A unified conversion and analysis method and system for fault signals of an electric energy meter sampling unit

Abstract

This invention relates to the field of fault signal processing technology for electricity meter sampling units, and discloses a unified conversion and analysis method and system for fault signals of electricity meter sampling units. The method simultaneously acquires voltage, current, and frequency signals and performs anomaly detection, generating a preliminary fault signal sequence containing various manifestations such as analog quantity anomalies, sampling deviations, and transient fluctuations. Feature indicators are extracted from the sequence, and a multi-channel unified fault feature space is constructed through fusion and mapping, representing each fault event as a feature vector in the space. Using a pre-established fault conversion model, the feature vectors are converted and mapped to a unified fault type code. The root cause type of the fault is identified and output based on the unified code, completing the unified identification of diverse fault signals. This invention realizes the mapping of multi-source heterogeneous fault signals to standardized type codes, thereby achieving normalized identification of fault root causes and effectively improving the accuracy and systematic nature of diagnosis.

Description

Technical Field

[0001] This invention relates to the field of fault signal processing technology for electricity meter sampling units, and in particular to a unified conversion and analysis method and system for fault signals of electricity meter sampling units. Background Technology

[0002] Effective and accurate fault diagnosis of electricity meter sampling units is a necessary technical step to ensure the economical, safe, and stable operation of power systems. Currently, the fault diagnosis methods commonly used in the industry mainly rely on setting static thresholds or simple rules for independent judgment of single electrical parameters or single sampling channels; for example, continuously monitoring whether the sampled values ​​of voltage or current exceed the allowable range, or judging whether the frequency measurement value deviates too much from the nominal value. These methods are effective in detecting obvious and isolated hardware faults.

[0003] However, in complex real-world operating environments, due to factors such as component performance degradation, electromagnetic interference, environmental stress changes, or software logic defects, sampling unit faults often exhibit cross-channel and multi-form correlation characteristics. A single fault root cause may trigger various forms of signal anomalies on different physical channels or at different time scales. Typically, the same hardware defect causes slow analog drift in the voltage channel, sudden jumps in the current channel, and periodic deviations in the synchronization signal in frequency measurements. This phenomenon of "single root cause, multiple manifestations" leads existing diagnostic systems based on single-point independent criteria to frequently generate discrete and seemingly contradictory alarm signals. The system may report multiple independent fault codes such as voltage over-limit, current distortion, and frequency lockout for the same underlying fault, but the inherent correlation between these codes is not effectively revealed. The direct consequence is increased complexity for maintenance personnel in analysis and diagnosis, easily leading to misjudgments—for example, misidentifying transient interference as permanent damage, or masking systemic failure risks by failing to correlate anomalies across multiple channels. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the technical defects of the prior art, such as alarm dispersion, lack of root cause correlation and redundant and contradictory diagnostic conclusions caused by the cross-channel and multi-form fault manifestations. The present invention provides a unified conversion and analysis method and system for fault signals of energy meter sampling units. By constructing a multi-channel unified feature space and fault conversion model, the present invention can map multi-source heterogeneous fault signals into standardized fault type codes, thereby achieving normalized identification of fault root causes and improving diagnostic accuracy and systematicity.

[0005] To address the aforementioned technical problems, this invention provides a unified conversion and analysis method for fault signals in an electricity meter sampling unit, comprising the following steps:

[0006] The voltage channel signal, current channel signal, and frequency signal of the energy meter sampling unit are collected synchronously.

[0007] Anomaly detection is performed on the voltage channel signal, current channel signal and frequency signal respectively, and a preliminary fault signal sequence corresponding to each channel is generated. In the preliminary fault signal sequence, the fault signal manifestations include analog quantity anomalies, sampling deviations and transient fluctuations.

[0008] From the initial fault signal sequence, feature indicators that can characterize the fault signal manifestation are extracted;

[0009] Based on feature indicators, the preliminary fault signal sequences sampled from the voltage channel, current channel, and frequency are fused and mapped to construct a multi-channel unified fault feature space, where each fault event corresponds to a feature vector located in the multi-channel unified fault feature space.

[0010] Based on a pre-established fault conversion model, feature vectors in a multi-channel unified fault feature space are converted and mapped to a unified fault type code. The fault conversion model defines the conversion rules and mapping relationships between multi-channel, multi-representation fault signal features and standardized fault type codes.

[0011] Based on a unified fault type code, the root cause type of the fault in the energy meter sampling unit is identified and output, thus completing the unified identification of fault signals with different manifestations.

[0012] In one embodiment of the present invention, anomaly detection includes the following steps:

[0013] Obtain the real-time sampled values ​​of each channel signal and the preset reference signal range;

[0014] Establish a dynamic threshold calculation mechanism based on continuous time windows;

[0015] When the sampled value exceeds the dynamic threshold range, an abnormal record containing a timestamp, channel identifier, and degree of deviation is generated;

[0016] Abnormal records are classified by form, distinguishing between three manifestations: analog quantity anomalies, sampling deviations, and transient fluctuations.

[0017] In one embodiment of the present invention, extracting feature indicators includes the following steps:

[0018] Establish a multidimensional feature extraction template, including time-domain statistical features, frequency-domain distribution features, and channel correlation features;

[0019] The initial fault signal sequence is segmented using a sliding window to obtain continuous fault segments;

[0020] Calculate the mean amplitude, variance, rate of change, and time-domain integral characteristics of each fault segment;

[0021] The dominant frequency components and their energy distribution ratios are obtained through spectrum analysis.

[0022] Calculate the synchronization correlation coefficient and phase difference characteristics between the voltage and current channels.

[0023] In one embodiment of the present invention, a multi-channel unified fault feature space is constructed, including:

[0024] The feature indicators extracted from each fault event are arranged according to their corresponding physical channels and feature categories to form an initial feature set. The initial feature set includes the time-domain and frequency-domain features of the voltage channel, the time-domain and frequency-domain features of the current channel, and the time-domain and frequency-domain features of the frequency channel.

[0025] Based on the predefined fault-channel impact relationship table, determine the priority order of fault characterization for voltage channel, current channel, and frequency channel under the fault event type;

[0026] Based on the priority order of fault representation, the time domain feature of the channel with the highest priority is selected from the initial feature set as the first dimension coordinate value, and the frequency domain feature of the channel with the highest priority is selected as the second dimension coordinate value.

[0027] Calculate the consistency coefficient of the feature change trend between the remaining channels and the highest priority channel, and use the consistency coefficient of the feature change trend as the third dimension coordinate value;

[0028] A three-dimensional coordinate point is formed by the first-dimensional coordinate value, the second-dimensional coordinate value, and the third-dimensional coordinate value, which serves as the feature vector of the fault event in the multi-channel unified fault feature space.

[0029] In one embodiment of the present invention, when contradictions arise among the features extracted from the initial feature set, the following consistency integration steps are performed:

[0030] Compare index values ​​from different channels that belong to the same feature category. If the difference in their values ​​exceeds the preset tolerance range, it is determined that there is a feature contradiction.

[0031] The channel feature verification mechanism is activated to check the integrity and noise level of the original sampling data of each channel, and the contradictory feature values ​​corresponding to the channel with the worst data quality are marked as data to be verified.

[0032] Based on the fault-channel impact relationship table and the temporal logical relationship of fault occurrence, select the feature value that best matches the physical mechanism of the fault root cause from the remaining valid feature values, and use it to calculate the coordinates of the corresponding dimension.

[0033] The data to be verified and its corresponding channel identifier are appended to the feature vector as auxiliary reference information for the diagnostic results.

[0034] In one embodiment of the present invention, a fault transfer model is established, including:

[0035] Based on the physical root cause of the fault in the sampling unit of the energy meter, a set of standardized fault types containing multiple codes is defined, with each code corresponding to a specific fault root cause;

[0036] For each standardized fault type, a standard feature vector template is established. The standard feature vector template includes the ideal value range of time domain dimension coordinates, frequency domain dimension coordinates, and correlation dimension coordinates.

[0037] In the multi-channel unified fault feature space, a feature space sub-region is defined with the standard feature vector template corresponding to each standardized fault type as the center and according to the ideal value range.

[0038] Each feature space sub-region is associated with a specific standardized fault type code, forming a one-to-one mapping relationship between the fault type code and the feature space sub-region;

[0039] Define the coordinate range of the feature space sub-region as the transformation judgment condition;

[0040] Generate a transformation rule base. Each transformation rule includes: the dimension coordinates of the input feature vector, the standardized fault type encoding for matching, and the matching confidence level.

[0041] In one embodiment of the present invention, if a feature vector falls within the reference range of multiple feature vectors simultaneously, the following steps are performed:

[0042] Obtain the coordinate values ​​representing the inter-channel correlation features in the feature vector;

[0043] Based on the predefined channel influence relationship table, determine which candidate fault type's typical path best matches the fault propagation path reflected by the coordinate values;

[0044] The code of the candidate fault type that best matches the typical path is determined as the final standardized fault type code and output.

[0045] In one embodiment of the present invention, when the feature vector to be transformed does not fall within any predetermined feature vector reference range, the following processing steps are performed:

[0046] The feature vector is identified as a new pattern vector, and a temporary fault code to be confirmed is generated.

[0047] The temporary fault code and its corresponding feature vector are submitted to the manual diagnostic port, and the original signal sequence generated is recorded.

[0048] Based on the feedback from the manual diagnostic port, if a new fault type is confirmed, a new standard code is assigned to the new fault type, and a new feature vector reference range is initialized in the multi-channel unified fault feature space with the new mode vector as the center, while the fault conversion model is updated.

[0049] To address the aforementioned technical problems, this invention also provides a unified conversion and analysis system for fault signals of an electricity meter sampling unit, used to implement the above method, comprising:

[0050] The signal acquisition module synchronously acquires the voltage channel signal, current channel signal, and frequency signal of the energy meter's sampling unit;

[0051] The anomaly detection module performs anomaly detection on the voltage channel signal, current channel signal, and frequency signal respectively, and generates a preliminary fault signal sequence corresponding to each channel. The fault signal manifestations include analog quantity anomalies, sampling deviations, and transient fluctuations.

[0052] The feature extraction module extracts feature indicators that can characterize the manifestation of the fault signal from the preliminary fault signal sequence;

[0053] The space construction module, based on feature indicators, fuses and maps the preliminary fault signal sequences of voltage channel, current channel and frequency sampling to construct a multi-channel unified fault feature space, where each fault event corresponds to a feature vector located in the multi-channel unified fault feature space.

[0054] The fault conversion and coding generation module, based on a pre-established fault conversion model, converts and maps feature vectors in a multi-channel unified fault feature space into a unified fault type code. The fault conversion model defines the conversion rules and mapping relationships between multi-channel, multi-representation fault signal features and standardized fault type codes.

[0055] The fault identification module identifies and outputs the root cause type of the fault in the energy meter sampling unit based on a unified fault type code, thus completing the unified identification of fault signals with different manifestations.

[0056] The technical solution of the present invention has the following advantages compared with the prior art:

[0057] This application synchronously acquires and performs multi-channel anomaly detection on voltage, current, and frequency signals from the energy meter sampling unit, and extracts multi-dimensional feature indicators from the preliminary fault signal sequence to construct a unified multi-channel fault feature space. Through a pre-established fault conversion model, complex fault feature vectors are mapped to a unified fault type code, thereby achieving unified identification and root cause localization of cross-channel, multi-morphological related faults in the energy meter sampling unit. This method effectively avoids the discrete alarms and false alarms generated by traditional single-point diagnostic methods, improves the accuracy and efficiency of fault diagnosis, reduces the analytical burden on maintenance personnel, and helps ensure the stable operation of the power system. Attached Figure Description

[0058] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0059] Figure 1 This is a flowchart of the steps in the unified conversion and analysis method for fault signals of the energy meter sampling unit of the present invention;

[0060] Figure 2 This is a flowchart of the anomaly detection steps of the present invention;

[0061] Figure 3 This is a flowchart of the steps for extracting feature indicators according to the present invention;

[0062] Figure 4 This is a flowchart of the steps involved in constructing a multi-channel unified fault feature space according to the present invention;

[0063] Figure 5 This is a flowchart of the steps for establishing a fault conversion model according to the present invention;

[0064] Figure 6 This is a structural framework diagram of the unified conversion and analysis system for fault signals of the energy meter sampling unit of the present invention. Detailed Implementation

[0065] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0066] Reference Figure 1 As shown, this invention proposes a unified conversion and analysis method for fault signals of energy meter sampling units, establishing a hierarchical conversion and mapping system from multi-source heterogeneous fault signals to standardized fault type codes, including the following steps:

[0067] First, raw sampling signals of voltage, current, and frequency are simultaneously acquired, and preliminary anomaly detection is performed on the signals of each channel based on preset rules, generating a preliminary fault signal sequence containing various anomaly manifestations. Second, feature indicators that can quantify fault characteristics are extracted from the fault signal sequences of each channel. Then, an algorithm is used to fuse and spatially map the heterogeneous features from different channels, constructing a multi-dimensional, multi-channel unified fault feature space. In this space, each independent fault event is represented as a feature vector with a defined dimension, thus achieving a unified mathematical description of cross-channel, multi-form fault information. Next, a pre-established fault transformation model is invoked. This model encapsulates the mapping relationship from multi-dimensional fault features to a finite number of standard fault root cause types, learned from historical fault data or defined by rules. It transforms the input feature vector according to internal mapping rules and outputs a corresponding unified fault type code, which uniquely corresponds to a specific fault root cause. Finally, based on this unified fault type code, the fault root cause type of the sampling unit is identified and output.

[0068] By implementing the above technical solution, the present invention can produce the following beneficial effects:

[0069] First, it achieves the normalization and root cause identification of fault diagnosis results. By mapping multi-channel and multi-form discrete alarm signals to a single root cause code, it fundamentally solves the problem of alarm redundancy and contradiction caused by the diversity of fault manifestations, significantly improves the clarity and accuracy of diagnostic conclusions, and reduces the burden of manual analysis and the risk of misjudgment.

[0070] Second, it enhances the comprehensive judgment ability and robustness of the diagnostic system. By constructing a unified feature space and transformation model, the system can make comprehensive inferences on complex, overlapping, or even partially missing fault features based on internal mapping relationships, thereby enhancing the ability to detect and distinguish hidden faults and complex faults that are difficult to identify by traditional methods.

[0071] In some of the embodiments described above in this application, anomaly detection is proposed for the voltage channel signal, current channel signal and frequency signal of the energy meter sampling unit to generate a preliminary fault signal sequence. However, if the anomaly detection mechanism is not refined enough or lacks dynamic adaptability, it may lead to the inability to accurately identify different types of abnormal signals.

[0072] In this regard, refer to Figure 2As shown, this application further proposes the following steps for anomaly detection: During the anomaly detection process, it is first necessary to acquire the real-time sampled values ​​and preset reference signal ranges of each channel signal. Real-time sampled values ​​refer to the instantaneous measurement data acquired from the energy meter's sampling unit at a specific moment; these data reflect the current operating state of the energy meter. The preset reference signal range refers to the expected value range that each channel signal should be within under normal operating conditions of the energy meter. This range is usually determined based on the equipment's design specifications, industry standards, or through statistical analysis of a large amount of historical normal operating data. Acquiring real-time sampled values ​​is the basis for any anomaly judgment, while the preset reference signal range provides an initial reference for determining whether a signal is abnormal.

[0073] Building upon this foundation, a dynamic threshold calculation mechanism based on continuous time windows is established. This is because the threshold used for anomaly detection is not fixed but can be adjusted in real-time based on the signal's historical behavior, environmental conditions, or specific algorithms. A continuous time window means that data over a period of time (rather than just a single sampling point) is considered when calculating the dynamic threshold, thus capturing signal trends, short-term fluctuations, and the potential impact of environmental factors on the signal. This dynamic adjustment capability allows anomaly detection to better adapt to normal signal fluctuations, avoiding false alarms or missed alarms caused by improper static threshold settings.

[0074] When the sampled value of any channel exceeds the dynamic threshold range, an anomaly record is immediately generated. This anomaly record is structured data that includes the exact timestamp of the anomaly, the specific channel identifier where the anomaly occurred, and the degree or magnitude of the sampled value deviating from the dynamic threshold range (i.e., the degree of deviation). This anomaly record provides crucial contextual information for subsequent fault analysis, ensuring the traceability and detail of the anomaly event.

[0075] Subsequently, the generated anomaly records are morphologically classified to distinguish three specific fault manifestations: analog quantity anomalies, sampling deviations, and transient fluctuations. Morphological classification categorizes anomaly records based on their characteristics (such as duration, magnitude of exceeding thresholds, and rate of change). For example, if a sampled value stably exceeds the threshold for a long period (e.g., several seconds or minutes), it may be classified as an analog quantity anomaly; if a sampled value repeatedly and slightly exceeds the threshold within a short period (e.g., several sampling cycles) with a certain regularity, it may be identified as a sampling deviation; and if a sampled value undergoes drastic changes at a single or very few sampling points and then quickly returns to normal, it is usually classified as a transient fluctuation.

[0076] Furthermore, when extracting feature indicators that characterize the fault signal manifestation from the initial fault signal sequence, if the characteristics of the fault signal in different dimensions are not captured comprehensively and precisely, the extracted feature indicators may not fully reflect the essence of the fault, thereby affecting the accuracy of subsequent fault fusion and mapping, and making the identification of the fault root cause type of the energy meter sampling unit not accurate enough.

[0077] In this regard, refer to Figure 3 As shown, this application further proposes steps for extracting feature indicators, including:

[0078] The purpose of establishing a multi-dimensional feature extraction template is to provide a systematic framework to ensure comprehensive and multi-faceted acquisition of fault signal characteristic information from raw data. This template encompasses time-domain statistical features, frequency-domain distribution features, and channel correlation features. Time-domain statistical features describe the amplitude distribution and fluctuations of the signal along the time axis, such as mean, variance, peak value, and RMS value. Frequency-domain distribution features reveal the energy distribution characteristics of the signal along the frequency axis, such as dominant frequency, bandwidth, spectral centroid, and energy ratio, reflecting the signal's periodicity, harmonic components, and frequency composition. Channel correlation features focus on describing the interrelationships and dependencies between different signal channels, such as correlation coefficient, phase difference, and mutual information, reflecting the synchronicity, coupling, and causal relationships between multi-channel signals. By pre-defining such a template, the feature extraction process can be guided, ensuring that the extracted features comprehensively characterize the fault signal from different dimensions, providing rich and discriminative information for subsequent fault analysis.

[0079] To better capture the dynamic characteristics of fault signals, this application performs sliding window segmentation on the initial fault signal sequence to obtain continuous fault segments. Fault signals are often dynamically changing. By setting a fixed-length window and a sliding step size, the continuous signal stream can be divided into a series of short-time segments with local characteristics. Each segment can be considered an independent analysis unit, facilitating independent feature analysis of each segment and capturing the transient nature and evolution of the fault. For example, the window length can be determined based on the typical duration or sampling frequency of the fault signal, and the sliding step size can be smaller than the window length to ensure overlap between segments, thereby capturing more subtle changes.

[0080] After acquiring consecutive fault segments, this application calculates the mean amplitude, variance, rate of change, and time-domain integral characteristics of each fault segment. The mean amplitude reflects the average intensity or DC component of the fault segment; the variance reflects the fluctuation degree or AC component energy of the fault segment; the rate of change describes the instantaneous change speed or trend of the fault segment; and the time-domain integral characteristics reflect the cumulative effect of the fault segment over time. These specific time-domain statistical characteristics can quantify the performance of the fault signal in the time dimension, such as signal stability, severity, and cumulative impact.

[0081] Simultaneously, this application obtains the dominant frequency components and their energy distribution ratios through spectrum analysis. Fault signals often exhibit specific characteristics in the frequency domain; for example, certain faults may lead to an enhancement of harmonic components at specific frequencies. Spectrum analysis can reveal this frequency information hidden in the time-domain signal, helping to identify the fault type. The dominant frequency component refers to the frequency element with the most concentrated energy in the signal, while the energy distribution ratio reflects the contribution of different frequency components to the total energy. This is typically achieved through digital signal processing techniques such as Fast Fourier Transform (FFT), converting the time-domain signal to the frequency domain, and then analyzing its frequency composition.

[0082] Furthermore, this application calculates the synchronization correlation coefficient and phase difference characteristics between the voltage and current channels. Faults in the energy meter sampling unit often affect both voltage and current signals simultaneously, and this effect may exhibit specific correlations between different channels. For example, certain faults may cause abnormal phase relationships between voltage and current. The synchronization correlation coefficient measures the temporal similarity between two signals, while the phase difference directly reflects their leading or lagging relationship. These characteristics are crucial for diagnosing faults involving multi-channel interactions (such as sensor wiring errors, transformer faults, etc.) and can be obtained by calculating the Pearson correlation coefficient or cross-correlation function peak values ​​of the two signals, and by comparing their phase angles at the dominant frequency using Fourier transform.

[0083] In some of the embodiments described above in this application, a method based on feature indicators is proposed to fuse and map the preliminary fault signal sequences of voltage channels, current channels, and frequency sampling to construct a multi-channel unified fault feature space. However, when fusing these feature indicators from different channels and of different types, how to effectively integrate these multi-source heterogeneous data and construct a multi-channel unified feature space that can clearly and accurately characterize the essence of the fault event, so as to avoid feature redundancy, information loss, or excessive dimensionality, is a key challenge to achieve accurate identification of subsequent fault types.

[0084] In this regard, refer to Figure 4 As shown, this application further proposes a method for constructing a multi-channel unified fault feature space, specifically including:

[0085] In the initial feature set formation step, upon detecting a fault event in the energy meter sampling unit, various feature indicators are extracted from the voltage channel signal, current channel signal, and frequency signal. These feature indicators may include, but are not limited to, time-domain statistical features (such as mean, variance, peak value, RMS value, zero-crossing rate, waveform factor, impulse factor, etc.), frequency-domain distribution features (such as dominant frequency, harmonic content, energy distribution ratio, spectral centroid, etc.), and channel correlation features (such as cross-correlation coefficient, phase difference, etc.). This step aims to systematically collect and organize all relevant features extracted from each channel for a specific fault event, forming a comprehensive, unfiltered raw feature dataset. For example, for a fault event, the initial feature set may include the mean, variance, dominant frequency, and harmonic content of the voltage channel; the mean, variance, dominant frequency, and harmonic content of the current channel; and the mean and variance of the frequency channel. This structured set lays the foundation for subsequent feature selection and space construction, ensuring that all potentially useful information is taken into consideration.

[0086] In determining the priority order of fault characterization, this application introduces a fault-channel influence relationship table, which is a pre-established knowledge base that records in detail the degree or sensitivity of different types of energy meter sampling unit faults (e.g., voltage channel open circuit, current transformer saturation, frequency source drift, etc.) to the three channels of voltage, current, and frequency. This table can be constructed through historical fault data analysis, expert experience summaries, or simulation experiments. For example, for a "voltage channel open circuit" fault, the voltage channel may have the highest priority, while the current and frequency channels have relatively lower priorities. When a fault event occurs, based on the preliminary judgment of the fault type (or through preliminary analysis of the initial feature set), this relationship table is consulted to dynamically determine which channel's features are most critical and reliable for characterizing the fault under the current fault event. This prioritization mechanism avoids treating all channel features equally, allowing the feature space to focus more on the essential manifestation of the fault.

[0087] In the step of selecting the first and second dimension coordinate values, once the highest priority channel is determined, this application selects the most representative time-domain and frequency-domain features from the characteristics of that channel. For example, if the voltage channel is determined to be the highest priority channel, its mean or variance can be used as the first dimension coordinate value, and its dominant frequency component or energy distribution ratio can be used as the second dimension coordinate value. This selection strategy ensures that the first two dimensions of the feature space can directly reflect the core time-domain and frequency-domain characteristics of the fault on the most sensitive channel, thereby providing the most direct and strongest signal for the initial differentiation of the fault.

[0088] In the step of calculating the consistency coefficient of characteristic change trends and using it as the third-dimensional coordinate value, in order to capture the correlation and propagation characteristics of faults across different channels, this application calculates the consistency coefficient of the change trends of specific characteristics (e.g., amplitude, effective value, or rate of change) between the remaining channels (i.e., channels with lower priority) and the highest-priority channel. For example, if the voltage channel is the highest-priority channel and its amplitude shows a decreasing trend, the system calculates the correlation between the amplitude change trend of the current channel and the amplitude change trend of the voltage channel. If the amplitude of the current channel also shows a similar decreasing trend, the consistency coefficient will be high, indicating that the fault may affect multiple channels. Using this consistency coefficient as the third-dimensional coordinate value can effectively characterize the cross-channel impact range and pattern of the fault, providing important auxiliary information for the refined identification of faults.

[0089] Through the above technical solution, this application can effectively integrate and reduce the dimensionality of the feature indicators of the preliminary fault signal sequence from voltage, current, and frequency channels. By introducing a fault-channel influence relationship table, the representation priority of each channel can be intelligently determined according to the type of fault event, thereby ensuring that the channel features most indicative of fault diagnosis are prioritized when constructing the feature space. Using the time-domain and frequency-domain features of the highest priority channel as the core dimension, and supplemented by the consistency coefficient of the feature change trend between the other channels and the highest priority channel as the correlation dimension, this application constructs a concise and informative three-dimensional feature vector. This construction method not only avoids the dimensionality curse and information redundancy caused by simply splicing multi-channel features, but also captures the correlation between channels, so that each fault event has a unique and physically meaningful feature vector in the unified fault feature space of multiple channels, which greatly improves the identification of fault features and the accuracy of subsequent fault type coding, thereby realizing the accurate location and unified identification of faults in the energy meter sampling unit.

[0090] However, in practical applications, due to sensor errors, signal interference, or the complexity of fault propagation paths, significant differences may arise between indicator values ​​from different channels that belong to the same feature category—a phenomenon known as feature inconsistency. If this inconsistency is not effectively addressed, it will directly affect the accuracy and reliability of the feature vectors, leading to deviations in subsequent fault identification.

[0091] In response, this application further proposes to perform the following consistency integration steps when contradictions arise among the features extracted from the initial feature set:

[0092] In constructing a unified fault feature space across multiple channels, the first step is to perform a consistency check on the features extracted from the initial feature set. This check is achieved by comparing indicator values ​​from different channels that belong to the same feature category. For example, if both the voltage and current channels provide time-domain statistical features (such as the mean amplitude) of a fault event, the mean amplitudes of these two channels will be compared. If the difference in their values ​​exceeds a preset tolerance range, it indicates a feature inconsistency. The preset tolerance range can be determined based on historical data statistical analysis, expert experience, or system design requirements, aiming to distinguish between normal fluctuations and actual inconsistencies.

[0093] Once a discrepancy is identified, a channel feature verification mechanism will be initiated. This mechanism aims to thoroughly examine the raw sampled data of each channel involved in the discrepancy. Specifically, this includes checking the integrity of the data, such as for lost data packets, data transmission errors, or sampling interruptions; simultaneously, it assesses the noise level of the raw sampled data for each channel, for example, by calculating the signal-to-noise ratio or analyzing noise components in the spectrum. Through these checks, channels with relatively poor data quality can be identified. Subsequently, the discrepancy feature value corresponding to the channel with the worst data quality is marked as data to be verified, meaning that this feature value may be unreliable and requires further processing.

[0094] After identifying the data to be verified, to ensure the accuracy of the feature vector, based on a pre-established fault-channel impact relationship table and the temporal logic of fault occurrence, a feature value that best matches the physical mechanism of the fault root cause is selected from the remaining valid feature values. This feature value is then used to calculate the coordinates of the corresponding dimension. The fault-channel impact relationship table is a knowledge base that defines how different types of faults affect each sampling channel (for example, a certain fault mainly manifests as anomalies in the current channel, while having a smaller impact on the voltage channel). The temporal logic of fault occurrence considers the order and duration of the fault signal appearing in different channels. By comprehensively utilizing this information, the feature value that best reflects the actual fault situation can be intelligently selected, thereby ensuring the physical meaning and accuracy of the feature vector.

[0095] Furthermore, this application does not simply discard the data marked as pending verification. Instead, the data to be verified and its corresponding channel identifier are appended to the feature vector. This additional information serves as supplementary reference information for the diagnostic results, providing important context in subsequent fault analysis or manual diagnosis. For example, when there is uncertainty in the diagnostic results, manual diagnostic personnel can consult this supplementary information to understand the source and handling process of feature inconsistencies, thereby making a more accurate judgment or providing a basis for optimizing the fault conversion model.

[0096] This application, based on a pre-established fault transfer model, transforms and maps feature vectors in a multi-channel unified fault feature space into a unified fault type code. In practical applications, a key issue is how to accurately define and construct the fault transfer model to ensure accurate identification of the root cause type of the energy meter sampling unit from complex, multi-faceted fault signal features. If the fault transfer model is not clearly defined or lacks sufficient robustness, it may lead to low accuracy and efficiency in fault identification.

[0097] In this regard, refer to Figure 5 As shown, this application further proposes a method for establishing a fault transition model, including:

[0098] When establishing a fault conversion model, the first step is to define a standardized set of fault types containing multiple codes, based on the physical root causes of faults in the energy meter sampling unit. Each code corresponds to a specific fault root cause. This step aims to provide a clear and standardized classification system for various faults that may occur in the energy meter sampling unit. Faults in the energy meter sampling unit may include, but are not limited to, voltage path open circuit, current path short circuit, frequency source drift, AD converter failure, and reference voltage anomaly. By deeply analyzing the physical mechanisms and manifestations of these faults, a unique, semantically meaningful code can be assigned to each fault type. For example, "U_OPEN" represents a voltage path open circuit, and "I_SHORT" represents a current path short circuit. This standardized coding facilitates the automation and unified management of subsequent fault diagnosis, avoiding confusion caused by inconsistent descriptions.

[0099] Based on this, a standard feature vector template is established for each standardized fault type. This template includes ideal value ranges for time-domain, frequency-domain, and correlation-domain coordinates. The standard feature vector template is an ideal or typical representation of each standardized fault type in a multi-channel unified fault feature space. For example, for the fault type "voltage channel open circuit," its standard feature vector template might exhibit the time-domain characteristics of the voltage channel (e.g., mean close to zero, variance extremely small), frequency-domain characteristics (e.g., no dominant frequency component), and correlation characteristics with the current channel (e.g., significantly reduced synchronization correlation coefficient). The ideal value range can be determined through extensive historical fault data analysis, expert experience, or simulation. It defines the typical distribution area of ​​this fault type in the feature space, providing a benchmark for subsequent fault matching.

[0100] Furthermore, within the multi-channel unified fault feature space, a sub-region of the feature space is delineated centered on the standard feature vector template corresponding to each standardized fault type, according to its ideal value range. This step concretizes the abstract feature template into an identifiable region in the feature space. Each sub-region of the feature space represents a specific fault type, and its boundary is determined by the ideal value range of the standard feature vector template. For example, if the ideal value range of the time domain dimension coordinates of a certain fault type is [0.1, 0.5], the frequency domain dimension coordinates are [0.8, 1.2], and the correlation dimension coordinates are [-0.2, 0.2], then these three ranges together define a cuboid region in three-dimensional space. When the feature vector of the fault event to be diagnosed falls into this sub-region, it can be preliminarily determined that it belongs to this fault type.

[0101] Simultaneously, each feature space sub-region is associated with a specific standardized fault type code, forming a one-to-one mapping between the fault type code and the feature space sub-region. This step establishes a direct association from the feature space to the fault type code. By binding each defined feature space sub-region with a unique standardized fault type code, a direct conversion between fault features and fault semantics is achieved. This one-to-one mapping is the foundation of fault identification, ensuring that when a feature vector is located in a certain sub-region, its corresponding fault type code can be obtained immediately.

[0102] Subsequently, the coordinate range of the feature space sub-region is defined as the transformation judgment condition. This step clarifies the judgment logic of fault transformation. The coordinate range of each feature space sub-region, that is, its upper and lower limits in each dimension, directly constitutes the condition for judging whether an input feature vector belongs to that fault type. For example, if the coordinate value of a certain dimension of the input feature vector falls within the corresponding dimension coordinate range of a certain sub-region, then the judgment condition for that dimension is satisfied. When all dimensions are satisfied, the feature vector is considered to fall into that sub-region.

[0103] Finally, a transformation rule base is generated. Each transformation rule includes: dimensional coordinate conditions of the input feature vector, a standardized fault type code for matching, and a matching confidence level. This step organizes the above definitions and mapping relationships into an executable set of rules. Each transformation rule is essentially a description of a sub-region of the feature space, specifying all the dimensional coordinate conditions that must be satisfied to determine whether a feature vector belongs to a certain fault type. Furthermore, the introduction of a matching confidence level quantifies the reliability of the diagnostic results. For example, if a feature vector falls entirely at the center of a sub-region, the confidence level may be high; if it falls at the edge of the sub-region, the confidence level may be low, providing a basis for subsequent decision-making or manual review.

[0104] In practical applications, due to the possibility of feature overlap or blurred boundaries between different fault types, the feature vector of a fault event may sometimes fall into multiple preset feature space sub-regions simultaneously, making it impossible for the system to clearly determine its fault type, thereby affecting the accuracy and reliability of fault identification.

[0105] Furthermore, when a feature vector falls within the reference range of multiple feature vectors simultaneously, the following steps are performed:

[0106] First, numerical values ​​specifically describing the relationships between different sampling channels (such as synchronization, phase difference, amplitude ratio, etc.) are extracted from the feature vector to be analyzed. These associated feature coordinate values ​​are usually obtained when constructing a unified fault feature space for multiple channels by calculating the synchronization correlation coefficient and phase difference characteristics between voltage and current channels, or by calculating the consistency coefficient of the feature change trends between the remaining channels and the highest priority channel. These coordinate values ​​can reflect the mutual influence and coordinated changes of signals between different physical channels when a fault occurs, providing key information for distinguishing faults with similar single-channel characteristics but different propagation paths. Simultaneously, this application pre-establishes a channel influence relationship table, which is a knowledge base that records in detail the signal propagation paths and influence patterns of different channels that various known fault types may cause in the energy meter sampling unit. For example, a specific type of fault may first affect the voltage channel, then affect the current channel through some physical coupling mechanism, and ultimately cause slight fluctuations in the frequency signal. This table defines the typical channel influence sequence, influence degree, and expected performance range of inter-channel correlation characteristics for each standardized fault type. Its establishment is usually based on a large amount of experimental data, simulation analysis, and the experience and knowledge of domain experts. Subsequently, the coordinate values ​​of the inter-channel correlation features obtained from the feature vectors are compared with the typical channel influence patterns of each candidate fault type recorded in the predefined channel influence relationship table. The comparison process can employ various matching algorithms, such as similarity calculation, distance metrics, or pattern recognition algorithms. The goal is to evaluate which candidate fault type's typical fault propagation path is most physically and logically consistent with the channel correlation features of the current fault event, i.e., which candidate fault type best explains the currently observed inter-channel correlation features. After evaluating the degree of agreement between the inter-channel correlation features and the typical propagation paths of each candidate fault type, the system selects the candidate fault type with the highest degree of agreement. The standardized fault type code corresponding to this candidate fault type will be finally determined and output as the result of this fault identification.

[0107] In actual operation, the sampling unit of the electricity meter may encounter novel fault modes that were not anticipated or trained upon, causing its feature vector to fail to fall within any predefined sub-region of the feature space. In such cases, existing fault transition models will be unable to effectively identify and classify these new fault modes, thus limiting the adaptability and robustness of the fault diagnosis system.

[0108] Furthermore, when the feature vector to be converted does not fall within any predetermined feature vector reference range, the following processing steps are performed: When the system receives a feature vector of a fault event, it compares it with all predefined feature space sub-regions in the fault conversion model. If the feature vector fails to fall within any known sub-region, it indicates that it does not conform to any known fault mode, and the feature vector is then identified as a new mode vector. To track and manage such unknown faults, the system generates a unique, unconfirmed temporary fault code for it. For example, it can be identified by a specific prefix plus a timestamp or sequence number, for subsequent manual intervention and processing. Once the new mode vector is identified and a temporary fault code is generated, the system submits the temporary fault code, the new mode vector itself, and the original voltage channel signal, current channel signal, and frequency signal sequence that caused the fault event to the manual diagnostic port. The manual diagnostic port is an interface or platform for professional technicians to perform manual analysis and judgment. It allows experts to view detailed raw data and feature information to deeply analyze the physical root cause of the fault. Recording the original signal sequence aims to provide the most comprehensive information to aid human diagnosticians in making accurate judgments and to provide a data foundation for subsequent model updates. After a detailed analysis of the submitted temporary fault codes, feature vectors, and original signal sequences, professionals at the human diagnostic interface will provide diagnostic feedback. If the human diagnostic team confirms that the new pattern vector indeed represents a previously unknown fault type with independent physical roots, the system will assign a unique, permanent standard code to this new fault type and incorporate it into the standardized fault type set. Simultaneously, to enable the fault transition model to identify and handle this new fault, the system will initialize a new feature vector reference range centered on the new pattern vector in a multi-channel unified fault feature space. This new reference range can be initially sized based on the characteristics of the new pattern vector and expert experience; for example, it can be set with a default radius or determined based on the statistical distribution of similar historical events. Finally, the fault transition model will be updated based on the newly assigned standard code and the newly initialized feature vector reference range, thereby enhancing the model's recognition capability and coverage.

[0109] To implement the above method, this application also proposes a unified conversion and analysis system for fault signals of an energy meter sampling unit, comprising: a signal acquisition module for synchronously acquiring voltage channel signals, current channel signals, and frequency signals of the energy meter sampling unit; an anomaly detection module for performing anomaly detection on the voltage channel signals, current channel signals, and frequency signals respectively, generating preliminary fault signal sequences corresponding to each channel, wherein the fault signal manifestations include analog quantity anomalies, sampling deviations, and transient fluctuations; a feature extraction module for extracting feature indicators that characterize the fault signal manifestations from the preliminary fault signal sequences; and a spatial construction module for, based on the feature indicators, converting the preliminary fault signals of the voltage channel, current channel, and frequency sampling units into a unified fault signal sequence. The fault signal sequence is fused and mapped to construct a multi-channel unified fault feature space, where each fault event corresponds to a feature vector located in the multi-channel unified fault feature space. The fault conversion and coding generation module, based on a pre-established fault conversion model, converts and maps the feature vectors in the multi-channel unified fault feature space into a unified fault type code. The fault conversion model defines the conversion rules and mapping relationships between multi-channel, multi-presentation fault signal features and standardized fault type codes. The fault identification module, based on the unified fault type code, identifies and outputs the fault root cause type of the energy meter sampling unit, completing the unified identification of fault signals with different presentation forms.

[0110] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A unified conversion and analysis method for fault signals of an electricity meter sampling unit, characterized in that, Includes the following steps: The voltage channel signal, current channel signal, and frequency signal of the energy meter sampling unit are collected synchronously. Anomaly detection is performed on the voltage channel signal, current channel signal and frequency signal respectively, and a preliminary fault signal sequence corresponding to each channel is generated. In the preliminary fault signal sequence, the fault signal manifestations include analog quantity anomalies, sampling deviations and transient fluctuations. From the initial fault signal sequence, feature indicators that can characterize the fault signal manifestation are extracted; Based on feature indicators, the preliminary fault signal sequences sampled from voltage channels, current channels, and frequency channels are fused and mapped to construct a unified multi-channel fault feature space. Each fault event corresponds to a feature vector within this space. This includes: arranging the feature indicators extracted from each fault event according to their corresponding physical channel and feature category to form an initial feature set. This initial feature set contains the time-domain and frequency-domain features of the voltage channel, the current channel, and the frequency channel; and determining the voltage channel, current channel, and frequency channel features under the fault event type based on a predefined fault-channel influence table. The fault characterization priority order of the flow channel and frequency channel is determined. Based on the fault characterization priority order, the time domain feature of the highest priority channel is selected from the initial feature set as the first dimension coordinate value, i.e., the time domain feature coordinate, and the frequency domain feature of the highest priority channel is selected as the second dimension coordinate value, i.e., the frequency domain feature coordinate. The consistency coefficient of the feature change trend between the remaining channels and the highest priority channel is calculated, and the consistency coefficient of the feature change trend is used as the third dimension coordinate value, i.e., the correlation dimension coordinate. The first dimension coordinate value, the second dimension coordinate value, and the third dimension coordinate value constitute a three-dimensional coordinate point, which serves as the feature vector of the fault event in the multi-channel unified fault feature space. Based on a pre-established fault conversion model, feature vectors in a multi-channel unified fault feature space are transformed and mapped to a unified fault type code. The fault conversion model defines the conversion rules and mapping relationships between multi-channel, multi-representation fault signal features and standardized fault type codes. Specifically, establishing the fault conversion model includes: defining a set of standardized fault types containing multiple codes based on the physical root cause of the fault in the energy meter sampling unit, with each code corresponding to a specific fault root cause; and establishing a standard feature vector template for each standardized fault type, which includes time-domain coordinates, frequency... Ideal value ranges for domain dimension coordinates and associated dimension coordinates; In a multi-channel unified fault feature space, a feature space sub-region is defined centered on the standard feature vector template corresponding to each standardized fault type, according to the ideal value range; Each feature space sub-region is associated with a specific standardized fault type code, forming a one-to-one mapping relationship between the fault type code and the feature space sub-region; The coordinate range of the feature space sub-region is defined as the transformation judgment condition; A transformation rule base is generated, and each transformation rule includes: the dimension coordinate condition of the input feature vector, the matched standardized fault type code, and the matching confidence level; Based on a unified fault type code, the root cause type of the fault in the energy meter sampling unit is identified and output, thus completing the unified identification of fault signals with different manifestations.

2. The unified conversion and analysis method for fault signals of energy meter sampling units according to claim 1, characterized in that: Anomaly detection includes the following steps: Obtain the real-time sampled values ​​of each channel signal and the preset reference signal range; Establish a dynamic threshold calculation mechanism based on continuous time windows; When the sampled value exceeds the dynamic threshold range, an abnormal record containing a timestamp, channel identifier, and degree of deviation is generated; Abnormal records are classified by form, distinguishing between three manifestations: analog quantity anomalies, sampling deviations, and transient fluctuations.

3. The unified conversion and analysis method for fault signals of energy meter sampling units according to claim 1, characterized in that: Extracting feature indicators includes the following steps: Establish a multidimensional feature extraction template, including time-domain statistical features, frequency-domain distribution features, and channel correlation features; The initial fault signal sequence is segmented using a sliding window to obtain continuous fault segments; Calculate the mean amplitude, variance, rate of change, and time-domain integral characteristics of each fault segment; The dominant frequency components and their energy distribution ratios are obtained through spectrum analysis. Calculate the synchronization correlation coefficient and phase difference characteristics between the voltage and current channels.

4. The unified conversion and analysis method for fault signals of energy meter sampling units according to claim 1, characterized in that: When inconsistencies arise among the features extracted from the initial feature set, the following consistency integration steps are performed: Compare index values ​​from different channels that belong to the same feature category. If the difference in their values ​​exceeds the preset tolerance range, it is determined that there is a feature contradiction. The channel feature verification mechanism is activated to check the integrity and noise level of the original sampling data of each channel, and the contradictory feature values ​​corresponding to the channel with the worst data quality are marked as data to be verified. Based on the fault-channel impact relationship table and the temporal logical relationship of fault occurrence, select the feature value that best matches the physical mechanism of the fault root cause from the remaining valid feature values, and use it to calculate the coordinates of the corresponding dimension. The data to be verified and its corresponding channel identifier are appended to the feature vector as auxiliary reference information for the diagnostic results.

5. The unified conversion and analysis method for fault signals of energy meter sampling units according to claim 1, characterized in that: If an eigenvector falls within the reference range of multiple eigenvectors, then the following steps are performed: Obtain the coordinate values ​​representing the inter-channel correlation features in the feature vector; Based on the predefined channel influence relationship table, determine which candidate fault type's typical path best matches the fault propagation path reflected by the coordinate values; The code of the candidate fault type that best matches the typical path is determined as the final standardized fault type code and output.

6. The unified conversion and analysis method for fault signals of energy meter sampling units according to claim 1, characterized in that: When the feature vector to be transformed does not fall within any predetermined feature vector reference range, the following processing steps are performed: The feature vector is identified as a new pattern vector, and a temporary fault code to be confirmed is generated. The temporary fault code and its corresponding feature vector are submitted to the manual diagnostic port, and the original signal sequence generated is recorded. Based on the feedback from the manual diagnostic port, if a new fault type is confirmed, a new standard code is assigned to the new fault type, and a new feature vector reference range is initialized in the multi-channel unified fault feature space with the new mode vector as the center, while the fault conversion model is updated.

7. A unified conversion and analysis system for fault signals of an electricity meter sampling unit, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The signal acquisition module synchronously acquires the voltage channel signal, current channel signal, and frequency signal of the energy meter's sampling unit; The anomaly detection module performs anomaly detection on the voltage channel signal, current channel signal, and frequency signal respectively, and generates a preliminary fault signal sequence corresponding to each channel. The fault signal manifestations include analog quantity anomalies, sampling deviations, and transient fluctuations. The feature extraction module extracts feature indicators that can characterize the manifestation of the fault signal from the preliminary fault signal sequence; The space construction module, based on feature indicators, fuses and maps the preliminary fault signal sequences of voltage channel, current channel and frequency sampling to construct a multi-channel unified fault feature space, where each fault event corresponds to a feature vector located in the multi-channel unified fault feature space. The fault conversion and coding generation module, based on a pre-established fault conversion model, converts and maps feature vectors in a multi-channel unified fault feature space into a unified fault type code. The fault conversion model defines the conversion rules and mapping relationships between multi-channel, multi-representation fault signal features and standardized fault type codes. The fault identification module identifies and outputs the root cause type of the fault in the energy meter sampling unit based on a unified fault type code, thus completing the unified identification of fault signals with different manifestations.

Images