An abnormal waveform matching method for identifying abnormal voltage in electric energy meters

Abstract

The present invention discloses an abnormal waveform matching method for identifying abnormal voltage of an electric energy meter, which relates to the field of waveform matching technology. The method comprises the following steps: collecting historical abnormal waveforms and extracting feature vectors, and constructing an abnormal waveform dictionary by using a dynamic dictionary learning algorithm; performing coverage analysis on the historical abnormal waveforms, screening core features and establishing a simplified feature set; collecting the voltage waveform to be identified in real time, extracting feature vectors and matching them with the simplified feature set, and initially judging whether there is an explicit abnormality; if so, screening candidate template types based on feature deviation ratios, and determining the abnormality type; if there is no explicit abnormality, constructing a normal waveform feature baseline model, performing deviation analysis to determine whether there is a latent abnormality, and if so, iteratively optimizing the abnormal waveform dictionary in combination with the waveform features to be identified; the present invention can effectively identify the voltage abnormality of the electric energy meter, balance recognition accuracy and computational efficiency, and adapt to new abnormal modes and gradual fault characteristics of equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of waveform matching, and in particular to an abnormal waveform matching method for identifying abnormal voltage of an electric energy meter. Background Art

[0002] In the smart grid sector, accurate identification of voltage anomalies by electricity meters is crucial for ensuring stable operation of distribution networks. Prior art methods for identifying voltage anomalies by electricity meters have the following challenges that need to be addressed.

[0003] Traditional methods mostly make anomaly judgments based on a single type of feature (such as only the frequency domain harmonic amplitude or time domain waveform amplitude), which makes it difficult to characterize the characteristics of complex voltage anomalies, such as the time domain morphology of transient voltage surges or dips, the frequency domain distribution of harmonic distortion, and the coupled characteristics of time-frequency domain energy changes. As a result, existing technologies have a high rate of missed judgments for complex or atypical anomalies.

[0004] Existing anomaly recognition relies on fixed thresholds or static templates, and cannot adapt to the waveform abnormality patterns caused by high-frequency oscillations and non-integer harmonic new voltages introduced by new equipment in the distribution network (such as new energy inverters and energy storage devices). It lacks the ability to autonomously learn and iterate dictionaries for unknown anomalies, resulting in a lag in the recognition of new anomalies.

[0005] To this end, the present invention provides an abnormal waveform matching method for identifying abnormal voltage of an electric energy meter. Summary of the Invention

[0006] The object of the present invention is to provide an abnormal waveform matching method for identifying abnormal voltage in an electric energy meter, so as to solve at least one of the above-mentioned problems in the prior art.

[0007] An abnormal waveform matching method for identifying abnormal voltage of an electric energy meter comprises the following steps:

[0008] Collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms and build an abnormal waveform dictionary;

[0009] Perform coverage analysis on historical abnormal waveforms to obtain feature priority values ​​and filter them to obtain a streamlined feature set;

[0010] The voltage waveform to be identified of the electric energy meter is collected in real time, and features are extracted from the voltage waveform to be identified to construct a waveform feature vector. The waveform feature vector is matched with a simplified feature set to initially determine whether there is an obvious anomaly in the voltage waveform to be identified.

[0011] If there is an obvious abnormality, the abnormality type is screened to obtain a candidate template type, and an abnormality determination analysis is performed on the candidate template waveform prototype and the voltage waveform to be identified to obtain the abnormality determination coefficient and finally determine the abnormality type of the electric energy meter;

[0012] If it is initially determined that the voltage waveform to be identified does not have an explicit abnormality, it is determined whether the voltage waveform to be identified has a hidden abnormality. If so, the abnormal waveform dictionary is iteratively optimized in combination with the characteristics of the voltage waveform to be identified.

[0013] As a further technical solution of the present invention, the abnormal waveform dictionary is constructed as follows:

[0014] Collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms, and obtain identification feature vectors containing frequency domain features, time-frequency domain features, and morphological features;

[0015] Based on the dynamic dictionary learning algorithm, an abnormal waveform dictionary containing identification feature vectors, abnormal types, and waveform prototypes is constructed.

[0016] As a further technical solution of the present invention, the method of obtaining the simplified feature set is:

[0017] The electric energy meter to be identified is used as a sample for abnormal identification, and historical abnormal waveforms are obtained from the abnormal waveform dictionary. The historical abnormal waveforms are classified and matched according to the samples;

[0018] Obtain the feature coverage and feature separability of each sample and sum them up to obtain the feature priority value;

[0019] Sub-features are sorted based on their priority values, core features are extracted from the abnormal waveform dictionary, and a reduced feature set is constructed based on the feature vectors corresponding to the core features.

[0020] As a further technical solution of the present invention, the method of obtaining the feature coverage and feature separability is:

[0021] Calculate the frequency of occurrence of different sub-features of each sample to obtain feature coverage;

[0022] Obtain the inter-class dispersion and intra-class dispersion of the samples, and calculate the ratio to obtain the feature separability.

[0023] As a further technical solution of the present invention, the method of initially judging whether there is an obvious abnormality in the voltage waveform to be identified is:

[0024] Perform waveform preprocessing on the voltage waveform to be identified that is collected in real time, perform feature extraction on the waveform after waveform preprocessing, and construct a waveform feature vector;

[0025] Extracting each abnormality type identification feature vector from the simplified feature set, calculating the matching degree between the identification feature vector and the waveform feature vector of each abnormality type, and obtaining the feature matching degree;

[0026] The characteristic matching degree is compared with a preset characteristic matching degree threshold value to obtain a result of judging whether there is a dominant abnormality in the voltage waveform to be identified.

[0027] As a further technical solution of the present invention, the matching degree calculation method is as follows:

[0028] The feature matching degree is obtained by calculating the cosine similarity between the recognition feature vector and the waveform feature vector.

[0029] As a further technical solution of the present invention, the method for finally determining the type of abnormality existing in the electric energy meter is:

[0030] If there is a corresponding abnormal type, calculate the deviation ratio between the feature matching degree of each abnormal type and the feature matching degree threshold to obtain the feature deviation ratio;

[0031] Screen the abnormal types based on the feature deviation ratio to obtain candidate template types;

[0032] Obtain the timing matching degree and feature matching degree of the candidate template type, perform weighted sum processing on the timing matching degree and feature matching degree to obtain the abnormality determination coefficient;

[0033] Based on the abnormality determination coefficient, the abnormality type of the candidate template type is obtained, and finally it is determined whether the voltage waveform to be identified has an obvious abnormality of the abnormal type.

[0034] As a further technical solution of the present invention, the method of obtaining the timing matching degree is:

[0035] The waveform prototype of the candidate template type and the voltage waveform to be identified collected by the electric energy meter are extracted from the abnormal waveform dictionary. The waveform prototype and the voltage waveform to be identified are dynamically time-warped to obtain the timing matching degree.

[0036] As a further technical solution of the present invention, the weight coefficient of the abnormal determination coefficient is obtained as follows:

[0037] Transform the waveform prototype and the voltage waveform to be identified to generate a time-frequency spectrum and extract the transient energy distribution entropy;

[0038] The absolute deviation ratio between the waveform prototype and the transient energy distribution entropy of the voltage waveform to be identified is calculated as the weight coefficient.

[0039] As a further technical solution of the present invention, the method of judging whether the voltage waveform to be identified has a hidden abnormality is:

[0040] Construct a baseline model of normal waveform characteristics;

[0041] Perform multi-dimensional feature deviation calculation to mark potential abnormal features;

[0042] Conduct dynamic trend analysis on potential abnormal features to determine whether there are trend abnormalities;

[0043] If there is a trend anomaly, the latent anomaly is determined by combining the feature separability and standardized deviation.

[0044] Beneficial effects of the present invention:

[0045] 1. Through multi-scale variational mode decomposition, morphological filtering and S transform, the frequency domain, time-frequency domain and morphological structure identification feature vectors are extracted from historical abnormal waveforms. Combined with the dynamic dictionary learning algorithm cluster analysis, an abnormal waveform dictionary containing feature vectors, waveform prototypes and abnormality types is constructed to achieve standardized modeling of multiple types of abnormalities such as overvoltage, undervoltage, harmonic distortion, voltage sag, etc., covering common and transient voltage abnormality scenarios.

[0046] 2. By calculating feature coverage (frequency of occurrence in fault samples) and feature separability (inter-class difference and internal fluctuation ratio), the feature priority value is obtained through normalization and summation. High-frequency and easily distinguishable core features are screened to construct a streamlined feature set. The core features are calculated on the electricity meter. This helps improve the anomaly recognition efficiency of the electricity meter, realize the lightweight deployment of complex feature vectors, and adapt to scenarios with limited edge device resources.

[0047] 3. Calculate cosine similarity through a streamlined feature set for coarse screening, call the complete dictionary for dynamic time warping (DTW) and time-frequency feature difference constraints on samples with insufficient matching, and dynamically adjust the weight according to the anomaly type, which is conducive to improving the accuracy of anomaly recognition; if there is no explicit anomaly, build a normal waveform feature baseline model to identify early anomalies with continuous feature value offsets, and achieve the expansion from threshold-type explicit detection to trend-type implicit warning. For detected implicit anomalies or unknown anomalies, compare them with existing templates through density clustering algorithm. If it is a known abnormal variation, update the template parameters; if it is a new abnormality, add a new template entry; form a closed loop of data collection-anomaly detection-dictionary update, and improve the adaptability of the electricity meter to dynamic changes in abnormal waveforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 This is a flow chart of an abnormal waveform matching method for identifying abnormal voltage in an electric energy meter provided by the present invention;

[0050] Figure 2It is a flowchart of implicit recognition provided by the present invention;

[0051] Figure 3 The present invention provides a module diagram of an abnormal waveform matching method for identifying abnormal voltage in an electric energy meter. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0053] Example 1: Figure 1 As shown, an abnormal waveform matching method for identifying abnormal voltage in an electric energy meter provided by an embodiment of the present invention specifically includes the following steps:

[0054] Step 1: Collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms and construct an abnormal waveform dictionary;

[0055] The method of decomposing the features of historical abnormal waveforms and constructing an abnormal waveform dictionary is as follows:

[0056] Preferably, the voltage anomaly waveform is read from the voltmeter log, and multi-scale variational modal decomposition is performed on the original waveform to decompose the complex voltage signal into multiple intrinsic mode functions. The eigenmode components of the transient voltage surge and sag abnormal waveforms are extracted, and morphological filtering is used to enhance the edge features of the waveform (such as repairing dents and sharpening mutation points).

[0057] The waveform after morphological filtering is transformed into a time-frequency spectrum through S-transformation, and the identification feature vector containing frequency domain features, time-frequency domain features, and morphological features is extracted:

[0058] Those skilled in the art will appreciate that, morphological filtering is first performed on the voltage waveform of the electric energy meter, and by selecting appropriate structural elements (such as rectangles, Gaussian kernels, etc.) for opening and closing operations, noise is filtered out and waveform contour features are retained; then, an S-transform is performed on the filtered waveform, and a Gaussian window function that is adaptively adjusted with frequency is constructed to convert the time domain signal into a time-frequency spectrum to characterize the energy distribution of the signal at different time-frequency points; finally, frequency domain features (such as the amplitude, phase, and energy proportion of each frequency component) and time-frequency domain features (such as time-frequency energy entropy, time-frequency distribution variance, and rate of change of frequency over time) are extracted from the time-frequency spectrum, and morphological features (such as the peak-to-trough spacing, the rising or falling edge slope, and the number of zero crossings) are extracted from the waveform after morphological filtering, and the three types of features are spliced ​​in a preset order to form an identification feature vector containing multi-dimensional information;

[0059] It needs to be explained that frequency domain features, time-frequency domain features, and morphological features include multiple sub-features;

[0060] The frequency domain features are composed of three sub-features: fundamental amplitude, harmonic amplitude ratio, and total harmonic distortion;

[0061] The time-frequency domain features are composed of two sub-features: transient energy distribution entropy and instantaneous frequency fluctuation coefficient;

[0062] The morphological features are composed of three sub-features: peak slope, phase jump point position, and offset;

[0063] Using a dynamic dictionary learning algorithm, we perform cluster analysis on the identification feature vectors of similar anomalies, select the cluster centers as standard feature templates, and simultaneously retain the time series data containing the complete voltage waveform, the anomaly type, and the time domain curve of the corresponding prototype. We then construct an abnormal waveform dictionary containing the identification feature vectors corresponding to the anomaly type, the anomaly type, and the waveform prototype.

[0064] Those skilled in the art will appreciate that when constructing an abnormal waveform dictionary using a dynamic dictionary learning algorithm, a dynamic clustering method (such as the K-means++ algorithm) is first used to perform cluster analysis on the identification feature vectors of similar abnormalities. Sample allocation and center update are achieved by iteratively calculating the distance between the feature vectors and the cluster centers. The number of clusters is dynamically adjusted based on the clustering quality to reduce local optima. Stable cluster centers are then selected from the clustering results as standard feature templates. Outlier detection is performed to ensure template robustness, and a periodic or threshold-triggered template update mechanism is established.

[0065] The selected standard feature templates are associated with the corresponding complete voltage waveform time series data, abnormality type labels and time domain curves to build an abnormal waveform dictionary containing multi-dimensional information such as identification feature vectors, abnormality types and waveform prototypes;

[0066] The purpose of constructing the abnormal waveform dictionary is:

[0067] Function 1: Providing a standardized feature template library: By dynamically clustering and selecting each cluster center as a standard feature template, the identification feature vectors of similar anomalies are abstracted and standardized, providing a unified feature matching benchmark for the subsequent initial judgment of explicit anomalies and final judgment of anomaly types in real-time waveforms, ensuring the comparability of anomaly features in different time periods and scenarios;

[0068] Function 2: Support dynamic learning and iterative optimization: Synchronously store the complete waveform time series data and time domain curves corresponding to the anomaly type, retaining not only the abstract information of the feature vector but also the detailed features of the original waveform. This provides a data basis for the dictionary iteration module when detecting hidden anomalies. By dynamically updating the dictionary based on new anomaly features, continuous learning of new anomaly patterns and expansion of the knowledge base are achieved.

[0069] Function 3: Realize multi-dimensional anomaly information association mapping: Structured association storage of identification feature vectors, anomaly types, and waveform prototypes to build a "feature-type-waveform" mapping relationship. This not only supports fast matching calculations based on feature vectors (such as feature vector matching in explicit initial judgments), but also enables backtracking to specific waveform prototypes for detailed comparisons when making final judgments on anomalies (such as anomaly decision analysis between candidate templates and waveforms to be identified). This provides cross-dimensional data support for the multi-level anomaly identification process, improving identification accuracy and decision reliability.

[0070] Step 2: Perform coverage analysis on historical abnormal waveforms to obtain feature priority values ​​and filter the core features of the abnormal waveform dictionary to establish a streamlined feature set;

[0071] Among them, the method of performing coverage analysis on historical abnormal waveforms to obtain feature priority values ​​is as follows:

[0072] The electric energy meter to be identified is used as a sample for abnormal identification, and historical abnormal waveforms are obtained from the abnormal waveform dictionary. The historical abnormal waveforms are classified and matched according to the samples;

[0073] Calculate the frequency of occurrence of different sub-features of each sample to obtain feature coverage;

[0074] By formula: Get feature separability K f ;

[0075] Where c represents the total number of abnormal types, that is, the number of fault categories covered in the abnormal waveform dictionary, such as overvoltage, undervoltage, harmonic distortion, and voltage sag;

[0076] n i Indicates the number of samples of the i-th abnormal type. For example, the i-th type "voltage sag" contains n iHistorical abnormal waveform samples;

[0077] Represents the mean of the i-th sample on the f-th sub-feature, for example, the mean of the third harmonic amplitude ratio of the i-th sample, where f is the sub-feature number and is used to traverse all sub-features in the frequency domain, time-frequency domain, and morphology. For example, f=1 corresponds to the fundamental amplitude, and f=2 corresponds to the harmonic amplitude ratio.

[0078] Represents the global mean of all samples on the f-th sub-feature, that is, the f-th seed feature mean of all historical samples, which is used to measure the benchmark of inter-class differences;

[0079] Represents the specific value of a single sample on the fth sub-feature, such as the measured value of the peak slope of a fault waveform, belonging to the i-th sample set c i Elements in

[0080] c i Represents the set of class i abnormal samples, that is, the subset of all historical waveform data marked as class i abnormal, and is expressed as To calculate the intra-class dispersion;

[0081] What needs to be explained is that 、 Represent the inter-class dispersion and intra-class dispersion of samples respectively;

[0082] Inter-class dispersion reflects the mean difference of different abnormal categories on the f-th seed feature. The larger the difference, the more effective the sub-feature is in distinguishing categories.

[0083] Intra-class dispersion reflects the degree of fluctuation of samples within the same class on the fth sub-feature (summing the deviations of samples within all classes. The smaller the fluctuation, the higher the consistency of the sub-feature within the class.

[0084] Feature separability K f The larger the value is, the higher the separability of the f-th sub-feature is (significant differences between classes and small differences within classes), and the greater its contribution to distinguishing fault types in anomaly recognition.

[0085] Among them, the method of screening the core features of the abnormal waveform dictionary is:

[0086] Obtain the feature coverage and feature separability of all samples and perform Min-Max normalization on them respectively;

[0087] The normalized feature coverage and feature separability are summed to obtain the feature priority value;

[0088] It can be understood that the physical meaning of the feature priority value is: the feature priority value is the result of the normalized sum of the feature coverage and the feature separability;

[0089] The feature priority value reflects the general frequency of occurrence in the sample, reflecting universality, and the feature separability reflects the ability to distinguish fault categories, reflecting discrimination;

[0090] The feature priority value quantifies the comprehensive value of the feature in fault identification. It reflects both the feature's extensive coverage of common faults (e.g., THD frequently appears in harmonic and overvoltage scenarios) and its ability to distinguish between different fault types (e.g., the mean value of THD in harmonic faults is significantly different from that of other faults).

[0091] By using the feature priority value, you can prioritize the core features that appear frequently and are easy to distinguish (such as frequency domain THD and morphological peak slope);

[0092] Feature priority guides model lightweighting, dynamic dictionary optimization, and matching algorithm tuning, ultimately achieving data-driven, efficient mapping from historical abnormal waveform data to real-time anomaly recognition. This balances recognition accuracy and computational efficiency in resource-constrained edge device scenarios, providing a feature screening and application basis with both universality and discriminative power for electricity meter voltage anomaly detection.

[0093] Sort the sub-features based on their priority values, extract the TOP-F features from the abnormal waveform dictionary as the core features, and extract the feature vectors of the core features to construct a streamlined feature set;

[0094] Those skilled in the art will appreciate that the implementation process for constructing a streamlined feature set based on feature priority values ​​is as follows: first, the feature priority value of each sub-feature in the abnormal waveform dictionary is calculated; then, the sub-features are sorted from high to low according to the feature priority value, and the top F (TOP-F) sub-features are extracted as core features; finally, the feature vectors corresponding to these core features are extracted from all abnormal types in the abnormal waveform dictionary, and these vectors are concatenated in a preset dimensional order to form the final streamlined feature set;

[0095] The purpose of constructing a simplified feature set is:

[0096] Function 1: Improve real-time matching efficiency: By filtering the TOP-F core features through feature priority values ​​and eliminating redundant or low-discrimination sub-features, it helps reduce the dimension and computational complexity of the feature vector. This enables the explicit initial judgment module to quickly complete the calculation and comparison of the waveform feature vector and the streamlined feature set during real-time matching, meeting the real-time requirements of electricity meter anomaly identification.

[0097] Function 2: Strengthen the ability to identify key features: Based on the ability of features to distinguish anomaly types (such as information gain, mutual information, and other indicators), ensure that only the most representative core features are retained in the streamlined feature set, reduce noise feature interference, improve the accuracy of the initial judgment of obvious anomalies, and reduce misjudgments or missed judgments;

[0098] Function 3: Supporting hierarchical processing of the hierarchical identification process: The streamlined feature set serves as a rapid screening layer for explicit initial judgments, forming a layered architecture with the deep matching layer of the abnormal final judgment module. The initial judgment quickly locates potential abnormalities through core features, and the final judgment identifies the abnormality type through the detailed features of the complete candidate template, achieving optimized resource allocation, and improving the overall processing efficiency of the system while ensuring recognition accuracy.

[0099] The technical solution of this embodiment is: collecting historical abnormal waveforms of the electricity meter, performing feature decomposition on the historical abnormal waveforms and constructing an abnormal waveform dictionary; performing coverage analysis on the historical abnormal waveforms, obtaining feature priority values ​​and filtering and processing them to obtain a streamlined feature set; this is conducive to improving the abnormality identification efficiency of the electricity meter, realizing lightweight deployment of complex feature vectors, and adapting to scenarios where edge device resources are limited.

[0100] Example 2: Figure 1 As shown, an abnormal waveform matching method for identifying abnormal voltage of an electric energy meter also includes the following steps:

[0101] Step 3: Real-time acquisition of the voltage waveform to be identified from the electric energy meter, feature extraction of the voltage waveform to be identified to construct a waveform feature vector, matching calculation of the waveform feature vector with the reduced feature set, and initial determination of whether the voltage waveform to be identified has obvious anomalies;

[0102] The method of extracting the features of the voltage waveform to be identified and constructing the waveform feature vector is as follows:

[0103] Perform waveform preprocessing on the voltage waveform to be identified that is collected in real time, perform feature extraction on the waveform after waveform preprocessing, and construct a waveform feature vector;

[0104] It should be explained that the implementation process of preprocessing and feature extraction of the real-time collected voltage waveform is as follows: first, the original waveform is subjected to morphological filtering) and Z-score normalization to eliminate noise interference and dimensionality effects; feature extraction is performed on the preprocessed waveform, including time domain features (such as mean, variance, and number of zero crossings), frequency domain features (S transform to obtain the amplitude and phase of each frequency component), and time-frequency domain features (such as time-frequency energy entropy and frequency change rate); finally, the extracted multi-dimensional features are spliced ​​into a fixed-dimensional feature vector in a preset order, and dimensionality reduction optimization (such as principal component analysis) is performed to construct the final waveform feature vector. The waveform feature vector retains the key features of the waveform while reducing redundant information, providing standard input for subsequent matching calculations;

[0105] Extracting each abnormality type identification feature vector from the simplified feature set, calculating the matching degree between the identification feature vector and the waveform feature vector of each abnormality type, and obtaining the feature matching degree;

[0106] It will be understood by those skilled in the art that the feature matching degree is obtained by calculating the cosine similarity between the recognition feature vector and the waveform feature vector;

[0107] Comparing the characteristic matching degree with a preset characteristic matching degree threshold, if the characteristic matching degree is higher than or equal to the preset characteristic matching degree threshold, it is initially determined that there is an obvious abnormality in the voltage waveform to be identified;

[0108] If the feature matching degree is lower than the preset feature matching degree threshold, the identification feature vectors of all abnormal types are extracted from the abnormal waveform dictionary, the feature matching degrees of all abnormal types are calculated, and compared with the feature matching degree threshold to initially determine whether the voltage waveform to be identified has an obvious abnormality;

[0109] It can be understood that when the feature matching degree is higher than the threshold, it means that the core features of the waveform to be identified are highly consistent with the simplified feature set, and it is determined to be an obvious anomaly and quickly enters the final judgment process, reducing the need for full feature matching for all anomaly types, reducing real-time computing overhead, and meeting the real-time requirements of electricity meter anomaly identification;

[0110] When the matching degree is lower than the threshold, there may be two situations: first, the simplified feature set (TOP-F core features) fails to cover the key abnormal features of the waveform (such as insufficient discrimination of the core features or slight differences in the abnormal features); second, the waveform itself may be a marginal abnormality or a new abnormality;

[0111] By extracting the complete feature vectors of all abnormal types in the abnormal waveform dictionary and recalculating the matching degree, it can reduce missed detections caused by core feature screening, effectively identify "atypical obvious abnormalities", and significantly improve the recall rate of abnormality detection at the expense of a small amount of computing resources, achieving the complementary effect of "rapid initial screening" and "comprehensive verification".

[0112] If there is an explicit exception, get the corresponding exception type;

[0113] If the feature matching degree of all abnormal types is still lower than the preset feature matching degree threshold, the voltage waveform of the electric energy meter is continuously monitored;

[0114] It should be explained that the purpose of the initial determination of the presence of an apparent abnormality in the voltage waveform to be identified is:

[0115] Function 1: Rapid pre-screening of anomalies: By efficiently matching and calculating the streamlined feature set with the waveform feature vector, abnormal waveforms with obvious feature matching can be quickly located, reducing the need for full-feature in-depth analysis of all waveforms, improving the real-time performance of anomaly identification, and meeting the needs of real-time monitoring of the operating status of electricity meters;

[0116] Function 2: Providing a candidate range for accurate judgment of abnormality types: The explicit initial judgment results serve as the "candidate abnormality set" for preliminary screening, providing clear candidate template types (such as preset abnormality types such as overvoltage and undervoltage) for the subsequent abnormality final judgment module, narrowing the matching range during the final judgment, reducing computational redundancy, and allowing the final judgment link to focus on the detailed feature comparison between the candidate template and the waveform to be identified (such as the time domain curve fitting of the waveform prototype and the calculation of the abnormality determination coefficient), thereby improving the efficiency and accuracy of abnormality type judgment.

[0117] Step 4: If there is an obvious abnormality, the abnormality type is screened to obtain a candidate template type, and an abnormality determination analysis is performed on the candidate template waveform prototype and the voltage waveform to be identified to obtain the abnormality determination coefficient and finally determine the abnormality type of the electric energy meter;

[0118] If there is a corresponding abnormal type, calculate the deviation ratio between the feature matching degree of each abnormal type and the feature matching degree threshold to obtain the feature deviation ratio;

[0119] Screen the abnormal types based on the feature deviation ratio to obtain candidate template types;

[0120] It can be understood that the feature deviation ratios are sorted from small to large, and the first several (such as the first three) anomaly types with the smallest deviations are retained as candidate templates, that is, these templates are considered to be most similar to the waveform features to be identified and are used as candidate ranges for further identification;

[0121] Extract the waveform prototype of the candidate template type from the abnormal waveform dictionary, and the voltage waveform to be identified collected by the electric energy meter. Perform dynamic time warping (DTW) calculation on the waveform prototype and the voltage waveform to be identified to obtain the timing matching degree.

[0122] Preferably, by the formula: Get timing matching ;

[0123] What needs to be explained is that Indicates the dynamic time warping (DTW) distance between the voltage waveform to be identified and a template waveform of a certain abnormal type in the abnormal waveform dictionary, which measures the degree of difference in the timing characteristics of the two. The smaller the value, the closer the timing morphology).

[0124] max_d represents the preset maximum allowable distance threshold, which is used for normalization to ensure that the matching calculation result is within the interval [0,1]. It represents the maximum possible distance between the anomaly type template and other waveforms and serves as the denominator for unifying the dimensions.

[0125] The larger the value of the timing matching degree is, the higher the matching degree of the timing characteristics between the waveform to be identified and the template waveform is. =0, timing matching =1, indicating a complete match;

[0126] Obtain the timing matching degree and feature matching degree of the candidate template type, perform weighted sum processing on the timing matching degree and feature matching degree to obtain the abnormality determination coefficient;

[0127] Based on the abnormality determination coefficient, the abnormality type of the candidate template type is obtained, and finally it is determined whether the voltage waveform to be identified has an obvious abnormality of the abnormal type;

[0128] The final method for determining whether the voltage waveform to be identified has an abnormal type of explicit abnormality is as follows:

[0129] Comparing the abnormality determination coefficient with a preset abnormality determination threshold, if the abnormality determination coefficient is higher than or equal to the preset abnormality determination threshold, it is considered that the voltage waveform to be identified has an obvious abnormality of the abnormal type;

[0130] Among them, the weight coefficient of the abnormal determination coefficient is obtained as follows:

[0131] Perform S-transform on the waveform prototype and the voltage waveform to be identified to generate a time-frequency spectrum and extract the transient energy distribution entropy;

[0132] The transient energy distribution entropy is normalized, and the absolute deviation ratio between the transient energy distribution entropy of the waveform prototype and the voltage waveform to be identified is calculated as the weight coefficient.

[0133] It should be explained that the transient energy distribution entropy values ​​of all waveform prototypes in the abnormal waveform dictionary and the transient energy distribution entropy values ​​of the voltage waveform to be identified are collected, the minimum and maximum entropy values ​​in the dictionary are calculated, and Min-Max normalization processing is performed on each entropy value (including prototypes and waveforms to be identified);

[0134] Calculate the absolute deviation between the normalized entropy value of the waveform to be identified and the normalized entropy value of each waveform prototype, and obtain the absolute deviation ratio;

[0135] Finally, the absolute deviation ratio is converted into a weight coefficient so that the weight coefficient is in the range of [0,1]. The larger the weight value, the more similar the transient energy distribution entropy of the two is, and the higher the contribution of this feature in the matching. Therefore, the influence of the transient energy feature is dynamically adjusted during multi-feature fusion to improve the accuracy of anomaly recognition.

[0136] The technical solution of this embodiment is as follows: real-time acquisition of the voltage waveform to be identified of the electric energy meter, feature extraction of the voltage waveform to be identified to construct a waveform feature vector, matching calculation of the waveform feature vector with a reduced feature set, and initial determination of whether the voltage waveform to be identified has an explicit abnormality; if an explicit abnormality exists, screening the abnormality type to obtain a candidate template type, performing abnormality determination analysis on the candidate template waveform prototype and the voltage waveform to be identified, obtaining an abnormality determination coefficient, and ultimately determining the abnormality type of the electric energy meter; this is conducive to forming a closed loop of data acquisition - abnormality detection - dictionary update, and improving the adaptability of the electric energy meter to dynamic changes in abnormal waveforms.

[0137] Example 3: Figure 1 As shown, an abnormal waveform matching method for identifying abnormal voltage of an electric energy meter also includes the following steps:

[0138] Step 5: If the voltage waveform to be identified does not have an explicit abnormality in the initial determination, determine whether the voltage waveform to be identified has a hidden abnormality. If so, iteratively optimize the abnormal waveform dictionary based on the characteristics of the voltage waveform to be identified.

[0139] like Figure 2 As shown, the method for determining whether the voltage waveform to be identified has a hidden abnormality is:

[0140] S1. Construct a baseline model of normal waveform characteristics;

[0141] Preferably, the historical normal waveform of the same electric energy meter is collected, that is, there is no obvious abnormal record and the obvious abnormal record is higher than a preset fixed period, and the frequency domain, time-frequency domain and morphological features of the electric energy meter are extracted;

[0142] Calculate the mean, standard deviation, and 95% quantile of each sub-feature of the electric energy meter in the frequency domain, time-frequency domain, and morphology, and construct a dynamic baseline interval of normal features as a baseline model of normal waveform features;

[0143] A sliding window algorithm is used to dynamically update the baseline interval to adapt to periodic changes in grid load (such as voltage fluctuations during peak hours in the morning and evening);

[0144] S2. Perform multi-dimensional feature deviation calculation to mark potential abnormal features;

[0145] Preferably, if the sub-feature of the electric energy meter deviates from the dynamic baseline interval, a normalized deviation between the sub-feature and the dynamic baseline interval is calculated;

[0146] Extract high-priority features based on standardized deviations and mark them as potential abnormal features;

[0147] S3. Perform dynamic trend analysis on potential abnormal features to determine whether there is a trend abnormality;

[0148] Preferably, for the characteristic values ​​of N consecutive periods (e.g., N=10), linear regression is used to calculate the trend slope. If the absolute value of the slope exceeds 3 times the baseline fluctuation rate, it is determined that there is a trend anomaly.

[0149] S4. If there is a trend anomaly, the hidden anomaly is determined by combining the feature separability and the deviation from the standardization;

[0150] It should be explained that if there is a trend anomaly (such as a characteristic value shifting in a single direction for multiple consecutive cycles), the method for determining latent anomalies by calculating the characteristic separability and standardized deviation is as follows: for core features with high characteristic separability (such as THD and peak slope), a lower standardized deviation threshold (such as Z ≥ 1.5) is set. When the standardized deviation of such features continuously exceeds the threshold and the trend slope exceeds the normal fluctuation range (such as weekly deviation ≥ 0.5%), combined with the weight assigned to the characteristic separability (the higher the separability, the greater the weight), if the weighted comprehensive score exceeds the preset threshold, it is determined to be a latent anomaly, thereby achieving the identification of early or progressive faults;

[0151] The method for iteratively optimizing the abnormal waveform dictionary in combination with the voltage waveform features to be identified is as follows:

[0152] Preferably, when a latent anomaly or an unknown anomaly is detected, the frequency domain, time-frequency domain and morphological feature vectors of the waveform to be identified are first extracted, and compared with the existing templates in the dictionary through a density clustering algorithm to determine whether a new anomaly cluster is formed;

[0153] If it is an early or variant form of a known anomaly, the feature mean, variation range, and time series evolution pattern of the corresponding template will be updated (such as the trend feature of slow growth of harmonics). If it is a new anomaly, a new template entry will be added (including feature vector, waveform prototype, and anomaly type label).

[0154] The feature coverage and feature separability are recalculated based on the new samples, the feature priority value sorting is dynamically adjusted, the feature set is optimized and streamlined, and abnormal samples are removed from the normal baseline data to update the statistical model, which is conducive to the continuous adaptation of the dictionary to new abnormal patterns and gradual fault characteristics of equipment in power grid operation.

[0155] Example 4: Figure 3 As shown, an abnormal waveform matching system for identifying abnormal voltage in an electric energy meter includes the following modules:

[0156] Dictionary construction module: used to collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms and construct an abnormal waveform dictionary;

[0157] Simplified splitting module: performs coverage analysis on historical abnormal waveforms, obtains feature priority values, and filters the core features of the abnormal waveform dictionary to establish a simplified feature set;

[0158] Explicit initial judgment module: used to collect the voltage waveform to be identified in real time, extract the features of the voltage waveform to be identified to construct a waveform feature vector, match the waveform feature vector with the simplified feature set, and initially judge whether the voltage waveform to be identified has explicit anomalies;

[0159] Abnormal final judgment module: If there is an obvious abnormality, the abnormal type is screened to obtain a candidate template type, and abnormal determination analysis is performed on the candidate template waveform prototype and the voltage waveform to be identified to obtain the abnormal determination coefficient and finally determine the abnormal type of the electric energy meter;

[0160] Dictionary iteration module: If it is initially determined that there is no explicit abnormality in the voltage waveform to be identified, it is determined whether there is a hidden abnormality in the voltage waveform to be identified. If so, the abnormal waveform dictionary is iteratively optimized in combination with the characteristics of the voltage waveform to be identified.

[0161] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An abnormal waveform matching method for identifying abnormal voltage in an electric energy meter, characterized in that: The following steps are involved: Collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms and build an abnormal waveform dictionary; Perform coverage analysis on historical abnormal waveforms to obtain feature priority values ​​and filter them to obtain a streamlined feature set; The voltage waveform to be identified of the electric energy meter is collected in real time, and features are extracted from the voltage waveform to be identified to construct a waveform feature vector. The waveform feature vector is matched with a simplified feature set to initially determine whether there is an obvious anomaly in the voltage waveform to be identified. If there is an obvious abnormality, the abnormality type is screened to obtain a candidate template type, and an abnormality determination analysis is performed on the candidate template waveform prototype and the voltage waveform to be identified to obtain the abnormality determination coefficient and finally determine the abnormality type of the electric energy meter; If it is initially determined that the voltage waveform to be identified does not have an explicit abnormality, it is determined whether the voltage waveform to be identified has a hidden abnormality. If so, the abnormal waveform dictionary is iteratively optimized in combination with the characteristics of the voltage waveform to be identified.

2. The abnormal waveform matching method for identifying abnormal voltage of an electric energy meter according to claim 1, characterized in that: The abnormal waveform dictionary is constructed as follows: Collect historical abnormal waveforms of the electric energy meter, perform feature decomposition on the historical abnormal waveforms, and obtain identification feature vectors containing frequency domain features, time-frequency domain features, and morphological features; Based on the dynamic dictionary learning algorithm, an abnormal waveform dictionary containing identification feature vectors, abnormal types, and waveform prototypes is constructed.

3. The abnormal waveform matching method for identifying abnormal voltage of an electric energy meter according to claim 1, characterized in that: The method for obtaining the reduced feature set is: The electric energy meter to be identified is used as a sample for abnormal identification, and historical abnormal waveforms are obtained from the abnormal waveform dictionary. The historical abnormal waveforms are classified and matched according to the samples; Obtain the feature coverage and feature separability of each sample and sum them up to obtain the feature priority value; Sub-features are sorted based on their priority values, core features are extracted from the abnormal waveform dictionary, and a reduced feature set is constructed based on the feature vectors corresponding to the core features.

4. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 3, characterized in that: The method for obtaining the feature coverage and feature separability is: Calculate the frequency of occurrence of different sub-features of each sample to obtain feature coverage; Obtain the inter-class dispersion and intra-class dispersion of the samples, and calculate the ratio to obtain the feature separability.

5. The abnormal waveform matching method for identifying abnormal voltage of an electric energy meter according to claim 1, characterized in that: The method for initially judging whether the voltage waveform to be identified has an obvious abnormality is: Perform waveform preprocessing on the voltage waveform to be identified that is collected in real time, perform feature extraction on the waveform after waveform preprocessing, and construct a waveform feature vector; Extracting each abnormality type identification feature vector from the simplified feature set, calculating the matching degree between the identification feature vector and the waveform feature vector of each abnormality type, and obtaining the feature matching degree; The characteristic matching degree is compared with a preset characteristic matching degree threshold value to obtain a result of judging whether there is a dominant abnormality in the voltage waveform to be identified.

6. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 5, characterized in that: The method for calculating the matching degree is: The feature matching degree is obtained by calculating the cosine similarity between the recognition feature vector and the waveform feature vector.

7. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 1, characterized in that: The method for finally determining the abnormality type of the electric energy meter is: If there is a corresponding abnormal type, calculate the deviation ratio between the feature matching degree of each abnormal type and the feature matching degree threshold to obtain the feature deviation ratio; Screen the abnormal types based on the feature deviation ratio to obtain candidate template types; Obtain the timing matching degree and feature matching degree of the candidate template type, perform weighted sum processing on the timing matching degree and feature matching degree to obtain the abnormality determination coefficient; Based on the abnormality determination coefficient, the abnormality type of the candidate template type is obtained, and finally it is determined whether the voltage waveform to be identified has an obvious abnormality of the abnormal type.

8. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 7, characterized in that: The method for obtaining the timing matching degree is: The waveform prototype of the candidate template type and the voltage waveform to be identified collected by the electric energy meter are extracted from the abnormal waveform dictionary. The waveform prototype and the voltage waveform to be identified are dynamically time-warped to obtain the timing matching degree.

9. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 7, characterized in that: The weight coefficient of the abnormal determination coefficient is obtained as follows: Transform the waveform prototype and the voltage waveform to be identified to generate a time-frequency spectrum and extract the transient energy distribution entropy; The absolute deviation ratio between the waveform prototype and the transient energy distribution entropy of the voltage waveform to be identified is calculated as the weight coefficient.

10. The abnormal waveform matching method for identifying abnormal voltage in an electric energy meter according to claim 1, characterized in that: The method for determining whether the voltage waveform to be identified has a hidden abnormality is: Construct a baseline model of normal waveform characteristics; Perform multi-dimensional feature deviation calculation to mark potential abnormal features; Conduct dynamic trend analysis on potential abnormal features to determine whether there are trend abnormalities; If there is a trend anomaly, the latent anomaly is determined by combining the feature separability and standardized deviation.

Images