Partial discharge detection method and device and electronic equipment

Abstract

The invention discloses a partial discharge detection method and device and electronic equipment, and relates to the field of power failure detection. The method comprises the following steps: acquiring M groups of partial discharge data of power equipment through M acquisition strategies; performing screening and dimension reduction on the features of each group of partial discharge data through a target verification strategy to obtain M groups of feature sets; determining a partial discharge detection result according to each group of feature sets to obtain M partial discharge detection results; converting each partial discharge detection result into a piece of posterior probability distribution data, obtaining M pieces of posterior probability distribution data in one-to-one correspondence with the M partial discharge detection results, and fusing the M pieces of posterior probability distribution data to obtain target probability distribution data; and determining a target partial discharge type and target discharge intensity of the power equipment according to the target probability distribution data. According to the invention, the problem of insensitive detection or wrong diagnosis of various defects in partial discharge detection by using a single detection means in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of power fault detection, and in particular to a partial discharge detection method, device and electronic equipment. Background Art

[0002] With the continued development of the power industry and the accelerated construction of smart grids, partial discharge detection technology is becoming increasingly prominent as a key tool for ensuring the safe operation of power equipment. Currently, partial discharge detection technology relies primarily on a series of mature yet distinctive detection methods, including but not limited to pulse current, optical measurement, ultra-high frequency, and ultrasonic methods.

[0003] Although these detection technologies have achieved remarkable results in partial discharge monitoring, their limitations cannot be ignored. The existing technology has the problem that using a single detection method in partial discharge detection is insensitive to various defects or may cause incorrect diagnosis.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present application provide a partial discharge detection method, apparatus, and electronic device to at least solve the technical problem in the prior art of using a single detection means to insensitively detect or misdiagnose various defects in partial discharge detection.

[0006] According to one aspect of an embodiment of the present application, a partial discharge detection method is provided, comprising: collecting M groups of partial discharge data of an electric power device using M collection strategies, where M is an integer greater than 1, and each collection strategy corresponds to a group of partial discharge data; screening and reducing the features of each group of partial discharge data using a target verification strategy to obtain M groups of feature sets, where the M groups of feature sets correspond one-to-one to the M groups of partial discharge data, and the target verification strategy is used to detect the correlation between the features and various partial discharge types, and select the feature set with the minimum mutual correlation and the maximum correlation with a preset target; determining a partial discharge detection result based on each group of feature sets to obtain M partial discharge detection results; converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding one-to-one to the M partial discharge detection results, and then fusing the M posterior probability distribution data to obtain target probability distribution data; and determining a target partial discharge type and a target discharge intensity of the electric power device based on the target probability distribution data.

[0007] Optionally, in the process of screening and reducing the dimension of the features of each group of partial discharge data through the target verification strategy to obtain M groups of feature sets, the process includes: calculating the correlation value between each feature of the i-th group of partial discharge data and various partial discharge types according to the target verification strategy, where i is a positive integer less than or equal to M, and the correlation value is used to quantitatively represent the correlation strength between each feature and various partial discharge types; for the i-th group of partial discharge data, screening multiple features with correlation values ​​greater than a preset threshold as candidate features; calculating the mutual information between the multiple candidate features; and selecting at least one target feature from the multiple candidate features according to the mutual information to form a set of features corresponding to the i-th group of partial discharge data.

[0008] Optionally, a partial discharge detection result is determined based on each group of feature sets to obtain M partial discharge detection results, including: inputting the i-th group of feature sets into a neural network model corresponding to the i-th group of feature sets, and determining a partial discharge detection result based on feature data in the i-th group of feature sets based on prior knowledge learned by the neural network model in a training phase, wherein the training data of the neural network model corresponding to the i-th group of feature sets includes: historical partial discharge data collected using the i-th collection strategy and actual partial discharge types and actual discharge intensities corresponding to the historical partial discharge data.

[0009] Optionally, M posterior probability distribution data are fused to obtain target probability distribution data, including: setting an identification framework, wherein the identification framework includes at least a set of local discharge types and a random element; determining a basic probability assignment corresponding to each posterior probability distribution data and the identification framework, wherein the basic probability assignment corresponding to each posterior probability distribution data is used to characterize, for each element in the identification framework, the probability that the posterior probability distribution data supports that the element is a true situation; and fusing the target probability distribution data based on the M basic probability assignments corresponding to the M posterior probability distribution data.

[0010] Optionally, M posterior probability distribution data are fused to obtain target probability distribution data, including: determining difference information between the M posterior probability distribution data for each element in the recognition frame; determining a target probability assignment for each element in the recognition frame based on the difference information; and determining the target probability distribution data based on the target probability assignment for each element in the recognition frame.

[0011] Optionally, the M collection strategies include at least two of the following collection strategies:

[0012] The first acquisition strategy is used to collect charge amount change data caused by partial discharge to capture the intensity and frequency of discharge events; and determine the partial discharge data of the power equipment based on the intensity and frequency of the discharge events.

[0013] The second acquisition strategy is used to acquire light radiation intensity variation data during the partial discharge process; and determine partial discharge data of the power equipment based on the light radiation intensity variation data.

[0014] The third acquisition strategy is used to acquire electromagnetic wave signals generated by partial discharge and determine partial discharge data of the power equipment based on the electromagnetic wave signals.

[0015] The fourth acquisition strategy is used to acquire ultrasonic signals generated during partial discharge, and determine partial discharge data of the power equipment based on the ultrasonic signals.

[0016] Optionally, the feature set includes at least one of the following features:

[0017] The average discharge amplitude is used to characterize the average value of the charge or voltage amplitude generated by the discharge event during the partial discharge detection cycle.

[0018] Discharge frequency is used to characterize the number of partial discharge events that occur per unit time.

[0019] Skewness is used to characterize the asymmetry of the probability density function distribution of the partial discharge signal.

[0020] Steepness is used to characterize the speed of the rising edge of the partial discharge pulse.

[0021] Phase asymmetry is used to characterize the difference between the discharge characteristics of the positive half-cycle and the negative half-cycle in the partial discharge signal.

[0022] Discharge asymmetry is used to characterize the asymmetry of the signal strength of partial discharge throughout the power cycle.

[0023] According to another aspect of an embodiment of the present application, a partial discharge detection device is provided, comprising: an acquisition unit for acquiring M groups of partial discharge data of an electric power device using M acquisition strategies, wherein M is an integer greater than 1, and each acquisition strategy corresponds to a group of partial discharge data; a first processing unit for screening and reducing the dimension of features of each group of partial discharge data using a target verification strategy to obtain M groups of feature sets, wherein the M groups of feature sets correspond one-to-one to the M groups of partial discharge data, and the target verification strategy is used to detect the correlation between features and various partial discharge types, and select the feature set with the minimum mutual correlation and the maximum correlation with a preset target; a first determination unit for determining a partial discharge detection result based on each group of feature sets to obtain M partial discharge detection results; a second processing unit for converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding one-to-one to the M partial discharge detection results, and then fusing the M posterior probability distribution data to obtain target probability distribution data; and a second determination unit for determining a target partial discharge type and a target discharge intensity of the electric power device based on the target probability distribution data.

[0024] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed, the device where the computer-readable storage medium is located executes the above-mentioned partial discharge detection method.

[0025] According to another aspect of the present application, an electronic device is also provided, wherein the electronic device includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned partial discharge detection method.

[0026] As can be seen from the above content, the present application collects M groups of partial discharge data of power equipment through M collection strategies, where M is an integer greater than 1, and each collection strategy corresponds to a group of partial discharge data; the features of each group of partial discharge data are screened and reduced in dimension through a target verification strategy to obtain M groups of feature sets, where the M groups of feature sets correspond one-to-one to the M groups of partial discharge data, and the target verification strategy is used to detect the correlation between the features and various partial discharge types, and select the feature set with the smallest mutual correlation and the largest correlation with the preset target; a partial discharge detection result is determined based on each group of feature sets to obtain M partial discharge detection results; after converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding one-to-one to the M partial discharge detection results, the M posterior probability distribution data are fused to obtain target probability distribution data; and the target partial discharge type and target discharge intensity of the power equipment are determined based on the target probability distribution data.

[0027] In an embodiment of the present application, M types of collection strategies are adopted to collect M groups of partial discharge data of power equipment, and the features of each group of partial discharge data are screened and reduced in dimension through a target verification strategy, thereby achieving the purpose of constructing an optimized set of M groups of features with the greatest correlation with the preset target. This ensures that the selected feature set can fully reflect the characteristics of partial discharge without increasing the model complexity or decreasing the diagnostic accuracy due to the high redundancy between the features, thereby solving the technical problem in the prior art of using a single detection means to detect various defects insensitively or erroneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0029] Figure 1 is a flow chart of an optional partial discharge detection method according to an embodiment of the present application;

[0030] Figure 2 is a schematic diagram of an optional construction of a corresponding neural network model according to an embodiment of the present application;

[0031] Figure 3 This is a schematic diagram of an optional partial discharge detection device based on multi-physics information fusion according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] It should also be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) collected by this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0035] According to an embodiment of the present application, an embodiment of a partial discharge detection method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] Optionally, according to an embodiment of the present application, a system is provided as the execution subject of the partial discharge detection method based on multi-physical information fusion in the embodiment of the present application, wherein the system can be a software system or an embedded system combining software and hardware. Of course, the method execution subject in the embodiment of the present application can also be other forms of execution subjects, such as devices, equipment, etc. Those skilled in the art should know that this application does not specifically limit the specific form of expression of the method execution subject.

[0037] Figure 1 is a flow chart of an optional partial discharge detection method according to an embodiment of the present application, such as Figure 1 As shown, the method includes the following steps:

[0038] Step S101 : collecting M groups of partial discharge data of power equipment using M collection strategies.

[0039] In step S101 , M is an integer greater than 1, and each acquisition strategy corresponds to a set of partial discharge data.

[0040] Optionally, the M acquisition strategies may refer to a variety of detection methods such as a pulse current method, a light measurement method, a UHF method, and an ultrasonic method.

[0041] Optionally, the system can collect partial discharge data from target power equipment using a variety of detection methods, including pulse current, optical measurement, ultra-high frequency, and ultrasonic methods. The collected raw data undergoes preprocessing, including filtering, standardization, and outlier removal, to ensure the accuracy and effectiveness of subsequent analysis.

[0042] Optionally, the M groups of data collected through M strategies can provide more comprehensive and detailed input information for subsequent feature analysis and diagnostic model construction.

[0043] Step S102 : screening and reducing the dimension of each set of partial discharge data using a target verification strategy to obtain M sets of feature sets.

[0044] In step S102, M groups of feature sets correspond one-to-one to M groups of partial discharge data. The target verification strategy is used to detect the correlation between features and various partial discharge types, and select the feature set with the minimum mutual correlation and the maximum correlation with the preset target.

[0045] Optionally, based on the preprocessed data, the system can extract a series of key discharge characteristic parameters, such as average discharge amplitude, discharge frequency, skewness, steepness, phase asymmetry and discharge asymmetry, according to the characteristics of different target verification strategies.

[0046] Optionally, in order to remove redundant information and improve the efficiency and accuracy of the diagnostic model, the system can use two feature screening methods, the chi-square test and the maximum relevance minimum redundancy method (mRMR), to quantitatively evaluate the correlation between each feature and the partial discharge type, and accordingly eliminate redundant features, optimize and simplify the extracted feature parameters, and finally form two optimized feature sets.

[0047] Optionally, during the feature screening phase, the system calculates the chi-squared test value for all candidate features against the discharge type, then sets a threshold as the screening criterion. Through this screening process, all features with chi-squared values ​​exceeding the threshold are retained in the optimized feature set. These features are considered to be statistically significantly correlated with the partial discharge type and can provide valuable information to the model without adding additional redundancy. The resulting chi-squared test optimized set, the first optimized feature set, ensures that only features that are closely related to the discharge type and information-independent are retained in the model training set, thereby improving the model's learning efficiency, reducing the risk of overfitting, and enhancing the model's generalization ability.

[0048] Optionally, through a chi-square test, the system can effectively identify and retain features that are highly relevant to the prediction target and have low redundancy in PD diagnosis. This provides a refined and powerful feature set for subsequent model building and training, improving the accuracy and efficiency of PD type identification. This process is crucial for building a high-precision, low-redundancy PD diagnosis model and is a core component of the design of an intelligent PD diagnosis system.

[0049] For example, using a chi-square value of 0.6 as the scoring standard, we can obtain the chi-square test optimization set after feature optimization. The data is shown in Table 1:

[0050] Table 1

[0051]

[0052]

[0053] Alternatively, the maximum relevance and minimum redundancy (mRMR) algorithm is a common method for optimizing feature sets. In the mRMR algorithm, mutual information is used as a quantitative metric to analyze the correlation between features and the target variable, as well as the degree of redundancy between features. The goal of this algorithm is to select an optimal set of features that are not only highly correlated with the target variable but also minimize redundancy between them, thereby maximizing the diversity and information content of the feature set.

[0054] Alternatively, in data collected using pulsed current, optical, ultra-high frequency, and ultrasonic methods, the maximum relevance and minimum redundancy (mRMR) algorithm can quantify the score of each feature by calculating mutual information. The score is then sorted from high to low to determine the importance of each feature in predicting the type of partial discharge. Features with high scores contribute significantly to the prediction of the target variable and have low redundancy with other features.

[0055] Optionally, based on the sorting, the maximum relevance minimum redundancy (mRMR) algorithm can determine the optimal size of the feature set by comparing the score differences of adjacent features, that is, retaining those features with higher scores and lower redundancy with each other, and eliminating those with repeated information content, thereby constructing an optimized feature set that contains key prediction information and avoids feature redundancy.

[0056] Optionally, the system applies a maximum relevance minimum redundancy (mRMR) algorithm to feature selection on partial discharge data. This algorithm can generate a set of features that effectively distinguishes different types of discharges while avoiding overfitting, thereby improving the performance of the diagnostic model, including classification accuracy and generalization capabilities. In practical applications, this directly translates into a more accurate assessment of the health status of power equipment, reducing the risk of misdiagnosis and missed diagnosis, and improving the safety and reliability of power grid operations.

[0057] For example, the condition with the largest difference between adjacent scores is selected as the feature screening boundary (except for the cases of infinite scores and finite scores), and thus the maximum correlation and minimum redundancy optimization set after feature selection can be obtained, as shown in Table 2:

[0058] Table 2

[0059]

[0060]

[0061] Step S103 : determining a partial discharge detection result according to each set of feature sets, and obtaining M partial discharge detection results.

[0062] Optionally, the system can use support vector machines (SVMs) as the core algorithm for the diagnostic model. Partial discharge data often exhibits complex nonlinear patterns, and SVMs are highly efficient at handling nonlinear problems. By incorporating kernel techniques, SVMs can find optimal classification boundaries in high-dimensional space. Even if the original data is inseparable in low-dimensional space, they can still find a suitable classification hyperplane in the transformed high-dimensional space, thereby improving classification accuracy.

[0063] Optionally, during the feature selection phase, the most valuable features for diagnosis are selected from the raw partial discharge data using the chi-square test and the maximum relevance minimum redundancy (mRMR) method, forming an optimized multi-physics feature set. The system then inputs this optimized multi-physics feature set into the SVM model.

[0064] Optionally, the system can map the optimized feature set input to the SVM model to a higher-dimensional space using a kernel function. The kernel function transforms the nonlinear, inseparable problem in the low-dimensional space into a linearly separable problem in the high-dimensional space, thereby simplifying the classification difficulty.

[0065] Optionally, the system can train the optimized multi-physics feature set by adjusting the parameters of the SVM model to find an optimal classification hyperplane that maximizes the ability to distinguish different types of partial discharges. During training, the model learns how to locate this hyperplane in high-dimensional space to accurately classify new, unknown data. Ultimately, using the SVM model, the system can accurately determine the type of partial discharge (such as corona discharge, creeping discharge, and suspended discharge) and its severity, which is crucial for preventing power equipment failures and maintaining power grid security.

[0066] In summary, the SVM algorithm can be used in partial discharge diagnosis, combined with the optimized multi-physical information feature set, and through the kernel function technology, it can greatly improve the model's ability to process complex nonlinear data, ensure the system's accurate classification of partial discharge phenomena, and provide strong technical support for the health management and status detection of power equipment.

[0067] Step S104 , after converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding to the M partial discharge detection results, the M posterior probability distribution data are fused to obtain target probability distribution data.

[0068] Optionally, for the trained SVM classification models, the system performs soft output conversion on the classification output of each model (i.e., converting the hard classification results into a posterior probability distribution). The posterior probability distribution data generated by each model for its optimized feature set reflects different confidence levels in the identification of the partial discharge type.

[0069] Optionally, the system can fuse the M posterior probability distributions generated by M models using the Dempster-Shafer evidence theory. This fusion method considers the diagnostic results of each model, as well as the correlations and conflicts between them, and combines the outputs of all models to produce a more comprehensive and reliable target probability distribution for the final comprehensive judgment of partial discharge type and severity.

[0070] In summary, the converted M posterior probability distributions carry information about the uncertainty of each model's classification of different discharge types. Compared to hard classification results, these probability distributions provide richer and more detailed information, reflecting the model's confidence level in the classification. The system then merges these M posterior probability distributions into a single target probability distribution using the Dempster-Shafer evidence theory. This fused target probability distribution achieves higher classification accuracy than the results of a single model and more accurately assesses the type and severity of partial discharges, providing a more robust scientific basis for power system maintenance and decision-making.

[0071] Step S105 : determining a target partial discharge type and a target discharge intensity of the power equipment according to the target probability distribution data.

[0072] Optionally, target probability distribution data is derived by fusing the posterior probability distributions of multiple partial discharge detection results. This data integrates information from multiple detection methods and reflects the system's assessment of the likelihood of different partial discharge types and its estimated discharge intensity. Based on the target probability distribution, the system selects the most likely discharge type as the target partial discharge type and estimates the discharge intensity level corresponding to the most consistent probability as the target discharge intensity, providing a scientific basis for health assessment of power equipment.

[0073] Optionally, the above steps can improve the accuracy and reliability of partial discharge diagnosis. By integrating the outputs of different partial discharge detection methods, the system's target probability distribution data can provide a more comprehensive and accurate estimate of discharge type and intensity. Compared to diagnostic methods that rely solely on a single detection method, this approach can effectively reduce false positives and missed detections, especially under complex operating conditions. This method is more adaptable to identifying various discharge phenomena, providing solid data support for maintenance and preventive repair plans for power equipment.

[0074] From the above content, it can be seen that the present application adopts M collection strategies to collect M groups of partial discharge data of power equipment, and screens and reduces the dimension of the features of each group of partial discharge data through the target verification strategy, thereby achieving the purpose of constructing an optimized set of M groups of features with the greatest correlation with the preset target, thereby ensuring that the selected feature set can fully reflect the characteristics of partial discharge without increasing the model complexity or decreasing the diagnostic accuracy due to the high redundancy between the features, thereby solving the technical problem in the prior art of using a single detection method to detect various defects insensitively or erroneously diagnose them in partial discharge detection.

[0075] In an optional embodiment, the process of screening and reducing the dimension of the features of each group of partial discharge data by using a target verification strategy to obtain M groups of feature sets includes: calculating, according to the target verification strategy, a correlation value between each feature of the i-th group of partial discharge data and various partial discharge types, where i is a positive integer less than or equal to M, and the correlation value is used to quantitatively represent the strength of the correlation between each feature and various partial discharge types; for the i-th group of partial discharge data, screening multiple features having correlation values ​​greater than a preset threshold as candidate features; calculating the mutual information between the multiple candidate features; and selecting, based on the mutual information, at least one target feature from the multiple candidate features to form a set of features corresponding to the i-th group of partial discharge data.

[0076] Optionally, the system can first employ a targeted validation strategy to quantitatively assess the strength of association between each feature and various PD types. A chi-square test is used to calculate the "association value," or chi-square value, between each feature and the PD classification label. A higher chi-square value indicates a stronger correlation between the feature and the classification, suggesting that the feature is important for distinguishing different types of PD. Based on this assessment, the system can set a preset threshold; all features with chi-square values ​​above this threshold are considered "candidate features" for further analysis and dimensionality reduction.

[0077] Next, for the i-th set of partial discharge data, the system compares the correlation values ​​of the features with a preset threshold and selects those with correlation values ​​exceeding the threshold as candidate features. The threshold directly affects the simplification of the feature set and the performance of the diagnostic model.

[0078] After initially screening candidate features, the system can refine them using the maximum relevance / minimum redundancy algorithm to remove redundancy and ensure a concise and efficient model. Specifically, the system uses the maximum relevance / minimum redundancy algorithm to calculate the mutual information between the candidate features to assess their correlation. Mutual information is a statistic used to measure the degree of dependence between two random variables. A larger value indicates less duplication of information between the two features, meaning they carry independent information.

[0079] Finally, based on the results of the mutual information analysis, the system selects target features with high information content and low cross-correlation from the candidate features to form a feature set corresponding to the i-th group of partial discharge data. This feature set not only contains information that is critical for partial discharge type identification, but also avoids feature redundancy, improving the efficiency and accuracy of the model.

[0080] For example, a partial discharge monitoring system for power transformers utilizes both ultra-high frequency (UHF) and ultrasonic methods. Initially, the partial discharge data from each method includes numerous characteristic parameters: the UHF method captures characteristics such as signal strength, frequency distribution, peak duration, and bandwidth; the ultrasonic method records information such as signal amplitude, frequency characteristics, and dispersion entropy.

[0081] The system first calculates the correlation between each feature and discharge types such as corona discharge, creeping discharge, and suspension discharge. For example, it found that the signal intensity feature in the UHF method has a strong correlation with the suspension discharge type, while the signal amplitude in the ultrasonic method is more sensitive to the corona discharge type.

[0082] Next, the system sets a reasonable correlation threshold, such as 0.6, and selects features with correlation values ​​greater than this threshold. For UHF data, features such as signal strength and peak time are considered; for ultrasonic data, features such as signal amplitude and dispersion entropy are retained.

[0083] Subsequently, the system further calculated the mutual information between these alternative features and found that the mutual information between signal strength and peak time was large, indicating that there was redundant information in these two features, while the mutual information between signal amplitude and dispersion entropy was small, which meant that the two could provide complementary information.

[0084] Ultimately, based on the results of the mutual information calculation, the system decided to use signal strength (UHF method) and signal amplitude (ultrasonic method) as target features, constructing a simplified feature set. This feature set not only covers sensitive information for various discharge types but also avoids redundancy between features, significantly improving the training efficiency and prediction accuracy of the subsequent support vector machine (SVM) model.

[0085] Optionally, by combining the chi-square test and the maximum relevance minimum redundancy algorithm for feature screening and dimensionality reduction, the system can effectively identify the features most critical for identifying partial discharge types while eliminating unnecessary redundant information. The resulting M sets of features not only improve the training efficiency of the subsequent SVM model but also enhance the model's generalization and diagnostic accuracy. Compared to models without feature optimization, models processed using the chi-square test and maximum relevance minimum redundancy algorithm perform better when processing complex nonlinear data, and can more accurately and comprehensively reflect the diversity and complexity of partial discharge in power equipment.

[0086] Optionally, Figure 2 is a schematic diagram of an optional construction of a corresponding neural network model according to the present application, such as Figure 2As shown, the system first collects data from power equipment using various partial discharge detection methods, including pulse current, optical measurement, ultra-high frequency (UHF), and ultrasonic methods. The resulting data includes pulse current, optical signal, UHF, and ultrasonic data. The preprocessing phase includes filtering, standardization, and outlier removal to ensure data purity and applicability. The system then extracts key discharge characteristic parameters, such as average discharge amplitude and discharge frequency. Feature selection and dimensionality reduction are performed using the chi-square test and maximum-relevant-minimum-redundancy (mRMR) method to construct an optimized feature set.

[0087] Then, based on the optimized feature set, the system can use machine learning algorithms such as support vector machines (SVM) to build a recognition model for the type and severity of partial discharge. During the model training process, it is necessary not only to pay attention to the model's classification ability, but also to use the posterior probability to convert the model's classification results into a probability distribution. That is, for a certain input, the model can not only give a hard classification label (such as "corona discharge"), but also provide a posterior probability to indicate the credibility of this classification result. For example, under ultrasonic testing, for a sample, the SVM model may predict it as "suspended discharge" and give a posterior probability of 97.3%, which means that the model has a high confidence that this sample belongs to the "suspended discharge" category.

[0088] Then, after obtaining the posterior probabilities under different detection methods, decision fusion is performed using the basic probability distribution in the Dempster-Shafer evidence theory (DS). The posterior probabilities obtained for the same sample by different detection methods are considered different sources of evidence. These evidence are first converted into basic probability distributions and then fused using the Dempster-Shafer synthesis rule to obtain a comprehensive decision result. The calculation of the basic probability distribution not only considers the output of each model but also the credibility of the model itself. In this way, even uncertain predictions from a model can be incorporated into the final decision.

[0089] Finally, after the model is built, the system uses a test set to evaluate its performance, specifically its accuracy, such as the true positive rate (TPR) and positive prediction rate (PPV), to ensure the model's accuracy and reliability in identifying PD types and assessing discharge severity. Based on the system's comprehensive PD diagnostic results, the model can be further optimized, such as by adjusting the SVM parameters and kernel function type, to improve its performance.

[0090] Alternatively, according to the test results of this application, the chi-square test optimized set performed better in distinguishing discharge types, while the mRMR optimized set performed very similarly to the chi-square test optimized set in comprehensively identifying discharge types and their severity. Therefore, the chi-square test optimized set is recommended as the feature parameter set for multi-physics partial discharge detection to achieve the best diagnostic results.

[0091] In an optional embodiment, a partial discharge detection result is determined based on each group of feature sets to obtain M partial discharge detection results, including: inputting the i-th group of feature sets into a neural network model corresponding to the i-th group of feature sets, and determining a partial discharge detection result based on feature data in the i-th group of feature sets based on prior knowledge learned by the neural network model during a training phase, wherein the training data of the neural network model corresponding to the i-th group of feature sets includes: historical partial discharge data collected using the i-th collection strategy, and actual partial discharge types and actual discharge intensities corresponding to the historical partial discharge data.

[0092] Optionally, the model is trained on historical partial discharge data, along with the corresponding actual partial discharge types and discharge intensities. During training, the model learns the characteristic patterns associated with specific partial discharge types (e.g., corona discharge, creeping discharge, and suspension discharge), as well as the relationship between these characteristics and discharge intensities (mild, moderate, and severe).

[0093] Alternatively, this approach can significantly improve the accuracy and reliability of partial discharge diagnosis. Each feature set is carefully screened and optimized to ensure the quality of the model input data and avoid interference from redundant information. Furthermore, the use of a neural network model, particularly its self-learning and adaptive capabilities, can process complex, nonlinear feature data, enabling more refined type identification and intensity assessment in partial discharge detection. Compared to traditional detection methods, this neural network-based intelligent diagnostic approach is more adaptable to changes in the operating environment of power equipment, promptly capturing potential discharge danger signals, and providing a more intelligent and effective monitoring method for the safe operation of power systems.

[0094] In an optional embodiment, M posterior probability distribution data are fused to obtain target probability distribution data, including: setting an identification framework, wherein the identification framework includes at least a set of local discharge types and a random element; determining a basic probability assignment corresponding to each posterior probability distribution data and the identification framework, wherein the basic probability assignment corresponding to each posterior probability distribution data is used to characterize, for each element in the identification framework, the probability that the posterior probability distribution data supports the element as a true situation; and fusing the target probability distribution data based on the M basic probability assignments corresponding to the M posterior probability distribution data.

[0095] Optionally, the identification framework is a fundamental concept in Dempster-Shafer evidence theory that defines all possible decision options. In this embodiment, the identification framework includes at least a set of partial discharge types (such as corona discharge, creeping discharge, and suspension discharge) and a random element representing uncertainty (such as "uncertain" or "unknown" discharge type).

[0096] Optionally, for each posterior probability distribution data output by each partial discharge detection method (a total of M), the system needs to determine a basic probability assignment. This basic probability assignment is a set of probability values ​​converted from the posterior probability for each element (discharge type or uncertainty) in the identification framework. It represents the degree of support the detection method has for that element (discharge type) being true. For example, for a particular partial discharge sample, the optical detection method might give a posterior probability of "corona discharge" of 95%. In this case, the basic probability assignment for "corona discharge" in the identification framework would be close to 0.95, while the basic probability assignments for other discharge types and supporting uncertainties would be correspondingly lower.

[0097] Optionally, after obtaining the posterior probability distribution data and corresponding basic probability assignments for each of the M detection methods, the system can fuse this scattered information to produce a target probability distribution that integrates information from all detection methods. This fusion process follows Dempster's synthesis rule, which effectively handles conflicting evidence and produces an updated probability distribution, the target probability distribution data. This data reflects the combined probability that supports the true situation for each partial discharge type after fusion.

[0098] In an optional embodiment, M posterior probability distribution data are fused to obtain target probability distribution data, including: determining difference information between the M posterior probability distribution data for each element in the recognition frame; determining a target probability assignment for each element in the recognition frame based on the difference information; and determining the target probability distribution data based on the target probability assignment for each element in the recognition frame.

[0099] Optionally, the system fuses M posterior probability distributions from different detection methods to produce a more comprehensive and accurate target probability distribution. First, the differences between the posterior probability distributions from the M detection methods must be identified and analyzed. This step aims to quantify the degree of inconsistency between the diagnostic results of each detection method for each element in the identification framework (i.e., PD type or uncertainty). This can be achieved by comparing the distance, or degree of conflict, between the posterior probabilities output by each model.

[0100] Optionally, based on this discrepancy information, the system can determine a target probability assignment for each element in the recognition framework by employing appropriate algorithms or rules. For example, if the posterior probability distributions of “corona discharge” among the M models differ significantly, the target probability assignment will incorporate these differences to produce a final probability that reflects the combined opinions of all models.

[0101] Finally, based on the target probability assignments for each element in the recognition framework, the system generates target probability distribution data. This data includes all possible discharge types and their fused probability values, providing a unified and more reliable result for intelligent diagnosis of partial discharge.

[0102] In summary, by analyzing the differences in the outputs of M detection methods, the system can identify and resolve conflicts between them, ensuring consistency and stability in the final diagnostic results across multiple physical signal analyses. Furthermore, the fused target probability distribution data reflects the combined judgment of all detection methods, enabling more accurate localization of the type of partial discharge. This approach is particularly advantageous in complex situations or situations where a single method is difficult to determine. Furthermore, the target probability assignment process quantifies uncertainty. Through comprehensive evaluation, the system can effectively reduce diagnostic uncertainty caused by the limitations of individual detection methods or noise interference.

[0103] In an optional embodiment, the M acquisition strategies include at least two of the following acquisition strategies: a first acquisition strategy is used to collect charge amount change data caused by local discharge to capture the intensity and frequency of discharge events; and determine the local discharge data of the power equipment based on the intensity and frequency of the discharge events; a second acquisition strategy is used to collect light radiation intensity change data during local discharge; and determine the local discharge data of the power equipment based on the light radiation intensity change data; a third acquisition strategy is used to collect electromagnetic wave signals generated by local discharge, and determine the local discharge data of the power equipment based on the electromagnetic wave signals; and a fourth acquisition strategy is used to collect ultrasonic signals generated during local discharge, and determine the local discharge data of the power equipment based on the ultrasonic signals.

[0104] Different strategies can be used to capture different types of partial discharge characteristics. The first strategy focuses on the electrical characteristics of the discharge, the second on optical radiation, the third on electromagnetic waves, and the fourth on ultrasound. This multi-dimensional acquisition approach ensures the system can address various discharge patterns and reduces the likelihood of missed detections. Furthermore, each acquisition strategy offers specific advantages, such as the pulsed current method's sensitivity to charge variations, the optical method's real-time monitoring of corona discharges, the ultra-high frequency method's ability to detect hidden discharge sources, and the ultrasonic method's ability to locate discharge locations. Combining this information allows for more precise localization of the discharge source and analysis of its type and severity.

[0105] In an optional embodiment, the feature set includes at least one of the following features: average discharge amplitude, used to characterize the average value of the charge or voltage amplitude generated by the discharge event in the partial discharge detection cycle; discharge frequency, used to characterize the number of partial discharge events occurring per unit time; skewness, used to characterize the asymmetry of the probability density function distribution of the partial discharge signal; steepness, used to characterize the speed of the rising edge of the partial discharge pulse; phase asymmetry, used to characterize the degree of difference between the discharge characteristics of the positive half-cycle and the negative half-cycle in the partial discharge signal; and discharge asymmetry, used to characterize the asymmetry of the signal strength of the partial discharge in the entire power cycle.

[0106] Optionally, the system can more meticulously depict the characteristic patterns of partial discharges through multi-angle analysis of discharge signals, providing a rich and comprehensive data foundation for the subsequent construction of intelligent diagnostic models. In addition, by combining multiple characteristic parameters, the system can construct a more complex model architecture. By using algorithms such as support vector machines (SVM) and optimizing feature set selection, such as the chi-square test and maximum relevance minimum redundancy (mRMR), the model's classification ability and noise resistance can be significantly improved, enhancing the model's adaptability and robustness in different application scenarios. The system can effectively filter out noise signals, reduce false alarm rates and missed detection rates, and improve the credibility of discharge diagnosis by comprehensively analyzing the time characteristics (such as steepness and phase asymmetry), amplitude characteristics (such as average discharge amplitude and discharge asymmetry), and statistical characteristics (such as discharge frequency and skewness) of the discharge signal.

[0107] According to another aspect of the present application, a partial discharge detection device based on multi-physics information fusion is also provided, wherein: Figure 3 is a schematic diagram of an optional partial discharge detection device based on multi-physics information fusion according to an embodiment of the present application, such as Figure 3 As shown, the partial discharge detection device based on multi-physics information fusion includes: an acquisition unit 301 , a processing unit 302 , a first determination unit 303 , a conversion unit 304 , and a second determination unit 305 .

[0108] Optionally, the acquisition unit 301 is configured to acquire M groups of partial discharge data of the power equipment using M acquisition strategies, where M is an integer greater than 1, and each acquisition strategy corresponds to a group of partial discharge data. The first processing unit 302 is configured to screen and reduce the dimensions of the features of each group of partial discharge data using a target verification strategy to obtain M groups of feature sets, where the M groups of feature sets correspond one-to-one to the M groups of partial discharge data, and the target verification strategy is configured to detect the correlation between the features and various partial discharge types, and select the feature set with the minimum mutual correlation and the maximum correlation with a preset target. The first determination unit 303 is configured to determine a partial discharge detection result based on each group of feature sets to obtain M partial discharge detection results. The second processing unit 304 is configured to convert each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding one-to-one to the M partial discharge detection results, and then fuse the M posterior probability distribution data to obtain target probability distribution data. The second determination unit 305 is configured to determine a target partial discharge type and target discharge intensity of the power equipment based on the target probability distribution data.

[0109] Optionally, the processing unit includes: a verification subunit, a first selection subunit, a first processing subunit, and a second selection subunit. The verification subunit is configured to calculate, based on a target verification strategy, a correlation value between each feature of the i-th group of partial discharge data and various partial discharge types, where i is a positive integer less than or equal to M, and the correlation value is used to quantitatively characterize the strength of the correlation between each feature and various partial discharge types; the first selection subunit is configured to screen, for the i-th group of partial discharge data, multiple features having correlation values ​​greater than a preset threshold as candidate features; the processing subunit is configured to calculate mutual information between the multiple candidate features; and the second selection subunit is configured to select, based on the mutual information, at least one target feature from the multiple candidate features to form a set of features corresponding to the i-th group of partial discharge data.

[0110] Optionally, the first determination unit includes: a first determination subunit. The first determination subunit is configured to input the i-th feature set into a neural network model corresponding to the i-th feature set, and determine a partial discharge detection result based on feature data in the i-th feature set according to prior knowledge learned by the neural network model during a training phase, wherein the training data of the neural network model corresponding to the i-th feature set includes: historical partial discharge data collected using the i-th collection strategy, and actual partial discharge types and actual discharge intensities corresponding to the historical partial discharge data.

[0111] Optionally, the second processing unit 304 includes: a second processing subunit, a second determination subunit, and a third processing subunit. The second processing subunit is configured to set an identification framework, wherein the identification framework includes at least a set of partial discharge types and a random element; the second determination subunit is configured to determine, for each posterior probability distribution data and the identification framework, a basic probability assignment corresponding to each posterior probability distribution data, wherein the basic probability assignment corresponding to each posterior probability distribution data is configured to represent, for each element in the identification framework, the probability that the posterior probability distribution data supports the element as being true; and the third processing subunit is configured to fuse the M basic probability assignments corresponding to the M posterior probability distribution data to obtain target probability distribution data.

[0112] Optionally, the second processing unit 304 further includes: a third determining subunit, a fourth determining subunit, and a fifth determining subunit. The third determining subunit is configured to determine difference information between the M posterior probability distribution data for each element in the recognition frame; the fourth determining subunit is configured to determine a target probability value for each element in the recognition frame based on the difference information; and the fifth determining subunit is configured to determine the target probability distribution data based on the target probability value for each element in the recognition frame.

[0113] According to another aspect of the present application, a computer-readable storage medium is provided, comprising: a computer program stored in the computer-readable storage medium, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the above-mentioned partial discharge detection method.

[0114] According to another aspect of the present application, an electronic device is provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned partial discharge detection method.

[0115] The above-mentioned embodiments or examples disclosed in this application are not exhaustive, but are only illustrations of some embodiments or examples, and are not intended to be specific limitations on the scope of protection disclosed in this application. In the absence of contradiction, each step in a certain embodiment or example in this application can be implemented as an independent example, and the steps can be arbitrarily combined. For example, the solution after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be arbitrarily combined; in addition, the various embodiments or examples can be arbitrarily combined. For example, some or all of the steps in different embodiments or examples can be arbitrarily combined, and a certain embodiment or example can be arbitrarily combined with the optional methods or optional examples of other embodiments or examples.

[0116] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0117] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0119] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected to achieve the purpose of the present embodiment according to actual needs.

[0120] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code.

[0122] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A partial discharge detection method, characterized in that: include: collecting M groups of partial discharge data of the power equipment using M collection strategies, where M is an integer greater than 1, and each collection strategy corresponds to a group of partial discharge data; The features of each set of partial discharge data are screened and dimensionally reduced using a target verification strategy to obtain M sets of feature sets, wherein the M sets of feature sets correspond one-to-one to the M sets of partial discharge data. The target verification strategy is used to detect the correlation between the features and various partial discharge types, and select the feature set with the minimum mutual correlation and the maximum correlation with a preset target. Determine a partial discharge detection result according to each set of feature sets, and obtain M partial discharge detection results; After converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding to the M partial discharge detection results, the M posterior probability distribution data are fused to obtain target probability distribution data; A target partial discharge type and a target discharge intensity of the electrical equipment are determined according to the target probability distribution data.

2. The method according to claim 1, characterized in that In the process of screening and reducing the dimension of the features of each set of partial discharge data by using the target verification strategy to obtain M sets of feature sets, the method further includes: calculating, according to the target verification strategy, a correlation value between each feature of the i-th group of partial discharge data and various partial discharge types, where i is a positive integer less than or equal to M, and the correlation value is used to quantitatively represent the strength of the correlation between each feature and various partial discharge types; For the i-th group of partial discharge data, screening multiple features with correlation values ​​greater than a preset threshold as candidate features; Calculating mutual information between a plurality of the candidate features; At least one target feature is selected from the plurality of candidate features according to the mutual information to form a feature set corresponding to the i-th group of partial discharge data.

3. The method according to claim 1, characterized in that A partial discharge detection result is determined based on each set of feature sets, and M partial discharge detection results are obtained, including: An i-th feature set is input into a neural network model corresponding to the i-th feature set, and a partial discharge detection result is determined based on feature data in the i-th feature set based on prior knowledge learned by the neural network model during a training phase, wherein the training data of the neural network model corresponding to the i-th feature set includes: historical partial discharge data collected using the i-th collection strategy, and actual partial discharge types and actual discharge intensities corresponding to the historical partial discharge data.

4. The method according to claim 1, wherein The M posterior probability distribution data are fused to obtain target probability distribution data, including: Setting an identification framework, wherein the identification framework includes at least a set of partial discharge types and a random element; Determining, for each posterior probability distribution data and the recognition framework, a basic probability assignment corresponding to each posterior probability distribution data, wherein the basic probability assignment corresponding to each posterior probability distribution data is used to represent, for each element in the recognition framework, the probability that the posterior probability distribution data supports the element as a true situation; The target probability distribution data is obtained by fusing the M basic probability assignments corresponding to the M posterior probability distribution data.

5. The method according to claim 4, characterized in that The M posterior probability distribution data are fused to obtain target probability distribution data, including: Determining difference information between the M posterior probability distribution data for each element in the recognition framework; determining a target probability assignment for each element in the recognition frame according to the difference information; The target probability distribution data is determined according to the target probability assignment of each element in the recognition frame.

6. The method according to any one of claims 1 to 5, characterized in that The M acquisition strategies include at least two of the following acquisition strategies: A first acquisition strategy is used to acquire charge amount change data caused by partial discharge to capture the intensity and frequency of discharge events; and to determine partial discharge data of the power equipment based on the intensity and frequency of the discharge events; The second acquisition strategy is used to acquire light radiation intensity variation data during the partial discharge process; and determine the partial discharge data of the power equipment based on the light radiation intensity variation data; a third acquisition strategy for acquiring electromagnetic wave signals generated by partial discharge, and determining partial discharge data of the power equipment based on the electromagnetic wave signals; The fourth acquisition strategy is used to acquire ultrasonic signals generated during partial discharge, and determine partial discharge data of the power equipment according to the ultrasonic signals.

7. The method according to any one of claims 1 to 5, characterized in that The feature set includes at least one of the following features: Average discharge amplitude, which is used to characterize the average value of the charge or voltage amplitude generated by the discharge event during the partial discharge detection cycle; Discharge frequency is used to characterize the number of partial discharge events per unit time; Skewness is used to characterize the asymmetry of the probability density function distribution of the partial discharge signal; Steepness, used to characterize the speed of the rising edge of the partial discharge pulse; Phase asymmetry is used to characterize the difference between the discharge characteristics of the positive half-cycle and the negative half-cycle in the partial discharge signal; Discharge asymmetry is used to characterize the asymmetry of the signal strength of partial discharge throughout the power cycle.

8. A partial discharge detection device, characterized in that: include: The collection unit is used to collect M groups of partial discharge data of power equipment through M collection strategies, wherein: M is an integer greater than 1, and each acquisition strategy corresponds to a set of partial discharge data; a first processing unit, configured to screen and reduce the dimension of features of each set of partial discharge data using a target verification strategy to obtain M sets of feature sets, wherein the M sets of feature sets correspond one-to-one to the M sets of partial discharge data, and the target verification strategy is configured to detect correlations between the features and various partial discharge types, and select a feature set having the least mutual correlation and the greatest correlation with a preset target; a first determining unit, configured to determine a partial discharge detection result according to each set of feature sets, to obtain M partial discharge detection results; a second processing unit for converting each partial discharge detection result into a posterior probability distribution data to obtain M posterior probability distribution data corresponding one-to-one to the M partial discharge detection results, and fusing the M posterior probability distribution data to obtain target probability distribution data; The second determining unit is configured to determine a target partial discharge type and a target discharge intensity of the power equipment according to the target probability distribution data.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the partial discharge detection method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The device comprises one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to perform the partial discharge detection method according to any one of claims 1 to 7.

Images