitive to electromagnetic transient processes during partial discharge, while ultrasonic signals are unaffected by on-site electromagnetic interference. By constructing RF PRPD maps and ultrasonic PRPD maps separately and then fusing them at the map layer level, the complementary information of the two types of signals in partial discharge perception can be fully integrated. This allows the fused PRPD map to possess both high sensitivity and high anti-interference capability, providing a richer and more comprehensive map foundation for subsequent image feature extraction and fault identification, thereby effectively improving the accuracy and robustness of partial discharge fault identification.

[0046] Optionally, in step 30, the image features of the PRPD grayscale image include: Hu invariant moments used to characterize the morphological features of the PRPD grayscale image; and the mean and standard deviation used to characterize the statistical features of the PRPD grayscale image.

[0047] After obtaining the PRPD grayscale image, feature extraction is performed on the PRPD grayscale image. Among them, Hu invariant moments characterize the morphological features of the PRPD grayscale image, and the mean and standard deviation characterize the statistical features of the PRPD grayscale image.

[0048] In some possible implementations, the Hu invariant moments include seven moment parameters. , respectively represented as:

[0049] In some possible implementations, the mean represents the average level of discharge intensity, and the formula for calculating the mean is as follows:

[0050] In the formula, The mean, The discharge amplitude, The number of data points.

[0051] In some possible implementations, the standard deviation characterizes the dispersion of the discharge phase, and the formula for calculating the standard deviation is as follows:

[0052] In the formula, The standard deviation is denoted as .

[0053] For example, by collecting experimental data on partial discharge faults inside pole-mounted circuit breakers and constructing PRPD grayscale images, the image feature values ​​of radio frequency signals are calculated, as shown in Table 1 below.

[0054] Table 1. Feature values ​​of PRPD images of radio frequency signals

[0055] For example, by collecting experimental data on partial discharge faults inside pole-mounted circuit breakers and constructing PRPD grayscale images, the image feature values ​​of ultrasonic signals were calculated, as shown in Table 2 below.

[0056] Table 2. Feature values ​​of ultrasound signal PRPD image

[0057] The Hu invariant moment, possessing rotation, scaling, and translation invariance, can stably capture the morphological differences of different partial discharge faults in the PRPD spectrum, effectively overcoming the influence of variations in the location and scale of discharge modes in the spectrum on the identification results. The mean and standard deviation, from a statistical perspective, reflect the overall brightness and contrast characteristics of the spectrum, respectively, and can quantitatively characterize the differences in discharge intensity and dispersion of different partial discharge faults. By extracting three types of image features—Hu invariant moments, mean, and standard deviation—from the PRPD grayscale image, these three features collaboratively characterize the information of the PRPD grayscale image from both morphological and statistical dimensions, achieving a multi-level, complementary representation of the partial discharge spectrum. This overcomes the limitation of single features in representing complex partial discharge modes, providing more discriminative input features for the partial discharge fault identification model, thereby effectively improving the accuracy and stability of partial discharge fault identification.

[0058] It is understood that the image features of a PRPD grayscale image may include, individually, Hu invariant moments characterizing the morphological features of the PRPD grayscale image; or, individually, the mean and standard deviation characterizing the statistical features of the PRPD grayscale image; or, simultaneously, Hu invariant moments characterizing the morphological features of the PRPD grayscale image and the mean characterizing the statistical features of the PRPD grayscale image; or, simultaneously, Hu invariant moments characterizing the morphological features of the PRPD grayscale image and the standard deviation characterizing the statistical features of the PRPD grayscale image. This invention does not impose any limitations on these aspects.

[0059] Optionally, combined Figure 3 As shown, in step S40, the PRPD map feature sample dataset is input into the pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker, including: S401, input the PRPD map feature sample dataset into the pre-trained first partial discharge fault identification model, and the first partial discharge fault identification model performs classification and identification based on the image features and outputs preliminary identification results.

[0060] Here, the first partial discharge fault identification model is a classification model combining a one-dimensional convolutional neural network and a support vector machine classifier, namely the 1DCNN-SVM network model. The PRPD map feature sample dataset is input into the 1DCNN-SVM network model, and the preliminary identification results of the partial discharge fault of the pole-mounted circuit breaker are output.

[0061] In some possible implementations, the first partial discharge fault identification model is constructed as follows: A series of sequentially connected convolutional layers (e.g., 2 layers) and pooling layers (e.g., 2 layers) are set; an SVM classifier connected to the output of the pooling layer is set as the classification layer; thus, a 1D CNN-SVM network model is obtained. Wherein: The convolutional layer performs convolution calculations according to the stride to obtain the corresponding feature matrix. The calculation formula is as follows:

[0062] In the formula, This is the j-th output of the l-th convolutional layer. For a one-dimensional convolutional neural network (1DCNN), [function name] is the activation function. For the i-th and j-th outputs of the l-th convolutional layer, This is the bias value.

[0063] For the feature matrix output by the convolutional layer, a pooling layer is used to reduce its dimensionality. The dimensionality reduction formula is as follows:

[0064] In the formula, This is a dimensionality reduction function.

[0065] The SVM classifier, used as the classification layer in 1DCNN, significantly improves the recognition accuracy of 1DCNN networks for small sample data.

[0066] In some practical applications, the 1DCNN network is configured with two convolutional layers, two pooling layers, and one fully connected layer. Features are processed through the pooling and convolutional layers and the fully connected layer, and then an SVM classifier is used to classify partial discharge faults. ReLU is used as the activation function, and the network function is optimized using SGD gradient descent with a learning rate of 0.01. The feature vector sets of radio frequency and ultrasound signals are divided into training, validation, and test sets in a 6:2:2 ratio. The 1DCNN-SVM network model is trained using the training and validation sets, and then the test set is input into the trained 1DCNN-SVM network model.

[0067] For example, combined Figure 6 The training loss function graph of the PRPD atlas features of the ultrasound signal is shown. The training set loss value of the 1DCNN-SVM network model is stable at around 0.85, and the validation set accuracy is stable at around 75%. The recognition results of the 1DCNN-SVM network model are shown in Table 3 below.

[0068] Table 3. Recognition Results of 1DCNN-SVM Network Model

[0069] In some possible implementations, the first partial discharge fault identification model is trained as follows: obtain PRPD spectral feature sample data containing labels of known partial discharge fault types; wherein, the PRPD spectral feature sample dataset includes a training set and a validation set; use the training set to perform supervised training on the initial classification model, and use the validation set to validate the trained initial classification model, thereby obtaining the trained first partial discharge fault identification model.

[0070] S402, the preliminary identification results are input into the second partial discharge fault identification model. The second partial discharge fault identification model performs decision fusion on the preliminary identification results, calculates the basic probability allocation (BPA) value for different partial discharge faults and performs evidence fusion, and outputs the final identification result based on the confidence level after evidence fusion.

[0071] Here, the second partial discharge fault identification model is a KNN-DS multi-information fusion model based on the K-nearest neighbor algorithm and DS evidence theory.

[0072] In some possible implementations, the second partial discharge fault identification model outputs the final identification result as follows: Based on Euclidean distance, it searches for the K nearest neighbor samples of the sample to be identified in the PRPD map feature sample dataset, and calculates the support of the sample to be identified for each partial discharge fault type based on the distribution of the K nearest neighbor samples, and generates a BPA value after normalization; it performs evidence fusion on multiple BPA values ​​based on the Dempster combination rule to obtain a joint BPA value; based on the joint BPA value, it selects the partial discharge fault type with the highest confidence as the final identification result according to the maximum confidence decision criterion.

[0073] For example, the sample to be identified is the first i The formula for calculating the support for partial discharge fault types is as follows:

[0074] In the formula, For the sample to be identified, the first i Support for partial discharge fault types The decay function, The indicator function is used. The weighted similarity sum of the K nearest neighbor samples of the sample to be identified for each type of partial discharge fault is calculated to obtain the similarity for the Kth sample. i Support for various failure modes.

[0075] For example, the sample to be identified is assigned to the first i The formula for calculating the BPA value of partial discharge faults is as follows:

[0076] In the formula, Assign the sample to be identified to the first iBPA value for partial discharge faults For the sample to be identified, the first i Support for partial discharge fault types Let Q represent the support of the sample to be identified for the q-th type of partial discharge fault, where Q is the total number of partial discharge fault types. This is a non-specific allocation term. Add it to the denominator. This significantly reduced the BPA values ​​for each fault type when the total support was small (i.e., when the evidence was insufficient), thereby allocating a portion of the probability to "uncertainty".

[0077] In practical applications, multiple BPA values ​​include a first BPA value generated based on radio frequency signal characteristics and a second BPA value generated based on ultrasonic signal characteristics. Evidence fusion is performed on these multiple BPA values ​​based on Dempster's combination rule to obtain a joint BPA value, including: calculating a conflict coefficient K based on the first and second BPA values; where the conflict coefficient K characterizes the degree of conflict between the first and second BPA values ​​regarding their support for each partial discharge fault type; when the conflict coefficient K is less than 1, an orthogonal sum operation is performed on the first and second BPA values ​​based on Dempster's combination rule, and the joint BPA value is obtained after normalization.

[0078] For example, the formula for calculating the combined BPA value is as follows:

[0079] In the formula, For the combined BPA value, For the first i The class information for the BPA value of the j-th class target. For the first i The BPA value represents the uncertainty of the class information, where K is the conflict coefficient. This applies when multiple BPA values ​​include a first BPA value generated based on radio frequency signal characteristics and a second BPA value generated based on ultrasound signal characteristics. The first BPA value, This is the second BPA value.

[0080] For example, the formula for calculating the conflict coefficient K is as follows:

[0081] The following is a specific example illustrating the two-level recognition process in step S40.

[0082] Suppose that the sample X to be identified has a true fault type of tip discharge, but due to the ambiguity of the signal characteristics, it is located in the boundary zone between tip discharge and surface discharge. The PRPD map feature vector of sample X (7 Hu invariant moments + mean + standard deviation, a total of 9 features) was obtained, and a labeled PRPD map feature sample dataset has been constructed as a reference library.

[0083] The feature vector of sample X is input into the pre-trained first partial discharge fault identification model (1DCNN-SVM). After the 1DCNN extracts deep features, the SVM classifier outputs the score (i.e., accuracy) of sample X belonging to each type of fault: tip discharge fault, score of 0.52; surface discharge fault, score of 0.43; air gap discharge fault, score of 0.05.

[0084] The preliminary identification results (including accuracy scores for each partial discharge fault type) were input into the second partial discharge fault identification model (KNN-DS). In the PRPD map feature sample dataset, the Euclidean distance between sample X and all known samples was calculated, and the five nearest neighbors were identified: Neighbor 1 (point discharge fault; Euclidean distance 0.3; distance weight 3.333); Neighbor 2 (surface discharge fault; Euclidean distance 0.4; distance weight 2.500); Neighbor 3 (point discharge fault; Euclidean distance 0.7; distance weight 1.429); Neighbor 4 (surface discharge fault; Euclidean distance 0.8; distance weight 1.250); Neighbor 5 (surface discharge fault; Euclidean distance 1.2; distance weight 0.833).

[0085] The accuracy of each partial discharge fault type in the preliminary identification results is used as a category confidence factor to dynamically adjust the voting weight of each neighbor. The adjustment formula is: Adjusted weight = Distance weight × SVM score of the corresponding fault type of the neighbor. Therefore, the adjusted weights are: Neighbor 1 (point discharge fault; distance weight 3.333 × 0.52 = 1.733); Neighbor 2 (surface discharge fault; distance weight 2.500 × 0.43 = 1.075); Neighbor 3 (point discharge fault; distance weight 1.429 × 0.52 = 0.743); Neighbor 4 (surface discharge fault; distance weight 1.250 × 0.43 = 0.538); Neighbor 5 (surface discharge fault; distance weight 0.833 × 0.43 = 0.358). After incorporating the preliminary identification results, the distinction between point discharge and surface discharge increases from almost no difference (1.04 times) to 1.26 times, effectively widening the classification boundary.

[0086] Dynamically adjusted support: Tip discharge = 1.733 + 0.743 = 2.476; Surface discharge = 1.075 + 0.538 + 0.358 = 1.971. After normalization, the first BPA value based on radio frequency signal characteristics is generated: m1 (tip discharge fault) = 0.48; m1 (surface discharge fault) = 0.38; m1 (uncertain) = 0.14. Here, the uncertain BPA value of 0.14 is dynamically allocated according to the category support ratio. Since the tip / surface ratio is still not particularly large, a certain degree of uncertainty is retained. Assuming that the PRPD spectral characteristics of the ultrasound signal are clearer, the second BPA value based on ultrasound signal characteristics is generated through the same process: m2 (tip discharge fault) = 0.80; m2 (surface discharge fault) = 0.10; m2 (uncertain) = 0.10.

[0087] Based on the first and second BPA values, the conflict coefficient K is calculated as follows: K = 0.48 × 0.10 + 0.38 × 0.80 = 0.048 + 0.304 = 0.352. After performing orthogonal summation and normalization (denominator = 1 - K = 0.648), the joint BPA value is obtained.

[0088] According to the maximum confidence decision criterion, the tip discharge fault was selected as the final identification result with a confidence level of 0.84.

[0089] For example, the preliminary identification results are as follows: A 1. A 2 and A 3. The final recognition results are shown in Table 4 below. A 1 indicates a tip discharge fault. A 2 indicates a surface discharge fault. A 3 indicates an air gap discharge fault. q It indicates uncertainty.

[0090] Table 4. KNN-DS Multi-Information Fusion Recognition Results

[0091] As shown in Table 4, in the examples of this invention, different partial discharge faults were effectively diagnosed and distinguished. The uncertainty of the final output result after KNN-DS multi-information fusion was significantly reduced, and the BPA values ​​of the uncertainties for tip discharge faults, surface discharge faults, and air gap discharge faults were all less than 0.05. Meanwhile, the maximum BPA values ​​corresponding to the three types of discharge faults were 0.8120, 0.7355, and 0.7857, respectively, with an overall accuracy of 95.5%, demonstrating that the proposed method has good engineering application value in the field of partial discharge fault diagnosis for pole-mounted circuit breakers.

[0092] By constructing a two-level cascaded architecture of a first partial discharge (PD) fault identification model and a second PD fault identification model, a progressive identification process from feature-level classification to decision-level fusion is achieved. The first PD fault identification model performs preliminary classification based on image features from PRPD grayscale images, quickly providing a preliminary judgment of the fault type. Building upon this, the second PD fault identification model uses a KNN-DS multi-information fusion mechanism to fuse the preliminary identification results with multi-source evidence such as radio frequency and ultrasonic data. It quantifies the confidence level of each fault type using Basic Probability Assignment (BPA) values ​​and outputs the final identification result after comprehensive evaluation through evidence fusion. In this way, the two-level identification architecture organically combines the feature extraction capabilities of deep learning models with the uncertainty handling capabilities of DS evidence theory, effectively overcoming the limitations of single models that are prone to misjudgment in ambiguous samples and strong interference environments, and significantly improving the accuracy and reliability of pole-mounted circuit breaker PD fault identification.

[0093] In some embodiments, combined with Figure 4 As shown, the method for identifying partial discharge faults in pole-mounted circuit breakers includes the following steps: S50, obtain the training dataset of the pre-trained partial discharge fault identification model, and use the training dataset to train the partial discharge fault identification model.

[0094] S10, acquire the radio frequency signal and ultrasonic signal collected when partial discharge occurs in the pole-mounted circuit breaker.

[0095] S20, based on radio frequency signals and ultrasonic signals, constructs a partial discharge phase distribution (PRPD) map.

[0096] S30. Construct a PRPD map feature sample dataset using the image features of the PRPD grayscale images corresponding to the PRPD map.

[0097] S40. Input the PRPD map feature sample dataset into the pre-trained partial discharge fault identification model to obtain the partial discharge fault identification result of the pole-mounted circuit breaker.

[0098] In some possible implementations, combining Figure 5 As shown, in step S50, the training dataset of the pre-trained partial discharge fault identification model is obtained as follows: S501 is an experimental platform for simulating partial discharge faults inside pole-mounted circuit breakers. Different types of partial discharge faults are set in the high-voltage pole of the pole-mounted circuit breaker, and partial discharge is generated after the operating voltage is applied.

[0099] In practical applications, the experimental platform is supplied with AC power through a high-voltage AC power supply. Each experiment only applies pressure to the high-voltage pole with faults, with each pressure application having an amplitude of 1kV. The pressure application is stopped when a partial discharge radio frequency signal is detected.

[0100] S502 acquires corresponding radio frequency signals and ultrasonic signals through radio frequency sensors and ultrasonic sensors.

[0101] S503 constructs a PRPD map based on the acquired radio frequency and ultrasonic signals, extracts the image features of the PRPD grayscale image corresponding to the PRPD map, and uses the image features to construct a PRPD map feature sample dataset with partial discharge fault type labels as a training dataset.

[0102] In practical applications, a PRPD (Partial Discharge Damage) time-domain signal is acquired for each experiment to construct a PRPD map. The constructed PRPD map is then converted to grayscale, becoming a 64×64 PRPD grayscale image. Color transitions from white to black are represented by levels 0-255. 300 PRPD grayscale images can be constructed for each type of partial discharge fault. Image features are extracted from the PRPD grayscale images, and these features are used to construct a PRPD map feature sample dataset labeled with the partial discharge fault type.

[0103] For the first partial discharge fault identification model, the training dataset serves as the training samples for supervised training, used to learn and optimize the parameters of the classification model combining a one-dimensional convolutional neural network and a support vector machine classifier. By iteratively training using image features in the training set labeled with known partial discharge fault types, the first partial discharge fault identification model establishes a mapping relationship from image features to fault types, enabling it to perform preliminary classification and identification of unknown samples.

[0104] For the second partial discharge (PD) fault identification model, the training dataset serves as the evidence source for the KNN-DS multi-information fusion model. Based on the K-nearest neighbor algorithm, the second PD fault identification model searches for the nearest neighbor samples of the sample to be identified in the training dataset and calculates the Basic Probability Assignment (BPA) value based on the fault type distribution of the nearest neighbor samples. Then, it utilizes DS evidence theory for multi-source evidence fusion. In this process, the training dataset does not participate in model parameter training but serves as a reference knowledge base during the inference stage, providing evidentiary support for decision-level fusion.

[0105] By constructing a partial discharge (PD) fault simulation experimental platform, different types of PD fault defects were set within the high-voltage pole of a pole-mounted circuit breaker, and radio frequency (RF) and ultrasonic signals were simultaneously acquired. This allowed for the creation of a PRPD spectral feature sample dataset labeled with PD fault types, which served as the training dataset. The acquisition method of this training dataset covers various typical PD fault types, including air gap discharge, tip discharge, and surface discharge, ensuring the diversity and completeness of the training samples. Furthermore, the PRPD spectral feature sample dataset, constructed based on measured signals, accurately reflects the signal characteristics of PD faults under field operating conditions, providing a reliable data foundation for the training and inference of PD fault identification models.

[0106] In this embodiment of the invention, a training dataset labeled with fault types is obtained by constructing a simulation experimental platform. This training dataset is then used to pre-train a partial discharge (PD) fault identification model, providing a reliable training foundation for the model to identify multiple types of PD faults. This allows the PD fault identification model to fully learn the differentiated feature patterns of different fault types on the PRPD spectrum. Based on this, radio frequency (RF) and ultrasonic signals are simultaneously acquired, utilizing the physical complementarity of these two types of signals to improve the completeness and anti-interference capability of fault information from the source. By constructing a PRPD spectrum, the one-dimensional time-domain signal is transformed into a two-dimensional image, and multi-level image features such as Hu invariant moments, mean, and standard deviation are extracted, allowing for a full representation of the differences between different PD faults. Finally, the pre-trained PD fault identification model automatically discriminates the spectrum features, transforming PD identification into a mature image pattern recognition problem. The entire scheme forms a closed loop from training data construction, signal fusion, feature extraction to intelligent recognition, effectively overcoming the limitations of single-sensor detection in balancing sensitivity and reliability, and significantly improving the accuracy and robustness of pole-mounted circuit breaker (PLC) PD fault identification.

[0107] Combination Figure 7 As shown, this embodiment of the disclosure provides a partial discharge fault identification device 70 for pole-mounted circuit breakers, including a processor 700 and a memory 701, and may also include a communication interface 702 and a bus 703. The processor 700, communication interface 702, and memory 701 can communicate with each other via the bus 703. The communication interface 702 can be used for information transmission. The processor 700 can call logical instructions in the memory 701 to execute the partial discharge fault identification method for pole-mounted circuit breakers described in the above embodiment.

[0108] Furthermore, the logic instructions in the aforementioned memory 701 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0109] The memory 701, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 700 executes functional applications and data processing by running the program instructions / modules stored in the memory 701, thereby implementing the partial discharge fault identification method for pole-mounted circuit breakers in the above method embodiments.

[0110] The memory 701 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 701 may include high-speed random access memory and may also include non-volatile memory.

[0111] Processor 700 is a circuit with signal processing capabilities. In one implementation, processor 700 can be a circuit with instruction read and execute capabilities, such as a central processing unit (CPU), microprocessor, graphics processing unit (GPU) (which can be understood as a type of microprocessor), or digital signal processor (DSP). In another implementation, processor 700 can implement certain functions through the logical relationships of hardware circuits. These logical relationships of hardware circuits are fixed or reconfigurable. For example, processor 700 can be a hardware circuit implemented as an ASIC or a programmable logic device (PLD), such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the processor loading instructions to implement the functions of some or all of the above modules. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as a type of ASIC, such as a neural network processing unit (NPU), tensor processing unit (TPU), or deep learning processing unit (DPU). The processor 700 is used to execute related programs to implement the functions required by the units in the partial discharge fault identification device for pole-mounted circuit breakers in the embodiments of this application, or to execute the partial discharge fault identification method for pole-mounted circuit breakers in the method embodiments of this application.

[0112] As can be seen, each module in the above device can be one or more processors (or processing circuits) configured to implement the above methods, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor types.

[0113] Furthermore, the modules in the above devices can be integrated in whole or in part, or they can be implemented independently. In one implementation, these modules are integrated together as a system-on-a-chip (SOC). The SOC may include at least one processor for implementing any of the above methods or for implementing the functions of the modules of the device. The at least one processor may be of different types, such as CPU and FPGA, CPU and artificial intelligence processor, CPU and GPU, etc.

[0114] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for identifying partial discharge faults in pole-mounted circuit breakers, characterized in that, include: Acquire radio frequency and ultrasonic signals when partial discharge occurs in a pole-mounted circuit breaker; A partial discharge phase distribution (PRPD) map was constructed based on radio frequency and ultrasonic signals. We construct a PRPD map feature sample dataset using the image features of the PRPD grayscale images corresponding to the PRPD map. Input the PRPD map feature sample dataset into the pre-trained partial discharge fault identification model to obtain the partial discharge fault identification results of the pole-mounted circuit breaker.

2. The partial discharge fault identification method according to claim 1, characterized in that, Based on radio frequency and ultrasonic signals, a partial discharge phase distribution (PRPD) map was constructed, including: Based on the time-domain data of radio frequency signals and ultrasonic signals, the power frequency phase range and discharge amplitude range of each discharge pulse are determined respectively, and the power frequency phase range is divided into multiple first intervals and the discharge amplitude range is divided into multiple second intervals. The frequency of each discharge pulse of the radio frequency signal and the ultrasonic signal in different combinations of the first interval and the second interval is counted respectively, and radio frequency PRPD map and ultrasonic PRPD map are generated with power frequency phase as the first dimension, discharge amplitude as the second dimension and frequency as the statistical value. The radiofrequency PRPD map and the ultrasound PRPD map are fused at the map layer to obtain the PRPD map.

3. The partial discharge fault identification method according to claim 1, characterized in that, The image features of PRPD grayscale images include: Hu invariant moments used to characterize the morphological features of PRPD grayscale images; and The mean and standard deviation are used to characterize the statistical features of PRPD grayscale images.

4. The partial discharge fault identification method according to claim 1, characterized in that, Inputting the PRPD map feature sample dataset into a pre-trained partial discharge fault identification model yields the partial discharge fault identification results for pole-mounted circuit breakers, including: The PRPD map feature sample dataset is input into the pre-trained first partial discharge fault identification model, which performs classification and identification based on the image features and outputs preliminary identification results. The preliminary identification results are input into the second partial discharge fault identification model. The second partial discharge fault identification model performs decision fusion on the preliminary identification results, calculates the basic probability allocation (BPA) value for different partial discharge faults, performs evidence fusion, and outputs the final identification result based on the confidence level after evidence fusion.

5. The partial discharge fault identification method according to claim 4, characterized in that, The first partial discharge fault identification model is a classification model that combines a one-dimensional convolutional neural network with a support vector machine classifier. The second partial discharge fault identification model is a KNN-DS multi-information fusion model based on the K-nearest neighbor algorithm and DS evidence theory.

6. The partial discharge fault identification method according to claim 5, characterized in that, The first partial discharge fault identification model was trained as follows: Obtain PRPD spectral feature sample data containing labels of known partial discharge fault types; wherein, the PRPD spectral feature sample dataset includes a training set and a validation set; The initial classification model is trained in a supervised manner using the training set, and the trained initial classification model is validated using the validation set to obtain the first trained partial discharge fault identification model.

7. The partial discharge fault identification method according to claim 5, characterized in that, The second partial discharge fault identification model outputs the final identification result in the following manner: Based on Euclidean distance, the K nearest neighbor samples of the sample to be identified are searched in the PRPD map feature sample dataset. Based on the distribution of the K nearest neighbor samples, the support of the sample to be identified for each partial discharge fault type is calculated and the BPA value is generated after normalization. Evidence fusion of multiple BPA values ​​is performed based on Dempster's combination rule to obtain a joint BPA value; Based on the joint BPA value, and according to the maximum trust decision criterion, the partial discharge fault type with the highest trust level is selected as the final identification result.

8. The partial discharge fault identification method according to claim 7, characterized in that, The multiple BPA values ​​include a first BPA value generated based on radio frequency signal characteristics and a second BPA value generated based on ultrasound signal characteristics; Evidence fusion of multiple BPA values ​​based on Dempster's combination rule yields a joint BPA value, including: Based on the first BPA value and the second BPA value, the conflict coefficient K is calculated; whereby the conflict coefficient K characterizes the degree of conflict between the first BPA value and the second BPA value for the support of each partial discharge fault type. When the conflict coefficient K is less than 1, based on the Dempster combination rule, the first BPA value and the second BPA value are orthogonally summed and then normalized to obtain the joint BPA value.

9. The partial discharge fault identification method according to any one of claims 1 to 8, characterized in that, The training dataset for the pre-trained partial discharge fault identification model is obtained as follows: Based on the experimental platform for simulating partial discharge faults inside pole-mounted circuit breakers, different types of partial discharge fault defects are set in the high-voltage pole of the pole-mounted circuit breaker, and partial discharge is generated after the operating voltage is applied. The corresponding radio frequency signals and ultrasonic signals are acquired using radio frequency sensors and ultrasonic sensors; PRPD maps are constructed based on the acquired radio frequency and ultrasonic signals. Image features of the corresponding PRPD grayscale images are extracted, and the image features are used to construct a PRPD map feature sample dataset with partial discharge fault type labels as a training dataset.

10. A partial discharge fault identification device for a pole-mounted circuit breaker, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to perform, when executing the program instructions, the partial discharge fault identification method for pole-mounted circuit breakers as described in any one of claims 1 to 9.

Images