Power equipment fault data processing method and system based on power big data

Abstract

The invention discloses a power equipment fault data processing method and system based on power big data, and the method comprises the steps: converting original fault data into a high-dimensional vector through structural coding, carrying out the dynamic distillation of the high-dimensional vector through a feature selection mechanism based on deep learning, screening out a distillation set representing the fault nature, and carrying out the dynamic distillation of the high-dimensional vector; the problem of feature redundancy of traditional fixed rule screening is solved, and accurate evaluation of the fault abnormal degree is further achieved through dynamic construction of fault data broken lines and quantitative calculation of fault characterization coefficients. Through the mode, the problems of low processing efficiency, insufficient fault mode identification capability under complex working conditions and the like caused by dependence on manual rules in a traditional method are solved, and high-precision and extensible technical support is provided for intelligent operation and maintenance of power grid equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and more specifically, to a power equipment fault data processing method and system based on power big data. BACKGROUND

[0002] With the intelligent development of the power system, the power equipment fault data processing technology has become a key link to ensure the reliable operation of the power grid. The traditional power equipment fault processing method usually relies on manual experience or a fixed rule-based screening mechanism, which faces significant bottlenecks in the power big data scenario: on the one hand, the data generated by the power equipment presents the characteristics of massiveness, high dimensionality, and heterogeneity, and the processing flow based on manual rules is difficult to realize real-time analysis and dynamic adaptation, which is prone to system response delay due to data redundancy; on the other hand, the traditional method lacks the ability to deeply mine the internal correlation of data, and it is difficult to distinguish the implicit fault mode under complex working conditions through simple logical screening or manual intervention, which is prone to misjudgment or missed detection problems.

[0003] Therefore, an optimized power equipment fault data processing method based on power big data is expected. SUMMARY

[0004] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a power equipment fault data processing method and system based on power big data, which first converts the original fault data into a high-dimensional vector through structured coding, and uses a feature selection mechanism based on deep learning to dynamically distill the high-dimensional vector, and selects a distillation set representing the essence of the fault, to overcome the feature redundancy problem of traditional fixed rule screening, and further realizes the accurate evaluation of the fault abnormality degree through the dynamic construction of the fault data fold line and the quantitative calculation of the fault representation coefficient. In this way, the problems of low processing efficiency and insufficient fault mode recognition ability under complex working conditions caused by relying on manual rules in the traditional method are solved, and high-precision and scalable technical support is provided for the intelligent operation and maintenance of the power grid equipment.

[0005] According to one aspect of the present application, a power equipment fault data processing method based on power big data is provided, which comprises:

[0006] real-time collection of fault data of an electric energy meter to obtain a set of fault data;

[0007] performing an unsupervised learning-based fault data filter on the set of fault data to obtain a distillation set of fault data;

[0008] constructing a fault data fold line based on the distillation set of fault data, and calculating a fault representation coefficient of the electric energy meter based on the fault data fold line;

[0009] The abnormality degree of the fault of the electric energy meter is determined based on comparison between the fault characterization coefficient of the electric energy meter and a fault characterization coefficient threshold of the electric energy meter.

[0010] According to another aspect of the present application, a power equipment fault data processing system based on power big data is provided, comprising:

[0011] A fault data collection module is configured to collect fault data of the electric energy meter in real time to obtain a set of fault data.

[0012] A fault data screening module is configured to perform a fault data screener based on unsupervised learning on the set of fault data to obtain a distilled set of fault data.

[0013] A fault characterization coefficient calculation module is configured to construct a fault data fold line based on the distilled set of fault data, and calculate the fault characterization coefficient of the electric energy meter based on the fault data fold line.

[0014] An abnormality degree determination module is configured to determine the abnormality degree of the fault of the electric energy meter based on comparison between the fault characterization coefficient of the electric energy meter and a fault characterization coefficient threshold of the electric energy meter.

[0015] Compared with the prior art, the power equipment fault data processing method and system based on power big data provided by the present application first converts the original fault data into a high-dimensional vector through structured coding, and dynamically distills the high-dimensional vector by using a feature selection mechanism based on deep learning to screen out a distilled set representing the essence of the fault, so as to overcome the feature redundancy problem of traditional fixed rule screening, and further realize accurate evaluation of the abnormality degree of the fault through dynamic construction of a fault data fold line and quantitative calculation of a fault characterization coefficient. In this way, the problems of low processing efficiency and insufficient fault mode recognition ability under complex working conditions caused by the dependence of traditional methods on artificial rules are solved, and high-precision and scalable technical support is provided for intelligent operation and maintenance of power grid equipment. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of embodiments of the present application and constitute a part of the specification, together with the description, to explain the present application and, together with the description, serve to explain the present application. The drawings do not constitute a limitation on the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 A flowchart of the power equipment fault data processing method based on power big data according to the embodiments of the present application;

[0018] Figure 2A data flow diagram of the power equipment fault data processing method based on power big data according to the embodiment of the present application;

[0019] Figure 3 A flow chart of sub-step S2 of the power equipment fault data processing method based on power big data according to the embodiment of the present application;

[0020] Figure 4 A block diagram of the power equipment fault data processing system based on power big data according to the embodiment of the present application. DETAILED DESCRIPTION

[0021] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of embodiments of the present application, and thus are not to be used to limit the whole embodiments of the present application, and it should be appreciated that the present application covers all the technical solutions falling within the scope of the present application.

[0022] As shown in the present application and claims, unless the context clearly indicates otherwise, the words "one", "an", "a", and / or "the" do not mean to specify a single number, but also include a plurality. Generally, the terms "comprising" and "including" only indicate including the steps and elements clearly indicated, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0023] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0024] Flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. Instead, various steps can be processed in reverse order or simultaneously, as needed. Meanwhile, other operations can be added to these processes, or one or more steps of the operations can be removed from these processes.

[0025] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of embodiments of the present application, and thus are not to be used to limit the whole embodiments of the present application, and it should be appreciated that the present application covers all the technical solutions falling within the scope of the present application.

[0026] In the technical solution of the present application, a power equipment fault data processing method based on power big data is proposed. Figure 1 A flow chart of the power equipment fault data processing method based on power big data according to the embodiment of the present application. Figure 2A data flow diagram of the power equipment fault data processing method based on power big data according to the embodiment of the application is shown in Figure 1. As shown in Figure 1, the power equipment fault data processing method based on power big data according to the embodiment of the application comprises the following steps: S1, collecting fault data of an electric energy meter in real time to obtain a set of fault data; S2, performing a fault data filter based on unsupervised learning on the set of fault data to obtain a distilled set of fault data; S3, constructing a fault data broken line based on the distilled set of fault data, and calculating a fault characteristic coefficient of the electric energy meter based on the fault data broken line; and S4, determining an abnormality degree of the electric energy meter fault based on a comparison between the fault characteristic coefficient of the electric energy meter and a fault characteristic coefficient threshold of the electric energy meter. Figure 1 and Figure 2 As shown in Figure 1, the power equipment fault data processing method based on power big data according to the embodiment of the application comprises the following steps: S1, collecting fault data of an electric energy meter in real time to obtain a set of fault data; S2, performing a fault data filter based on unsupervised learning on the set of fault data to obtain a distilled set of fault data; S3, constructing a fault data broken line based on the distilled set of fault data, and calculating a fault characteristic coefficient of the electric energy meter based on the fault data broken line; and S4, determining an abnormality degree of the electric energy meter fault based on a comparison between the fault characteristic coefficient of the electric energy meter and a fault characteristic coefficient threshold of the electric energy meter.

[0027] In particular, the S1, collecting fault data of an electric energy meter in real time to obtain a set of fault data. In a traditional operation and maintenance mode, fault data is often obtained through periodic inspection or passive reception of alarm signals. This discrete and fragmented collection method is difficult to capture dynamic fault characteristics in the process of equipment operation. In one specific example of the present application, by deploying edge computing nodes and Internet of Things communication protocols, the system can continuously capture event logs, voltage and current waveform distortion, temperature sensor readings, and communication status codes of the electric energy meter at a millisecond level of time resolution, and construct an original fault data set containing time stamps, device IDs, fault codes, environmental variables, and other multi-modal features. By collecting fault data of the electric energy meter in real time, the system can record the evolution trajectory of the fault from the initial signs to complete failure, providing necessary time sequence correlation support for deep learning models to mine fault evolution rules.

[0028] In particular, the S2, performing a fault data filter based on unsupervised learning on the set of fault data to obtain a distilled set of fault data. In one specific example of the present application, as shown in Figure 2, the S2 comprises the following steps: S21, structurally encoding each fault data in the set of fault data to obtain a set of fault data structured encoding vectors; S22, performing fault data distillation extraction based on a fault characteristic topology driven fault data filter on the set of fault data structured encoding vectors to obtain a distilled set of fault data structured encoding vectors; and S23, determining the distilled set of fault data based on the distilled set of fault data structured encoding vectors. Figure 3

[0029] ​Specifically, the S21, each fault data in the set of fault data is structured coding to get a set of fault data structured coding vector. It should be understood that the power meter original fault data usually contains event log, voltage and current waveform segment, temperature sensor discrete sampling value and communication protocol state code and other multi-source heterogeneous information, these data exist significant difference in time resolution, dimension and semantic level, directly lead to traditional rule-based feature extraction method difficult to effectively fuse multi-dimensional fault feature. Therefore, in the technical scheme of the present application, the fault data embedding matrix is used for structured coding to establish a unified representation space of cross-modal data, and a set of fault data structured coding vectors is obtained. In this process, the fault data embedding matrix maps fault attributes of different dimensions (such as category type label of fault code, continuous type value of waveform sampling, time sequence label of communication state) into high-dimensional vectors with semantic association, thereby solving the feature alignment failure problem caused by data heterogeneity in traditional methods. The generated set of fault data structured coding vectors not only retains the statistical characteristics of the original fault data, but also implicitly captures the nonlinear relationship between fault patterns, thereby providing a solid foundation for subsequent fault analysis.

[0030] Specifically, the S22 performs fault data distillation extraction on the set of structured and coded fault data vectors based on a fault characterization topology to obtain a distilled set of structured and coded fault data vectors. It should be understood that although the set of structured and coded fault data vectors achieves unified representation of multi-source heterogeneous data, there are still a large number of redundant features and noise interference in the high-dimensional vectors. For example, when a certain electric energy meter has communication abnormalities, it may trigger multiple-dimensional data fluctuations such as signal strength fluctuations, instantaneous power jumps, and slight changes in environmental temperature. Traditional feature selection based on artificial experience is difficult to distinguish which are core fault indicators and which are incidental associated noise. Therefore, in the technical solution of the present application, the set of structured and coded fault data vectors is subjected to fault data distillation extraction based on a fault characterization topology to obtain a distilled set of structured and coded fault data vectors. That is, a group distribution semantic graph is constructed, and individual fault vectors are placed in a global fault mode network for dynamic evaluation. In this process, first, the group distribution semantic graph is learned from massive historical fault data. This graph is modeled by a scale-invariant correlation matrix, which captures the high-order correlation of fault features in the spatial topology (co-occurrence probability of voltage harmonic distortion and communication delay, nonlinear coupling relationship between temperature gradient change and measurement error, etc.). Then, the individual feature vector of the fault data is extracted and mapped to the feature space of the semantic graph to quantify the critical dimension of its local associated defects in the global universality, generating a fault modulation individual feature vector that integrates group context information. In particular, in this process, if the correlation of the fault feature (such as instantaneous current spike) with the core fault mode (such as insulation aging) in the global semantic graph is lower than a preset threshold, it will be determined as incidental noise and removed, obtaining a distilled set of structured and coded fault data vectors. It is worth mentioning that the distilled set of structured and coded fault data vectors can exhibit clear fault cluster separation characteristics in visual analysis, such as communication module faults and measurement unit faults forming independent clusters in the vector space, laying a data foundation for subsequent accurate calculation of fault characterization coefficients. This group semantic driven selection mechanism not only breaks through the static limitations of traditional rule engines, but also adapts to the evolution of power equipment operating states, providing a feature optimization solution with self-evolution ability for power grid operation and maintenance.

[0031] Specifically, first, a fault data population distribution semantic graph of a set of fault data structured encoding vectors is calculated. It should be understood that a single fault data structured encoding vector can only reflect an isolated segment of the transient operating condition of the device, and the actual fault mode is often hidden in the spatial distribution, time sequence correlation and cross-device data coupling of multi-dimensional features. In the traditional method, due to the lack of understanding of the overall distribution law of the feature set, the potential synergistic mechanism between features is easily broken. Therefore, in the technical solution of the present application, a fault data population distribution semantic graph is constructed to capture the complex semantic network between fault features through unsupervised graph structured learning. In this process, the system inputs the set of structured encoding fault vectors (including fault codes, waveform distortion spectrum, environmental temperature and humidity time sequence, etc. multi-dimensional features) into a graph neural network, models the spatial topological relationship of the fault data structured encoding vectors using a scale-invariant correlation matrix, for example, by calculating the co-occurrence probability of the voltage fluctuation amplitude and the communication error rate, analyzing the phase correlation of the temperature gradient change and the measurement error curve, to obtain the fault data population distribution semantic graph. The fault data population distribution semantic graph captures the implicit knowledge network formed by the power equipment group during operation, reveals the cluster distribution law of the fault data at the feature space level, and constructs the context association system of the fault features at the semantic level, overcoming the disadvantages of frequent manual parameter adjustment of fixed rule systems. In one specific example of the present application, the fault data population distribution semantic graph of the set of fault data structured encoding vectors is calculated as follows:

[0032] X={x1,x2,...,x i ,...,x n}

[0033]

[0034] Wherein, X represents the set of fault data structured encoding vectors, x1, x2, x i , x j , x n represent the first, second, i-th, j-th, and n-th fault data structured encoding vectors in the set of fault data structured encoding vectors, M represents the fault data population distribution semantic graph, R(x1, x1), R(x n , x1), R(x1, x n ), R(x n , x n ) and R(x i , x j ) represent the feature values at positions (1, 1), (n, 1), (1, n), (n, n) and (i, j) in the fault data population distribution semantic graph.

[0035] Then, the ith fault data structured encoding vector is extracted from the set of fault data structured encoding vectors as a fault data individual feature vector. Each fault data individual feature vector represents a multi-dimensional feature representation of a fault instance, for example, in the context of power meter fault data analysis, this can involve a combination of current intensity, voltage level, temperature variation, and other parameters, and the ith fault data structured encoding vector specifically describes the working state of the power meter at a certain time or under certain conditions. By extracting the ith fault data structured encoding vector from the set of fault data structured encoding vectors, the system can analyze each fault data point independently, gain a deeper understanding of its specific form of expression when the fault occurs, and distinguish it from other faults. In addition, by mapping the fault data individual feature vector into the space of the entire fault data group distribution semantic graph, the uniqueness of the individual feature vector relative to the overall distribution can be quantified, thereby more accurately determining whether it truly reflects the underlying fault pattern. This method not only improves the accuracy of fault detection, reduces the possibility of false positives and false negatives, but also enhances the ability to identify implicit fault patterns under complex working conditions through detailed analysis of individual features, providing strong support for the intelligent operation and maintenance of power grid equipment.

[0036] Then, the fault data individual feature vector is mapped into the fault feature space of the fault data group distribution semantic graph to obtain a fault modulation individual feature vector. It should be understood that in the actual application of power equipment fault diagnosis, individual fault features often fail to accurately reveal the nature of the fault due to the lack of a global perspective. For example, a power meter in an industrial park frequently experiences voltage sag alarms, and traditional methods may only rely on single-point thresholding for judgment, ignoring the special association pattern of this feature in the group semantic graph - when the individual voltage sag feature vector is mapped into the group distribution semantic graph, it is found that there is a spatiotemporal coupling between this feature and the nearby substation capacitor switching event, which cannot be revealed in single-device data. Therefore, in the technical solution of the present application, by mapping the fault data individual feature vector into the fault feature space of the fault data group distribution semantic graph, the limitations of isolated fault data analysis are overcome, enabling the system to place individual fault features in the fault data group distribution semantic graph to achieve fault context enhancement. This group context enhancement mechanism not only improves the representation ability of the features, but also establishes an individual-group feature dialogue channel, providing precise data support for differentiated operation and maintenance of power equipment. In one specific example of the present application, the fault data individual feature vector is mapped into the fault feature space of the fault data group distribution semantic graph to obtain a fault modulation individual feature vector according to the following formula:

[0037]

[0038] where [x i ; xj ] represents the concatenation of x i and x j , W r represents a trainable weight matrix, b represents a trainable bias vector, r i,j represents x i and x j , sigmoid represents a sigmoid activation function, r i,j,h represents a feature value at the hth position in r i,j , T represents the length of r i,j , represents matrix multiplication, v i represents the ith failure modulation individual feature vector.

[0039] Then, the fault modulation individual feature vector is globally and universally calibrated to obtain an optimized fault modulation individual feature vector. It should be understood that the power equipment fault data has the characteristics of high dimension, heterogeneity and dynamic change. The group distribution semantic graph constructed directly through local feature correlation may cause abnormal energy channel coding of the global semantic background field due to local correlation defects (such as imperfection of the feature space topological structure or distortion of near-field energy transmission). Specifically, in the screening process of the fault feature vector, the regional correlation constraint may cause insufficient compatibility of the fault modulation individual feature vector and the global semantic configuration, thereby affecting the accuracy of the overall-local difference significance measure of the fault mode recognition. For example, some hidden fault modes may not be effectively absorbed by the global semantic field due to local energy fluctuations, resulting in the deletion of key features in the distillation process or the retention of redundant features, and ultimately reducing the fault feature projection accuracy. Therefore, in one preferred example of the present application, the fault modulation individual feature vector is globally and universally calibrated to obtain an optimized fault modulation individual feature vector. Specifically, in this process, the fault modulation individual feature vector is taken as a steady-state convergence kernel under the group distribution semantic graph, and the global semantic conjugate configuration is expanded by Beta functional, so as to correct the abnormal dimension of local correlation defects in global energy transmission; in addition, the critical dimension of global correlation is determined by calculating the i th fault tuning global outlier feature vector, and the local energy fluctuation is re-associated to the fixed point by combining the energy scale dimension propagation channel, realizing the global semantic unification in the dimensionless invariant. Through scale conservation correlation renormalization, the problem of uneven semantic near-field energy distribution caused by local correlation defects is effectively suppressed, so that the fault modulation individual feature vector can more accurately reflect the dynamic characteristics of the global semantic background field. For example, under complex working conditions, the optimized feature vector can avoid feature drift caused by local noise interference, thereby enhancing the recognition ability of hidden fault modes (such as transient harmonic anomalies or early-stage features of insulation aging). At the same time, through the regularization processing of the global coupling dimension, the screening process of the fault data distillation set can more efficiently eliminate redundant features and retain key information strongly related to the fault nature, ultimately improving the calculation accuracy of the fault representation coefficient and reducing the misjudgment rate. In one specific example of the present application, the fault modulation individual feature vector can be globally and universally calibrated to obtain an optimized fault modulation individual feature vector by the following steps: based on the fault modulation individual feature vector and the fault data individual feature vector, scale conservation correlation renormalization is performed on the fault data group distribution semantic graph to obtain a fault data group distribution semantic fixed correlation matrix; based on the fault modulation individual feature vector and the fault data individual feature vector, the fault data individual feature vector is corrected by outlier dimension normalization to obtain a fault tuning global outlier feature vector; based on the fault data group distribution semantic fixed correlation matrix and the fault tuning global outlier feature vector, the fault modulation individual feature vector is optimized to obtain an optimized fault modulation individual feature vector.

[0040] In this example, the fault modulation individual feature vector is globally universally calibrated to obtain the optimized fault modulation individual feature vector with the following formula; wherein the formula is:

[0041]

[0042] x ′i = ln(v i )⊙[ln(x i ) ⊙-1 ]

[0043]

[0044] Wherein, is the point multiplication according to the position, M' is the fault data group distribution semantic invariable correlation matrix, -1 indicates the reciprocal of each position characteristic value in the vector, ln indicates the logarithmic function value with natural constant e as the base, x' i is the i th fault tuning global outlier feature vector, v' i is the i th optimized fault modulation individual feature vector.

[0045] Further, the system calculates a fault data semantic discrete increment operator of the optimized fault modulation individual feature vector relative to the set of fault data structured coding vectors as a fault semantic contribution degree. It should be understood that when the power equipment generates a large amount of heterogeneous data, there are often a large number of redundant or low-value features in the high-dimensional vector set of the fault data after structured coding, for example, certain voltage fluctuation data may be irrelevant to the fault under certain working conditions, and the traditional fixed rules cannot dynamically identify such noise. At this time, the system quantifies the importance of each structured coding vector in the overall fault semantic expression by calculating the fault data semantic discrete increment operator (such as the information entropy contribution degree or attention weight of the feature vector in the fault group distribution) corresponding to each structured coding vector. In this process, the system maps the single fault data structured coding vector into the semantic graph with the fault group distribution semantic graph as the global background (such as the fault feature space topological relationship constructed by clustering or graph neural network), and analyzes the deviation degree of the single fault data structured coding vector from the group distribution. For example, if a certain coding vector is located in the core cluster area of the group distribution and is closely associated with most fault modes, its semantic discrete increment operator is high; on the contrary, if it is located at the edge of the distribution and is weakly associated with known fault modes, the increment operator may be lower than the threshold. It is worth mentioning that the preset threshold is usually dynamically optimized based on the historical fault data validation set, for example, by determining the critical value corresponding to the balance point of the false positive rate and the false negative rate through the ROC curve, or using an adaptive mechanism to dynamically adjust the threshold boundary according to the real-time data stream characteristics. This data-driven threshold decision mechanism not only avoids the subjectivity of artificial experience rules, but also balances the processing efficiency and fault semantic fidelity by quantifying feature importance, providing high-quality feature input for subsequent fault data fold line construction and abnormality degree evaluation. In one specific example of the present application, the fault data semantic discrete increment operator of the optimized fault modulation individual feature vector relative to the set of fault data structured coding vectors is calculated as a fault semantic contribution degree by the following formula:

[0046]

[0047] wherein v′ i,k represents the feature value at the kth position in the ith modulation individual feature vector, x i,k represents the feature value at the kth position in the ith fault data structured coding vector, L represents the length of the fault data structured coding vector, λ i represents the fault data suppression factor corresponding to v i , π represents the circular constant, and arctan represents the inverse tangent function.

[0048] Subsequently, based on a comparison between the semantic discrete incremental operator of the fault data and a preset threshold, it is determined whether to delete the i th fault data structured coding vector. Wherein, the result corresponding to the deletion of the i th fault data structured coding vector indicates that the fault data is distilled. Here, considering that the power equipment fault data has high dimensionality, strong noise and complex working conditions (such as transformer partial discharge or circuit breaker contact oxidation scenes), the traditional fixed rule feature selection method is difficult to adapt to the dynamic change of data distribution, and is easy to cause the key fault features to be mistakenly deleted or the redundant features to be retained. Through the dynamic comparison of the threshold value and the semantic discrete incremental operator, the system can quantify the contribution of the features to the expression of the fault semantics, avoid the screening failure caused by local correlation deviation, and provide flexible feature retention strategies for different working conditions. It is worth mentioning that the threshold value as a dynamic decision boundary, its setting needs to be combined with the specific application scene: for example, in the fault positioning scene requiring high precision (such as key equipment monitoring in a substation), the threshold value can be set lower to retain more potential related features; while in the wide area fault early warning scene requiring higher real-time performance, the threshold value can be dynamically adjusted higher to quickly eliminate low-contribution features. In addition, the adaptive mechanism of the threshold value (such as based on the historical fault mode library or online feedback learning) can further improve the adaptability of the system to complex working conditions. For example, when a new type of fault (such as photovoltaic inverter harmonic resonance) is detected, the system can update the threshold value parameters online, preferentially retain the current phase features related to the resonance frequency, and avoid missing detection caused by the rigidity of the preset threshold value. Through the dynamic screening driven by the threshold value, the system can accurately distinguish between key fault features and redundant noise. For example, in the case of cable partial discharge detection, the optimized screening mechanism can effectively eliminate the capacitance value drift features caused by environmental temperature and humidity fluctuations (whose semantic incremental operator is lower than the threshold value), while retaining the amplitude-time joint features of the discharge pulse (whose operator value is higher than the threshold value). This feature selection mechanism based on quantitative evaluation provides scalable technical support for real-time monitoring and accurate operation and maintenance of power systems. In one specific example of the present application, based on a comparison between the semantic discrete incremental operator of the fault data and a preset threshold, it is determined whether to delete the i th fault data structured coding vector according to the following formula:

[0049]

[0050] wherein, represents v′ i The corresponding fault data semantic discrete incremental operator is θ, and the mask represents the mask function.

[0051] Specifically, the S23 determines the distillation set of the fault data based on the distillation set of the structured coding vector of the fault data. That is, in the technical solution of the present application, the distillation set of the structured coding vector of the fault data is input into the fault feature reconstruction engine to obtain the distillation set of the fault data. It can be understood that, since the power equipment fault data is subjected to structured coding and dynamic screening, its distillation set exists in the form of a high-dimensional vector (such as an abstract feature representation of transformer oil temperature anomaly or circuit breaker contact aging), which can represent the essence of the fault, but cannot be directly used for subsequent fault data fold line construction and quantitative analysis. The decoding module restores the abstract features to fault parameters (such as voltage fluctuation value, current harmonic component) aligned with the original data space through inverse mapping relationship, thereby providing structured input for the calculation of fault representation coefficients. In the decoding process, the design of the decoding module reprojects the distilled high-dimensional vector to the original fault data space through inverse mapping, thereby establishing a bridge between the data-driven model and the physical mechanism of the power equipment fault. Through the intelligent restoration of the decoding module, the system can convert the abstract fault feature vector into time series data or statistical indicators that can be directly analyzed. Taking the construction of the fault data fold line as an example, the decoded data set can generate a curve reflecting the development trend of the fault (such as the exponential decay trend of the insulation resistance over time), and then accurately calculate the fault representation coefficients through the slope change point detection. In actual operation and maintenance of the power grid, this decoding mechanism not only solves the compatibility problem of high-dimensional data and downstream analysis modules, but also suppresses the noise interference that may be introduced in the feature screening process through information fidelity technology (such as residual compensation network).

[0052] In particular, the S3 constructs a fault data polyline based on the distilled set of fault data, and calculates a fault characterization coefficient of the electric energy meter based on the fault data polyline. Here, the construction of the fault data polyline aims to convert abstract distilled features into intuitive fault timing trends. It should be understood that power equipment faults often manifest as continuous abnormal fluctuations or mutations of specific parameters, and traditional static threshold detection is difficult to capture dynamic abnormal patterns under complex working conditions. By constructing a polyline graph with time as the horizontal axis and the numerical values of the distilled feature vectors as the vertical axis, the multi-dimensional fault features can be reduced to a two-dimensional visual space, and the evolution process of the device state can be intuitively reflected. For example, the periodic spikes of fault current, the sustained drop of voltage and other implicit fault patterns are highlighted through the morphological changes of the polyline, thereby breaking through the limitations of traditional rule-based analysis of isolated data at a single moment. In the technical solution of the present application, the calculation of the fault characterization coefficient of the electric energy meter based on the fault data polyline realizes the quantitative evaluation of the fault degree. The coefficient extracts fault intensity and persistence features by analyzing the geometric properties of the polyline (such as the rate of change of the slope, the curvature extreme point, and the variance of adjacent nodes). For example, when the polyline shows a steep rising trend, the characterization coefficient calculates the severity of transient faults by combining the absolute value of the slope and the duration; if the polyline shows high-frequency oscillation, the frequency energy distribution is extracted through Fourier transform to quantify the influence of periodic interference. By mapping high-dimensional fault features into analyzable timing polylines and designing multi-dimensional fusion characterization coefficients, the problems of ambiguous feature expression and lagging abnormal pattern recognition in traditional methods under complex data scenarios are effectively solved, providing a quantifiable decision basis for the precise operation and maintenance of power equipment.

[0053] In particular, the S4 determines the abnormal degree of the electric energy meter fault based on a comparison between the fault characterization coefficient of the electric energy meter and the fault characterization coefficient threshold of the electric energy meter. Specifically, in response to the fault characterization coefficient of the electric energy meter being greater than or equal to the fault characterization coefficient threshold of the electric energy meter, it is determined that the abnormal degree of the electric energy meter fault is high; in response to the fault characterization coefficient of the electric energy meter being less than the fault characterization coefficient threshold of the electric energy meter, it is determined that the abnormal degree of the electric energy meter fault is low. This quantitative determination method provides a priority handling basis for maintenance personnel.

[0054] In summary, the power equipment fault data processing method based on power big data according to the embodiments of the present application is illustrated, which first converts the original fault data into a high-dimensional vector through structured coding, and adopts a feature selection mechanism based on deep learning to dynamically distill the high-dimensional vector, and screens out a distillation set representing the essence of the fault, so as to overcome the feature redundancy problem of traditional fixed rule screening, and further realizes accurate evaluation of the fault abnormality degree through dynamic construction of the fault data fold line and quantitative calculation of the fault representation coefficient. In this way, the problems of low processing efficiency and insufficient fault mode recognition ability under complex working conditions caused by the dependence on artificial rules in traditional methods are solved, and high-precision and scalable technical support is provided for intelligent operation and maintenance of power grid equipment.

[0055] Further, a power equipment fault data processing system based on power big data is also provided.

[0056] Figure 4 A block diagram of the power equipment fault data processing system based on power big data according to the embodiments of the present application is shown. As shown in Figure 4 The power equipment fault data processing system based on power big data 300 according to the embodiments of the present application includes: a fault data acquisition module 310, configured to acquire fault data of an electric energy meter in real time to obtain a set of fault data; a fault data screening module 320, configured to perform an unsupervised learning-based fault data screening on the set of fault data to obtain a distillation set of fault data; a fault representation coefficient calculation module 330, configured to construct a fault data fold line based on the distillation set of fault data, and calculate a fault representation coefficient of the electric energy meter based on the fault data fold line; and an abnormality degree determination module 340, configured to determine an abnormality degree of the electric energy meter fault based on a comparison between the fault representation coefficient of the electric energy meter and a fault representation coefficient threshold of the electric energy meter.

[0057] As described above, the power equipment fault data processing system based on power big data 300 according to the embodiments of the present application can be implemented in various wireless terminals, such as a server with a power equipment fault data processing algorithm based on power big data, etc. In one possible implementation, the power equipment fault data processing system based on power big data 300 according to the embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the power equipment fault data processing system based on power big data 300 can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the power equipment fault data processing system based on power big data 300 can also be one of the many hardware modules of the wireless terminal.

[0058] Alternatively, in another example, the power equipment failure data processing system 300 based on power big data and the wireless terminal can also be separate devices, and the power equipment failure data processing system 300 based on power big data can be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in a conventional data format.

[0059] Embodiments of the present disclosure have been described above, the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles, practical applications, or improvements to the technology in the market of the embodiments, or to enable other ordinary skilled in the art to understand the embodiments disclosed herein.

Claims

1.A power equipment fault data processing method based on power big data, characterized in that, The method comprises the following steps: Real-time acquisition of fault data of the electric energy meter to obtain a fault data set; Performing an unsupervised learning-based fault data filter on the fault data set to obtain a distilled fault data set, including: determining whether to distill each fault data in the fault data set based on a comparison between the fault semantic contribution degree of the fault data and a preset threshold value; Based on the distilled fault data set, a fault data fold line is constructed, and a fault characterization coefficient of the electric energy meter is calculated based on the fault data fold line; Based on a comparison between the fault characterization coefficient of the electric energy meter and a fault characterization coefficient threshold value of the electric energy meter, the abnormality degree of the electric energy meter fault is determined. 2.The power equipment failure data processing method based on power big data according to claim 1, wherein, Performing an unsupervised learning-based fault data filter on the fault data set to obtain a distilled fault data set, including: Structurally encoding each fault data in the fault data set to obtain a set of fault data structured encoding vectors; Performing fault data distillation extraction based on a fault characterization topology-driven fault data distillation extraction on the set of fault data structured encoding vectors to obtain a distilled set of fault data structured encoding vectors; Based on the distilled set of fault data structured encoding vectors, the distilled fault data set is determined. 3.The power equipment failure data processing method based on power big data according to claim 2, wherein, Structurally encoding each fault data in the fault data set to obtain a set of fault data structured encoding vectors, including: Using a fault data embedding matrix to structurally encode each fault data in the fault data set to obtain a set of fault data structured encoding vectors. 4.The power equipment failure data processing method based on power big data according to claim 3, wherein, Performing fault data distillation extraction based on a fault characterization topology-driven fault data distillation extraction on the set of fault data structured encoding vectors to obtain a distilled set of fault data structured encoding vectors, including: Calculating a fault data population distribution semantic graph of the set of fault data structured encoding vectors; Extracting the i-th fault data structured encoding vector from the set of fault data structured encoding vectors and mapping it to the fault data population distribution semantic graph to obtain a fault modulation individual feature vector; Based on the fault modulation individual feature vector and the set of fault data structured encoding vectors, determining a fault data semantic discrete incremental operator of the i-th fault data structured encoding vector as a fault semantic contribution degree; Based on a comparison between the fault data semantic discrete incremental operator and a preset threshold value, determining whether to delete the i-th fault data structured encoding vector. 5.The power equipment failure data processing method based on power big data according to claim 4, wherein, Extracting the i-th fault data structured encoding vector from the set of fault data structured encoding vectors and mapping it to the fault data population distribution semantic graph to obtain a fault modulation individual feature vector, including: Extracting the i-th fault data structured encoding vector from the set of fault data structured encoding vectors as a fault data individual feature vector; Mapping the fault data individual feature vector to a fault feature space of the fault data population distribution semantic graph to obtain a fault modulation individual feature vector. 6.The power big data-based power equipment failure data processing method of claim 5, wherein, Based on the fault modulation individual feature vector and the set of fault data structured encoding vectors, determining a fault data semantic discrete incremental operator of the i-th fault data structured encoding vector as a fault semantic contribution degree, including: Performing global universal calibration on the fault modulation individual feature vector to obtain an optimized fault modulation individual feature vector; A fault data semantic discrete increment operator is calculated for the optimized fault modulation individual feature vector relative to the set of fault data structured coding vectors. 7.The power equipment failure data processing method based on power big data according to claim 6, wherein, A global universal calibration is performed on the fault modulation individual feature vector to obtain an optimized fault modulation individual feature vector, including: A scale-conserved associated reorganization is performed on a fault data population distribution semantic graph based on the fault modulation individual feature vector and the fault data individual feature vector to obtain a fault data population distribution semantic invariable association matrix; An outlier dimension normalization correction is performed on the fault data individual feature vector based on the fault modulation individual feature vector and the fault data individual feature vector to obtain a fault tuning global outlier feature vector; An optimized expression is performed on the fault modulation individual feature vector based on the fault data population distribution semantic invariable association matrix and the fault tuning global outlier feature vector to obtain an optimized fault modulation individual feature vector. 8.The power equipment failure data processing method based on power big data according to claim 7, wherein, Based on the distilled set of fault data structured coding vectors, a distilled set of fault data is determined, including: The distilled set of fault data structured coding vectors is input into a fault feature reconstruction engine to obtain the distilled set of fault data. 9.The power big data-based power equipment failure data processing method of claim 8, wherein, Based on a comparison between the fault characterization coefficient of the electric energy meter and the fault characterization coefficient threshold value of the electric energy meter, the abnormality degree of the electric energy meter fault is determined, including: In response to the fault characterization coefficient of the electric energy meter being greater than or equal to the fault characterization coefficient threshold value of the electric energy meter, it is determined that the abnormality degree of the electric energy meter fault is high; in response to the fault characterization coefficient of the electric energy meter being less than the fault characterization coefficient threshold value of the electric energy meter, it is determined that the abnormality degree of the electric energy meter fault is low. 10.A power equipment fault data processing system based on power big data, characterized in that, Including: A fault data acquisition module is configured to acquire fault data of an electric energy meter in real time to obtain a set of fault data; A fault data screening module is configured to perform a fault data screening based on unsupervised learning on the set of fault data to obtain a distilled set of fault data; A fault characterization coefficient calculation module is configured to construct a fault data polyline based on the distilled set of fault data, and calculate a fault characterization coefficient of the electric energy meter based on the fault data polyline; An abnormality degree determination module is configured to determine the abnormality degree of the electric energy meter fault based on a comparison between the fault characterization coefficient of the electric energy meter and the fault characterization coefficient threshold value of the electric energy meter.

Images