Intelligent fusion terminal-based anomaly identification method and device, equipment and medium

Abstract

The application provides an abnormality identification method and device based on an intelligent fusion terminal, equipment and a medium, and belongs to the technical field of power abnormality identification. The method comprises the following steps: acquiring multi-dimensional electrical parameters under a current working condition; determining a feature vector corresponding to the multi-dimensional electrical parameters based on the multi-dimensional electrical parameters and through a pre-trained double-channel feature learning network model; inputting the feature vector into a pre-trained abnormality scoring model to determine an abnormality score corresponding to the feature vector; if the abnormality score is higher than a preset score threshold, regarding the feature vector as an abnormal feature vector; determining a target abnormality mode of a low-voltage line of a transformer area according to the abnormal feature vector and an abnormal feature library, and the abnormal feature library is used for representing a corresponding relationship between the abnormal feature vector and an abnormal mode. The abnormality identification method and device based on the intelligent fusion terminal, equipment and the medium provided by the application can improve the accuracy of abnormality identification of the intelligent fusion terminal.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power anomaly identification, and more particularly relates to an anomaly identification method and device based on an intelligent fusion terminal, equipment and a medium. BACKGROUND

[0002] In modern industry and power fields, stable operation of an electrical system is crucial, and an intelligent fusion terminal is a smart data acquisition, calculation and communication hub installed at the end of a power system (for example, a low-voltage line of a transformer area). As a key device of the electrical system, the intelligent fusion terminal undertakes the core role of real-time monitoring of electrical parameters. The intelligent fusion terminal can monitor the low-voltage line of the transformer area in real time to ensure stable operation of the power system. According to the monitoring of the low-voltage line of the transformer area by the intelligent fusion terminal, once an anomaly occurs in the low-voltage line of the transformer area, not only may equipment failure shutdown be caused, but also power interruption and economic loss may be caused.

[0003] In the prior art, there are still certain limitations in the anomaly identification technology for the low-voltage line of the transformer area. The traditional anomaly identification method is based on a single parameter threshold monitoring method, only focuses on part of the data, and cannot comprehensively capture the influence of multi-dimensional parameters of the equipment on the operation of the equipment. In a complex mode, the traditional anomaly identification method is prone to false positives and false negatives, resulting in low accuracy of anomaly identification. SUMMARY

[0004] The application aims to provide an anomaly identification method and device based on an intelligent fusion terminal, equipment and a medium to improve the accuracy of anomaly identification of the intelligent fusion terminal.

[0005] In a first aspect, an anomaly identification method based on an intelligent fusion terminal is provided, comprising:

[0006] Obtaining multi-dimensional electrical parameters under a current working condition, the multi-dimensional electrical parameters comprising first sampling data and second sampling data, the first sampling data and the second sampling data being data of the low-voltage line of the transformer area collected under the current working condition, wherein the first sampling data comprises voltage steady-state data, current steady-state data, power data and harmonic data, and the second sampling data comprises current instantaneous value and voltage instantaneous value;

[0007] Based on the multi-dimensional electrical parameters and through a pre-trained double-channel feature learning network model, a feature vector corresponding to the multi-dimensional electrical parameters is determined, and the feature vector is used to represent the operating state of the low-voltage line of the transformer area;

[0008] The feature vector is subjected to preset number of forward propagation processing according to the pre-trained anomaly scoring model, to obtain an anomaly score set, the standard deviation of each anomaly score in the anomaly score set is taken as an uncertainty score, and the arithmetic mean of each anomaly score in the anomaly score set is taken as an anomaly score of the feature vector; the uncertainty score is used to represent the reliability of the anomaly score set output by the anomaly scoring model, and the anomaly score is used to represent the difference between the multi-dimensional electrical parameters and the normal multi-dimensional electrical parameters corresponding to the normal mode;

[0009] If the anomaly score is higher than a preset score threshold and the uncertainty score is lower than a preset standard score, the feature vector is taken as an abnormal feature vector.

[0010] According to the abnormal feature vector and the abnormal feature library, a target abnormal mode of the low-voltage line of the transformer area is determined, and the abnormal feature library is used to represent the corresponding relationship between the abnormal feature vector and the abnormal mode.

[0011] In a second aspect, an abnormality identification device based on an intelligent fusion terminal is provided, which includes:

[0012] A data acquisition module is configured to acquire multi-dimensional electrical parameters under a current working condition, the multi-dimensional electrical parameters including first sampling data and second sampling data, the first sampling data and the second sampling data being data of the low-voltage line of the transformer area collected under the current working condition, wherein the first sampling data includes voltage steady-state data, current steady-state data, power data and harmonic data, and the second sampling data includes current instantaneous value and voltage instantaneous value.

[0013] A feature vector determination module is configured to determine a feature vector corresponding to the multi-dimensional electrical parameters based on the multi-dimensional electrical parameters and through a pre-trained double-channel feature learning network model, the feature vector being used to represent the running state of the low-voltage line of the transformer area.

[0014] An anomaly score determination module is configured to subject the feature vector to preset number of forward propagation processing according to a pre-trained anomaly scoring model, to obtain an anomaly score set, take the standard deviation of each anomaly score in the anomaly score set as an uncertainty score, and take the arithmetic mean of each anomaly score in the anomaly score set as an anomaly score of the feature vector; the uncertainty score is used to represent the reliability of the anomaly score set output by the anomaly scoring model, and the anomaly score is used to represent the difference between the multi-dimensional electrical parameters and the normal multi-dimensional electrical parameters corresponding to the normal mode.

[0015] An abnormal feature vector determination module is configured to take the feature vector as an abnormal feature vector when the anomaly score is higher than a preset score threshold and the uncertainty score is lower than a preset standard score.

[0016] The target abnormal mode determination module is configured to determine a target abnormal mode of the low-voltage line of the transformer area according to the abnormal feature vector and an abnormal feature library, and the abnormal feature library is configured to represent a corresponding relationship between the abnormal feature vector and the abnormal mode.

[0017] In a third aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the abnormality identification method based on the intelligent fusion terminal are implemented.

[0018] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the abnormality identification method based on the intelligent fusion terminal are implemented.

[0019] The abnormality identification method and device based on the intelligent fusion terminal, the equipment, and the medium provided by the embodiments of the present application have the following advantages. According to the first collection data and the second collection data of the low-voltage line of the transformer area, the embodiments of the present application comprehensively monitor the low-voltage line of the transformer area in the steady state and the transient state, avoiding the problem of one-sided identification perspective. The embodiments of the present application use a pre-trained double-channel feature learning network model to process multi-dimensional electrical parameters to obtain a feature vector that can represent the operating state of the low-voltage line of the transformer area. The double channel can differentially and specifically extract features from the steady-state data and the transient data, more comprehensively mine effective features in the first multi-dimensional data, and avoid the situation that the abnormality identification is misjudged due to the loss of feature information. The embodiments of the present application score the feature vector by using a pre-trained abnormality scoring model, quantify the difference between the current feature vector and the feature vector of the normal mode by combining the abnormality score and the uncertainty score, more accurately quantify the abnormality degree, and improve the accuracy of abnormality identification. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor.

[0021] Figure 1 The flowchart of the abnormality identification method based on the intelligent fusion terminal provided by an embodiment of the present application is shown in the figure.

[0022] Figure 2 The flowchart of determining the target abnormal mode according to the cosine similarity provided by an embodiment of the present application is shown in the figure.

[0023] Figure 3A structural block diagram of an abnormality identification device based on an intelligent fusion terminal is provided in an embodiment of the present application.

[0024] Figure 4 A schematic block diagram of an electronic device is provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the application. However, persons having ordinary skill in the art will readily recognize that embodiments of the application can be practiced without these specific details. In other instances, well-known structures, devices, circuits, and processes have not been described in detail so as not to unnecessarily obscure aspects of the application.

[0026] It can be understood that, in the embodiments of the present application, data related to user information and the like is involved, and when the embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards.

[0027] It should be noted that the terms "first", "second", and the like in the description, claims and drawings of the present application are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0028] In order to make the purposes, technical solutions and advantages of the present application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0029] Reference should be made to Figure 1 , Figure 1 A flowchart of an abnormality identification method based on an intelligent fusion terminal is provided in an embodiment of the present application, which can include the following steps.

[0030] S101: Obtain multi-dimensional electrical parameters under a current working condition.

[0031] In the present embodiment, the multi-dimensional electrical parameters include first sampling data and second sampling data, which are data of a low-voltage line of a transformer area in a target time period collected under a current working condition, wherein the first sampling data includes voltage steady-state data, current steady-state data, power data and harmonic data, and the second sampling data includes current instantaneous value and voltage instantaneous value.

[0032] The embodiment can adopt high-precision voltage transformers and current transformers to collect voltage signals and current signals of the low-voltage line of the transformer area in real time, perform analog-to-digital conversion on the collected voltage signals and current signals, and extract steady-state components of the voltage signals and the current signals respectively through a digital filtering algorithm (for example, mean filtering) to obtain voltage steady-state data and current steady-state data.

[0033] According to the voltage steady-state data, the current steady-state data, and a power calculation formula, the power of the low-voltage line of the transformer area under the current working condition is determined, and the power calculation mode is as follows: , wherein, P P is the power of the low-voltage line of the transformer area under the current working condition, U V is the voltage steady-state data under the current working condition, I I is the current steady-state data, is the phase difference between the voltage and the current.

[0034] The embodiment performs fast Fourier transform on the collected voltage signals and current signals to decompose fundamental wave and harmonic components, thereby obtaining harmonic data such as harmonic order, harmonic amplitude, and harmonic content (the harmonic content is the ratio of the harmonic amplitude to the fundamental wave amplitude).

[0035] The embodiment adopts a high-speed data acquisition module with a sampling frequency generally not lower than 10 kHz to perform high-speed sampling on the voltage analog signals and the current analog signals of the low-voltage line of the transformer area, thereby obtaining continuous current instantaneous value sequences and voltage instantaneous value sequences, and obtaining current instantaneous values and voltage instantaneous values to reflect the instantaneous change characteristics of electrical quantities.

[0036] S102: Based on the multi-dimensional electrical parameters and through a pre-trained double-channel feature learning network model, a feature vector corresponding to the multi-dimensional electrical parameters is determined.

[0037] In the embodiment, the obtained multi-dimensional electrical parameters are preprocessed, and the multi-dimensional parameters are normalized to [0, 1] to eliminate dimensional differences. In the embodiment, the multi-dimensional parameters can be normalized according to a normalization formula, and the normalization formula in the embodiment can be as follows: , wherein, x is the normalized electrical parameter value, is the electrical parameter value, is the minimum value of the preset value range of the electrical parameter value, is the maximum value of the preset value range of the electrical parameter value. For example, the value range of the steady-state data of the voltage of the low-voltage line of the transformer area is [180V, 240V], and the voltage steady-state data obtained by the embodiment is 220V, and the normalized voltage steady-state data is In the embodiment, the current steady-state data, the power data, and the harmonic data are normalized by using the normalization method to determine the normalized values.

[0038] If there is data missing, the missing data can be completed according to the mean completion method or the interpolation method; if there is data anomaly, the abnormal data is rejected to ensure the integrity and consistency of the input data.

[0039] The feature vector corresponding to the multi-dimensional electrical parameter is determined based on the multi-dimensional electrical parameter and through a pre-trained double-channel feature learning network model, including:

[0040] The first sampling data is input into a first channel in the pre-trained double-channel feature learning network model to obtain a first feature vector;

[0041] The second sampling data is input into a second channel in the pre-trained double-channel feature learning network model to obtain a second feature vector;

[0042] The first feature vector and the second feature vector are spliced and fused to obtain the feature vector corresponding to the multi-dimensional electrical parameter.

[0043] The first sampling data after preprocessing is spliced and fused into a one-dimensional feature array in this embodiment, which can be expressed as [voltage steady-state data, current steady-state data, power data, harmonic data]. The one-dimensional feature array is a horizontal splicing of steady-state data with different physical meanings, and is used to represent the overall state of the current low-voltage line working condition of the transformer area.

[0044] The voltage instantaneous value sequence and the current instantaneous value sequence in the second sampling data after preprocessing are obtained respectively, and a table with a sequence length number and a dimension number is constructed according to the voltage instantaneous value sequence and the current instantaneous value sequence. The sequence length number represents the total number of sampling points, i.e. the number of time points; the dimension number represents several physical dimensions (voltage, current). In this embodiment, the two-dimensional matrix is a longitudinal record of instantaneous parameters with the same physical meaning changing over time, and is used to represent the dynamic change of the instantaneous parameters on the time axis.

[0045] In one possible implementation, if the sequence length number is 3 and the dimension number is 2, the table is a table with three rows and two columns, which can be shown in Table 1.

[0046] Table 1

[0047]

[0048] In Table 1, the first column represents the voltage instantaneous value of the low-voltage line of the transformer area, and the second column represents the current instantaneous value of the low-voltage line of the transformer area. According to the two-dimensional table, the two-dimensional vector corresponding to the 0 time point is [220.1, 9.95], t the two-dimensional vector corresponding to the 1 time point is [219.3, 10.08], t the two-dimensional vector corresponding to the 1 time point is [219.3, 10.08], t3 The two-dimensional vector corresponding to the time point is [219.0, 9.85].

[0049] In this embodiment, the one-dimensional feature array is input into the first channel of the pre-trained double-channel feature learning network model. The first channel of this embodiment can adopt a neural network structure composed of multiple fully connected layers, or can adopt a convolutional neural network to extract the nonlinear relationship between features. After the one-dimensional feature array input is subjected to nonlinear transformation by multiple fully connected layers, the first feature vector is output at the hidden layer or output layer of the first channel. The first feature vector is used to represent the energy, quality and distortion of the steady-state operation of the low-voltage line in the transformer area.

[0050] The two-dimensional vectors are input into the second channel of the pre-trained double-channel feature learning network model. The second channel of this embodiment adopts a long short-term memory network to determine the dynamic characteristics in each two-dimensional vector and outputs a second feature vector. The second feature vector is used to represent the dynamic response and transient behavior characteristics of the low-voltage line in the transformer area, and can determine the decay trend of the second sampling data or the trend change of the second sampling data.

[0051] In this embodiment, the first feature vector and the second feature vector are dimensionally aligned, the first feature vector and the second feature vector after dimensional alignment are spliced and fused to obtain a spliced and fused feature vector, and the spliced and fused feature vector is taken as a feature vector corresponding to the multi-dimensional electrical parameter.

[0052] S103: input the feature vector into a pre-trained anomaly score model to determine an anomaly score corresponding to the feature vector; if the anomaly score is higher than a preset score threshold, the feature vector is taken as an abnormal feature vector.

[0053] The anomaly score in this embodiment is used to represent the difference between the multi-dimensional electrical parameter and the normal multi-dimensional electrical parameter corresponding to the normal mode.

[0054] In this embodiment, the above feature vector is input into a pre-trained anomaly score model, and the model is used to perform inference operation on the input feature vector to determine an anomaly score corresponding to the feature vector. The higher the anomaly score, the greater the difference between the current working condition and the normal mode of working condition. The greater the difference, the greater the possibility of failure of the current working condition.

[0055] In this embodiment, a gradient boosting decision tree model is used as the anomaly score model.

[0056] The historical multi-dimensional electrical parameters under a large number of historical working conditions (including normal working conditions and abnormal working conditions) of the low-voltage line of the transformer area are acquired, and the feature vectors corresponding to each historical multi-dimensional electrical parameter are determined based on the above-mentioned double-channel feature learning network model. A label is assigned to each feature vector. In this embodiment, label A0 represents normal working conditions, and label A1 represents abnormal working conditions. The feature vector and the corresponding label are used as input data of the gradient boosting decision tree model.

[0057] The historical multi-dimensional electrical parameters after the label assignment are divided into a training set and a validation set according to a preset proportion. In this embodiment, the model is trained through the training set, so that the model learns the complex mapping relationship from the feature vector to the corresponding label. The loss function of this embodiment can use the binary cross-entropy loss function to minimize the difference between the predicted probability of the model and the true label. Gradient descent algorithm and its variants (for example, Adam optimizer) are used to iteratively optimize the model parameters on the training set, and the performance is evaluated on the validation set. Through early stopping and other techniques, overfitting is prevented, and finally the optimal gradient boosting decision tree model is obtained.

[0058] The feature vector corresponding to the current working condition is input into the pre-trained abnormal score model, and each base model of the model independently performs a forward propagation inference, and outputs a corresponding prediction value. An abnormal score set {d1, d2,..., d D} containing multiple prediction values is obtained. Each d is a continuous probability value between 0 and 1, and the probability value is used as an abnormal score. The continuous probability value in this embodiment is used as the confidence of the model determining that the input feature vector belongs to the abnormal category.

[0059] The arithmetic mean of the abnormal score set is determined according to each abnormal data in the abnormal score set, and the arithmetic mean is used as the abnormal score of the feature vector. The abnormal score is used to represent the difference between the multi-dimensional electrical parameters and the normal multi-dimensional electrical parameters corresponding to the normal mode.

[0060] The standard deviation of the abnormal score set is determined according to each abnormal data in the abnormal score set, and the standard deviation is used as the uncertainty score of the feature vector. The uncertainty score is used to represent the reliability of the abnormal score set output by the abnormal score model. The larger the standard deviation, the greater the disagreement in the model integration for the input, and the higher the uncertainty.

[0061] If the abnormal score is higher than the preset score threshold, and the uncertainty score is lower than the preset standard score, the current feature vector is regarded as an abnormal feature vector, which represents that the current low-voltage line of the transformer area enters a high-confidence abnormal processing flow. Subsequently, it can be directly matched with the abnormal feature library to determine the specific abnormal mode.

[0062] If the abnormality score is higher than the preset score threshold, but the uncertainty score is not lower than the preset standard score, the current feature vector is taken as an alert feature vector. The alert feature vector is used to represent that the low-voltage line of the transformer area is in a risk-to-be-diagnosed state; at this time, a maintenance manual intervention review process needs to be intervened, or a more complex diagnosis program needs to be started.

[0063] S104: Determine the target abnormal mode of the low-voltage line of the transformer area according to the abnormal feature vector and the abnormal feature library.

[0064] The abnormal feature library in the embodiment is used to represent the corresponding relationship between the feature vector and the abnormal mode type, and the abnormal mode type can include overload, three-phase imbalance, or harmonic overlimit.

[0065] The obtained abnormal feature vector is matched with the feature vector of each abnormal mode type in the abnormal feature library, and the abnormal mode type of the low-voltage line of the transformer area is determined as the target abnormal mode according to the matching result.

[0066] In the embodiment, the target abnormal mode of the low-voltage line of the transformer area is determined according to the abnormal feature vector and the abnormal feature library, which includes:

[0067] The cosine similarity values corresponding between the abnormal feature vector and each feature vector in the abnormal feature library are calculated to determine a cosine similarity value group.

[0068] The two cosine similarity values with the highest cosine similarity values are selected from the cosine similarity value group, the two cosine similarity values with the highest cosine similarity values include a first similarity value and a second similarity value, and the first similarity value is greater than the second similarity value.

[0069] If the absolute value of the difference between the two cosine similarity values is higher than a preset similarity threshold, the abnormal mode corresponding to the first similarity value is taken as the target abnormal mode of the low-voltage line of the transformer area.

[0070] If the absolute value of the difference between the two cosine similarity values is not higher than the preset similarity threshold, the cosine similarity values in the cosine similarity value group are arranged in descending order, the abnormal modes corresponding to the first preset number of cosine similarity values are determined, a candidate abnormal mode set is formed according to the abnormal modes, and the target abnormal mode of the low-voltage line of the transformer area is determined according to the candidate abnormal mode set.

[0071] Referring to Figure 2 In the embodiment, the cosine similarity values between the input abnormal feature vector and all feature vectors in the abnormal feature library are determined using a vectorization calculation formula, and the cosine similarity calculation formula is: wherein, s is the cosine similarity value, is the input abnormal feature vector, The feature vector in the abnormal feature library.

[0072] The embodiment sets the feature vector in the abnormal feature library. k The cosine similarity value group is determined according to the obtained cosine similarity values, and the cosine similarity value group can be represented as

[0073] The first similarity value and the second similarity value in the cosine similarity value group are selected according to the cosine similarity value group, the first similarity value and the second similarity value are the two largest similarity values in the cosine similarity value group, wherein the first similarity value is higher than the second similarity value, and the abnormal mode corresponding to the first similarity value and the second similarity value is recorded.

[0074] The absolute value of the similarity difference value is determined according to the first similarity value and the second similarity value, the absolute value of the similarity difference value is compared with the preset similarity threshold value, if the absolute value of the difference value is greater than the preset similarity threshold value, it is determined that the abnormal mode corresponding to the first similarity value is the target abnormal mode of the low-voltage line in the transformer area. In the embodiment, the preset similarity threshold value can be set to 0.1 according to experimental experience, and the setting of the preset similarity threshold value can directly affect the determination accuracy, which can be adjusted according to different application scenarios.

[0075] If the absolute value of the difference between the two cosine similarity values is less than or equal to the preset similarity threshold value, it is determined that the first similarity value and the second similarity value are close, and in this case, the target abnormal mode of the low-voltage line in the transformer area cannot be determined only according to the first similarity value and the second similarity value. At this time, all cosine similarity values in the entire cosine similarity value group are sorted in descending order, a preset number of cosine similarity values are selected from the sorted cosine similarity value group, a candidate cosine similarity value group is determined according to the selected cosine similarity values, the abnormal mode corresponding to each cosine similarity value in the candidate cosine similarity value group is determined, and a candidate abnormal mode set is determined according to each abnormal mode. In the embodiment, the N cosine similarity values with the highest cosine similarity values can be selected from the sorted cosine similarity value group, wherein N is the preset number, which can be adjusted according to the data type and the scene, and the preset number in the embodiment is 5. The target abnormal mode of the low-voltage circuit in the transformer area is determined according to the feature vector corresponding to each abnormal mode in the candidate abnormal mode set.

[0076] ​From the above, it can be concluded that the embodiment collects the first sampling data (steady-state data) and the second sampling data (instantaneous data) of the low-voltage line of the transformer area, and inputs the first sampling data and the second sampling data into the corresponding channels of the dual-channel feature learning network model respectively, more specifically obtains the features in various types of data, and comprehensively monitors the operation state of the low-voltage line of the transformer area; the embodiment splices and fuses the first feature vector and the second feature vector output by the dual-channel feature learning network model, which can capture the complex correlation between multiple electrical parameters, improve the dynamic monitoring between multiple dimensions of electrical parameters, help monitor potential faults of the low-voltage line of the transformer area, and reduce the false positive rate and the false negative rate.

[0077] In an embodiment of the present application, the target abnormal mode of the low-voltage line of the transformer area is determined according to the candidate abnormal mode set, comprising:

[0078] According to the feature vectors corresponding to each abnormal mode in the candidate abnormal mode set, a feature vector set is determined, which is a set composed of the feature vectors corresponding to all abnormal modes in the candidate abnormal mode set;

[0079] According to the clustering algorithm, the feature vectors in the feature vector set are clustered to obtain the class clusters corresponding to each feature vector respectively;

[0080] The class clusters corresponding to each feature vector are mapped to the candidate abnormal mode set to obtain an abnormal mode grouping with a clustering label;

[0081] For each group of abnormal modes, an abnormal representative mode of the abnormal modes in the group is determined to obtain the abnormal representative modes corresponding to each group respectively;

[0082] According to the distance between each feature vector in each group of abnormal modes and the feature vector of the abnormal representative mode, an average distance in the group is determined;

[0083] The average distance is converted into an intra-group consistency score according to a consistency conversion formula, and the consistency conversion formula is: wherein, is the intra-group consistency score, is a scaling factor, is the normalized average distance in the group;

[0084] The minimum distance between the feature vector of the abnormal representative mode in each group of abnormal modes and the feature vector of the other abnormal representative mode is determined;

[0085] The minimum distance is converted into an inter-group separation degree score according to a separation degree conversion formula, and the separation degree conversion formula is: wherein, is the inter-group separation degree score, is a separation degree scaling factor, and the separation degree scaling factor is a regulation parameter, used to represent the rate of growth of the inter-group separation score with the minimum distance, is the minimum distance between the feature vector of the abnormal representative mode and the feature vector of other abnormal representative modes;

[0086] According to the intra-group consistency score, the inter-group separation score, and the confidence formula, the confidence value of each abnormal representative mode is determined, and the confidence formula is: wherein, c is the confidence value of each abnormal representative mode, is the intra-group consistency score, is the inter-group separation score, is the weight corresponding to the intra-group consistency score, is the weight corresponding to the inter-group separation score;

[0087] According to the confidence value and the preset confidence threshold, the target abnormal mode of the low-voltage line of the transformer area is determined.

[0088] The candidate abnormal mode set in the embodiment can be M ={ m 1, m 2, m 3, m 4, m 5} wherein, m i is the i-th abnormal mode in the candidate abnormal mode set, i≤5, and each abnormal mode in the candidate abnormal mode set represents a preliminary judgment of the abnormal situation of the low-voltage line of the transformer area.

[0089] The key features corresponding to each abnormal mode in the candidate abnormal mode set are obtained, the corresponding feature vectors are determined according to the key features of each abnormal mode, and the feature vector set is determined according to each feature vector. The feature vector set in the embodiment can be V ={ v 1, v 2, v 3, v 4, v 5} wherein, v i is the i-th feature vector in the feature vector set, i≤5.

[0090] The key features in the embodiment can include electrical quantity features (for example, electrical deviation rate, current effective value, and three-phase unbalance degree, etc.), time features (for example, time of abnormal occurrence, duration, etc.), statistical features (for example, mean and variance of data before and after the abnormal occurrence, etc.), and mode features (for example, the feature of overload mode can be load rate, and the feature of electric leakage mode can be the size of zero sequence current, etc.).

[0091] This embodiment extracts two features from each feature vector, including average current (A) and current fluctuation rate (%). v 1=[10,1], this eigenvector represents a low current and steady state. v 2=[12,1.5], this eigenvector represents a low current and relatively stable state. v 3=[85,2], this eigenvector represents a high current and steady state. v 4=[90,25], this eigenvector represents a state of high current and violent fluctuations. v 5=[88,30], this eigenvector represents a state of high current and violent fluctuations.

[0092] In this embodiment, cluster analysis is performed on each feature vector according to the K-means algorithm. The number of clusters is set to K=3, and three random initial clusters are set. The initial clusters include a first cluster, a second cluster, and a third cluster. The centroid coordinates of the first cluster are [12,1], the centroid coordinates of the second cluster are [85,2], and the centroid coordinates of the third cluster are [90,25].

[0093] The distance from each feature vector in the feature vector set to each initial cluster is calculated based on Euclidean distance. In this embodiment, the feature vectors are used as the basis for this calculation. v Taking 1 as an example, determine the feature vector. v The distance from 1 to the centroid coordinates of each cluster, where the eigenvectors v The distance from 1 to the centroid coordinates of the first type of cluster is: eigenvectors v The distance from the centroid coordinates of cluster 1 to the second type is: eigenvectors v The distance from the centroid coordinates of cluster 1 to the third type is: The eigenvectors are obtained based on the calculation. v The distances from 1 to the centroid coordinates of each cluster are 2, 75.01, and 83.52, respectively. Then, the eigenvectors... v 1. Assign to the first cluster. Based on the Euclidean distance calculation method described above, determine the distances of each other feature vector to the centroid coordinates of each cluster, and determine the cluster corresponding to each feature vector based on these distances. In this embodiment, based on the above determination method, the feature vectors are assigned to the first cluster. v 2 is assigned to the first cluster, and the feature vector is... v 3. Assign to the second cluster, and transfer the feature vector. v 4. Assigning feature vector v5 to the third cluster, thus determining the feature vectors in each cluster as follows: The first cluster includes feature vector v5. v 1 and eigenvectors v2; The second type of cluster includes feature vectors v 3; The third type of cluster includes feature vectors v 4 and eigenvectors v 5.

[0094] The algorithm iterates by re-determining new cluster centroids based on the average values ​​of each cluster. Using the new centroids and eigenvectors, and determining the distance from each eigenvector to the new centroid based on Euclidean distance, the algorithm identifies the cluster corresponding to each eigenvector. In this embodiment, after determining the new centroids, the assigned results are consistent with those before the iteration, and the positions and changes of the centroids are minimal. This indicates that the clustering algorithm has converged, and the output cluster set is determined. C ={ C 1, C 2, C 3}, where, C 1 represents the first type of cluster. C 2 represents the second type of cluster. C 3 represents the third type of cluster. C 1={ v 1, v 2}, C 2={ v 3}, C 3={ v 4, v 5}. In this embodiment, a corresponding tag is matched for each type of cluster. The tags corresponding to each type of cluster include normal light load operation, stable overload operation, and intermittent overload operation or failure. Specifically, tag 1 corresponds to normal light load operation, tag 2 corresponds to stable overload operation, and tag 3 corresponds to intermittent overload operation or failure.

[0095] This embodiment maps labels to corresponding anomalous patterns based on the correspondence between feature vectors and feature vectors of anomalous patterns, obtaining anomalous pattern grouping with clustering labels. The anomalous pattern grouping can be as follows: ,in, , indicating that the j-th abnormal pattern group is composed of all those that satisfy conditional m i The composition is defined as j≤3.

[0096] For each abnormal pattern group, this embodiment can determine the representative abnormal pattern in that group using the centroid method. Specifically, the abnormal pattern corresponding to the centroid (mean vector) of all feature vectors in the group is taken as the representative abnormal pattern. If the centroid does not correspond to any abnormal pattern, the abnormal pattern closest to the centroid is selected as the representative abnormal pattern. This embodiment determines the first abnormal pattern group according to this method. G The exception representation pattern in 1 is v 2. Second Abnormal Pattern GroupingG 2, the abnormal representative mode in the group is v 3, the third abnormal mode group G 3, the abnormal representative mode in the group is v 5.

[0097] For each group, the distance between each feature vector in the group and the abnormal representative vector is calculated according to the Euclidean distance, and the average value of all distances is calculated to obtain the average distance in the group d , the average distance is normalized to obtain the normalized average distance . The normalized average distance is converted into the consistency score in the group according to the consistency conversion formula. The consistency conversion formula used in this embodiment is: , wherein , wherein is the group consistency score of the group, is the consistency scaling factor, which is used to represent the sensitivity of the group consistency score to the average distance. The greater the consistency scaling factor, the faster the group consistency score decreases when the group average distance increases, is the normalized average distance in the group.

[0098] For each group, the distance between the feature vector of the abnormal representative mode in the group and the feature vector corresponding to the abnormal representative mode in other groups is determined according to the Euclidean distance, and the minimum distance among these distances is determined d 1, the minimum distance is normalized to obtain the normalized minimum distance . The normalized minimum distance is converted into the inter-group separation degree score according to the separation degree conversion formula. The separation degree conversion formula used in this embodiment is: , wherein , wherein is the separation degree scaling factor, which represents the rate at which the inter-group separation degree score increases with the minimum distance. The greater the separation degree scaling factor, the faster the inter-group separation degree score increases to 1 with the minimum distance, is the normalized minimum distance between the feature vector of the abnormal representative mode and the feature vector of other abnormal representative modes. This embodiment uses the inter-group separation degree score to represent the difference between the abnormal representative mode of the group and the abnormal representative mode of other groups. The higher the inter-group separation degree score, the more unique the group is.

[0099] This embodiment determines the confidence value corresponding to each abnormal representative mode based on the above-mentioned group consistency score and inter-group separation degree score according to the confidence formula. Specifically, the confidence formula used in this embodiment can be: , wherein c is the confidence value of each abnormal representative mode, is an intra-group consistency score, is an inter-group separation score, is a weight corresponding to the intra-group consistency score, is a weight corresponding to the inter-group separation score, The embodiment can be adjusted according to actual scenes. For example, if more attention is paid to the intra-group consistency score, the value of 0.7 can be set as the preset confidence threshold. If more attention is paid to the inter-group separation score, the value of 0.7 can be set as the preset confidence threshold. If the intra-group consistency score and the inter-group separation score are equally important, the value of 0.5 can be set as the preset confidence threshold. .

[0100] The embodiment can set the preset confidence threshold to 0.7, and filter out abnormal representative modes with a confidence value greater than or equal to the preset confidence threshold from all confidence values corresponding to the abnormal representative modes obtained above. The filtered abnormal representative modes are taken as target abnormal modes of the low-voltage line of the transformer area.

[0101] When the absolute value of the difference between the two cosine similarity values is less than or equal to the preset similarity threshold, the embodiment starts a mechanism similar to a more accurate expert group consultation. The feature vectors corresponding to the abnormal modes are clustered. Through clustering, similar abnormal modes are grouped to improve the distinguishability of the abnormal modes. The abnormal representative modes after clustering can effectively extract the core features of each group of abnormal modes. The embodiment also quantifies the confidence values of the abnormal representative modes by combining the intra-group consistency score, the inter-group separation score, and the confidence formula, thereby improving the stability and reliability of the identification result. Then, the confidence values determined are compared with the preset confidence threshold to determine target abnormal modes with high confidence, thereby improving the accuracy of target abnormal mode identification.

[0102] In an embodiment of the present application, the method further comprises:

[0103] If the abnormal score is not higher than the preset score threshold, the abnormal score is converted into a health index according to a health index conversion formula, and the health index conversion formula is: wherein, H is a health index, h is an abnormal score, is a preset score threshold;

[0104] A historical health index sequence and a time sequence of the low-voltage line of the transformer area in a preset historical time period are obtained, and the time sequence corresponds to the health index of the historical time period in a one-to-one manner.

[0105] The time sequence is converted into a time value to form a time value sequence.

[0106] According to the historical health index sequence, the time value sequence and the linear regression equation, a linear relationship between the health index and the time value is determined.

[0107] According to the linear relationship, a slope is determined.

[0108] According to the health index and the slope, a health level of the low-voltage line of the transformer area is determined.

[0109] In the embodiment, if the calculated abnormal score of the low-voltage line of the transformer area is not higher than a preset score threshold, the abnormal score is converted into a corresponding health index according to a health index conversion formula, and the health index conversion formula in the embodiment is: wherein, H H is the health index, h is the abnormal score, and the preset score threshold.

[0110] In the embodiment, historical health indexes of the low-voltage line of the transformer area in a preset historical time period are obtained, and the preset historical time period can be the past 30 days. The health index of each day in the preset historical time period is obtained to form a historical health index sequence. In the embodiment, the date corresponding to each health index is recorded to form a time sequence.

[0111] The manner of converting the obtained time sequence into a value sequence can be that the number of days from the starting point is taken as the time value. For example, the first day is taken as , the second day is taken as , the n th day is taken as , and the value sequence is determined according to the conversion manner as .

[0112] In the embodiment, the linear relationship between the health index and the time value is determined according to the health index in the historical health index sequence, the time value in the time value sequence and a linear regression equation. The linear regression equation can be: wherein, H is the health index, b is the slope of the linear regression equation, representing the change trend of the health index, is the i th time value, a is the intercept of the linear regression equation. If the slope in the embodiment is-0.5, it means that the health index decreases by 0.5 per day.

[0113] According to the least square method and the linear relationship, the slope is determined, and the calculation formula of the slope in the embodiment is: wherein, n is the total number of time values, t i is the i th time value,H i is the sum of the time values, is the sum of the time values, is the sum of the health indexes, is the sum of the products of the time values and the health indexes, is the sum of the squares of the time values, is the square of the sum of the time values.

[0114] If the slope is less than 0, it is determined that the health index decreases over time, that is, the health status of the low-voltage line of the transformer area is deteriorating; if the slope is approximately equal to 0, it is determined that the health index is basically stable, that is, the health status of the low-voltage line of the transformer area remains stable; if the slope is greater than 0, it is determined that the health index increases over time, that is, the health status of the low-voltage line of the transformer area is improving.

[0115] In this embodiment, the health level of the low-voltage line of the transformer area is determined according to the slope and the health index of the low-voltage line of the transformer area.

[0116] In this embodiment, the date and time points that cannot be directly used for mathematical calculation are converted into a series of continuous and equally spaced numerical values, which helps to perform regression analysis; in this embodiment, the health status of the low-voltage line of the transformer area is quantified as an index and a slope, and a time dimension is introduced, and the trend of the health status of the low-voltage line of the transformer area is analyzed through the slope, which helps to foresee potential abnormalities in advance, realize early warning, and help to intervene in the area where abnormalities may occur in advance, thereby improving the reliability of power supply of the transformer area.

[0117] In this embodiment, the health level of the low-voltage line of the transformer area is determined according to the health index and the slope, including:

[0118] If the health index is greater than a preset first health index threshold and the trend slope is greater than a preset first slope threshold, the health level is determined to be a first health level;

[0119] If the health index is less than or equal to a preset first health index threshold and greater than a preset second health index threshold, or the trend slope is less than or equal to a preset second slope threshold, the health level is determined to be a second health level;

[0120] If the health index is less than or equal to a preset second health index threshold, the health level is determined to be a third health level.

[0121] The health status of the first health level, the second health level and the third health level in this embodiment decreases in turn.

[0122] The preset first health degree threshold in the embodiment is greater than the preset second health degree threshold. The preset first health degree threshold can be set to 80, and the preset second health degree threshold can be set to 60. The preset first slope in the embodiment is greater than the preset second slope threshold. The preset first slope threshold can be set to 0, and the preset second slope threshold can be set to -0.5.

[0123] In the embodiment, if the health degree index is greater than the preset first health degree threshold and the trend slope is greater than the preset first slope threshold, at this time, the state of the low-voltage line of the table area is very healthy, and the trend of the future health degree tends to be stable, or when the slope is positive, the health degree has an upward trend, indicating that there is no sign of deterioration of the health degree at this time, and the health level of the low-voltage line of the table area is determined as the first health level.

[0124] If the health degree index is less than or equal to the preset first health degree threshold and greater than the preset second health degree threshold, or the trend slope is less than or equal to the preset second slope threshold, at this time, there are two cases that satisfy the condition. The first case is that the health degree index of the low-voltage line of the table area is less than or equal to the preset first health degree threshold and greater than the preset second health degree threshold, at this time, the health degree index of the low-voltage line of the table area is low, but higher than the preset second health degree threshold. The second case is that the health degree index of the low-voltage line of the table area is greater than the preset first health degree threshold, but the corresponding slope is less than or equal to the preset second slope threshold, at this time, the health degree of the low-voltage line of the table area is high, but there is a significant deterioration trend. The health level of the low-voltage line of the table area is determined as the second health level.

[0125] If the health degree index is less than or equal to the preset second health degree threshold, at this time, no matter how the trend of the slope changes, the state of the low-voltage line of the table area is in a poor state, and the health level of the low-voltage line of the table area is determined as the third health level. The health states of the first health level, the second health level, and the third health level decrease in turn.

[0126] For example, if the health degree index of the current low-voltage line of the table area is 76 according to the health degree index conversion formula, and the slope is -0.72 according to the slope calculation formula. At this time, for the health degree index, the health degree index is less than the preset first health degree threshold and greater than the preset second health degree threshold, which satisfies the condition of the second health level. For the slope, the slope is less than the preset second slope threshold, which also satisfies the condition of the second health level. The health level of the low-voltage line of the table area is determined as the second health level, but there is a deterioration trend, which needs to be paid attention to.

[0127] Based on the same inventive concept, the embodiments of the present application also provide an intelligent fusion terminal-based anomaly identification device for implementing the intelligent fusion terminal-based anomaly identification method described above. The implementation scheme for solving problems provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more intelligent fusion terminal-based anomaly identification device embodiments provided below can be referred to the limitations of the intelligent fusion terminal-based anomaly identification method in the above, which will not be described here.

[0128] The embodiments of the present application provide an intelligent fusion terminal-based anomaly identification device, as shown in Figure 3 The intelligent fusion terminal-based anomaly identification device 30 includes a data acquisition module 31, a feature vector determination module 32, an anomaly score determination module 33, an anomaly feature vector determination module 34, and a target anomaly mode determination module 35.

[0129] In an embodiment of the present application, the data acquisition module 31 is configured to acquire multi-dimensional electrical parameters under a current working condition, the multi-dimensional electrical parameters including first sampling data and second sampling data, the first sampling data and the second sampling data being data of a low-voltage line in a transformer area collected under the current working condition, wherein the first sampling data includes voltage steady-state data, current steady-state data, power data, and harmonic data, and the second sampling data includes current instantaneous value and voltage instantaneous value.

[0130] The feature vector determination module 32 is configured to determine a feature vector corresponding to the multi-dimensional electrical parameters based on the multi-dimensional electrical parameters and through a pre-trained double-channel feature learning network model, the feature vector being used to represent an operating state of the low-voltage line in the transformer area.

[0131] The anomaly score determination module 33 is configured to perform forward propagation processing on the feature vector for a preset number of times according to a pre-trained anomaly scoring model, obtain an anomaly score set, take a standard deviation of each anomaly score in the anomaly score set as an uncertainty score, and take an arithmetic mean of each anomaly score in the anomaly score set as an anomaly score of the feature vector; the uncertainty score is used to represent a reliability degree of the anomaly score set output by the anomaly scoring model, and the anomaly score is used to represent a difference between the multi-dimensional electrical parameters and normal multi-dimensional electrical parameters corresponding to a normal mode.

[0132] The anomaly feature vector determination module 34 is configured to take the feature vector as an anomaly feature vector when the anomaly score is higher than a preset score threshold and the uncertainty score is lower than a preset standard score.

[0133] The target anomaly mode determination module 35 is configured to determine a target anomaly mode of the low-voltage line in the transformer area according to the anomaly feature vector and an anomaly feature library, the anomaly feature library being used to represent a corresponding relationship between the anomaly feature vector and an anomaly mode.

[0134] In an embodiment of the present application, when the feature vector corresponding to the multi-dimensional electrical parameter is determined based on the multi-dimensional electrical parameter and through the pre-trained double-channel feature learning network model, the feature vector determination module 32 is specifically configured to:

[0135] input the first sampling data into the first channel of the pre-trained double-channel feature learning network model to obtain the first feature vector;

[0136] input the second sampling data into the second channel of the pre-trained double-channel feature learning network model to obtain the second feature vector;

[0137] splice and fuse the first feature vector and the second feature vector to obtain the feature vector corresponding to the multi-dimensional electrical parameter.

[0138] In an embodiment of the present application, when the target abnormal mode of the low-voltage line of the transformer area is determined according to the abnormal feature vector and the abnormal feature library, the target abnormal mode determination module 35 is specifically configured to:

[0139] calculate the cosine similarity values corresponding between the abnormal feature vector and each feature vector in the abnormal feature library to determine a cosine similarity value group;

[0140] select the two cosine similarity values with the highest values from the cosine similarity value group, the two cosine similarity values with the highest values include a first similarity value and a second similarity value, the first similarity value is greater than the second similarity value;

[0141] if the absolute value of the difference between the two cosine similarity values is higher than a preset similarity threshold, the abnormal mode corresponding to the first similarity value is taken as the target abnormal mode of the low-voltage line of the transformer area;

[0142] if the absolute value of the difference between the two cosine similarity values is not higher than the preset similarity threshold, the cosine similarity values in the cosine similarity value group are arranged in descending order, the abnormal modes corresponding to the first preset number of cosine similarity values are determined, the candidate abnormal mode set is formed according to the abnormal modes, and the target abnormal mode of the low-voltage line of the transformer area is determined according to the candidate abnormal mode set.

[0143] In an embodiment of the present application, when the target abnormal mode of the low-voltage line of the transformer area is determined according to the candidate abnormal mode set, the target abnormal mode determination module 35 is specifically configured to:

[0144] determine the feature vector set according to the feature vectors corresponding to each abnormal mode in the candidate abnormal mode set, the feature vector set is a set composed of the feature vectors corresponding to all abnormal modes in the candidate abnormal mode set;

[0145] According to the clustering algorithm, the feature vectors in the feature vector set are clustered to obtain a class cluster corresponding to each feature vector respectively;

[0146] The class cluster corresponding to each feature vector is mapped to the candidate abnormal mode set to obtain an abnormal mode group with a clustering label;

[0147] For each abnormal mode group, an abnormal representative mode of the abnormal modes in the group is determined to obtain an abnormal representative mode corresponding to each group respectively;

[0148] The intra-group consistency score of each abnormal mode in the abnormal mode group and the inter-group separation degree score of the abnormal modes in the abnormal mode group are obtained;

[0149] According to the intra-group consistency score, the inter-group separation degree score, and a confidence formula, a confidence value of each abnormal representative mode is determined, and the confidence formula is: wherein c is the confidence value of each abnormal representative mode, is the intra-group consistency score, is the inter-group separation degree score, is a weight corresponding to the intra-group consistency score, is a weight corresponding to the inter-group separation degree score;

[0150] According to the confidence value and a preset confidence threshold, a target abnormal mode of the low-voltage line of the transformer area is determined.

[0151] In an embodiment of the present application, when the intra-group consistency score of each abnormal mode in the abnormal mode group and the inter-group separation degree score of the abnormal modes in the abnormal mode group are obtained, the target abnormal mode determination module 35 is specifically configured to:

[0152] According to the distance between each feature vector in each abnormal mode and the feature vector of the abnormal representative mode, an average distance in the group is determined;

[0153] The average distance is converted into an intra-group consistency score according to a consistency conversion formula, and the consistency conversion formula is: wherein, is the intra-group consistency score, is a scaling factor, is the intra-group normalized average distance;

[0154] The minimum distance between the feature vector of the abnormal representative mode in each abnormal mode and the feature vector of other abnormal representative modes is determined;

[0155] The minimum distance is converted into an inter-group separation degree score according to a separation degree conversion formula, and the separation degree conversion formula is: wherein, is the inter-group separation degree score, is a degree of separation scaling factor, is the minimum distance between the feature vector of the abnormal representative mode and the feature vector of other abnormal representative modes.

[0156] In an embodiment of the present application, the abnormality identification device based on the intelligent fusion terminal further comprises a health state judgment module, which is configured to:

[0157] According to the judgment result that the abnormal score is not higher than the preset score threshold, the abnormal score is converted into a health index through a health index conversion formula, and the health index conversion formula is: wherein, H is the health index, h is the abnormal score, is the preset score threshold;

[0158] A historical health index sequence and a time sequence of the low-voltage line of the transformer area in a preset historical time period are obtained, and the time sequence corresponds to the health index of the historical time period one by one.

[0159] The time sequence is converted into a time value to form a time value sequence.

[0160] According to the historical health index sequence, the time value sequence, and a linear regression equation, a linear relationship between the health index and the time value is determined.

[0161] According to the linear relationship, a slope is determined.

[0162] According to the health index and the slope, a health level of the low-voltage line of the transformer area is determined.

[0163] In an embodiment of the present application, when the health level of the low-voltage line of the transformer area is determined according to the health index and the slope, the health state judgment module is specifically configured to:

[0164] The health level of the low-voltage line of the transformer area is determined.

[0165] If the health index is greater than a preset first health index threshold and the trend slope is greater than a preset first slope threshold, the health level is determined to be a first health level.

[0166] If the health index is less than or equal to the preset first health index threshold and greater than a preset second health index threshold, or the trend slope is less than or equal to a preset second slope threshold, the health level is determined to be a second health level.

[0167] If the health index is less than or equal to the preset second health index threshold, the health level is determined to be a third health level, wherein the health states of the first health level, the second health level, and the third health level decrease in turn.

[0168] See Figure 4 , Figure 4 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 4 The electronic device 400 in this embodiment may include one or more processors 401, one or more input devices 402, one or more output devices 403, and one or more memories 404. The processors 401, input devices 402, output devices 403, and memories 404 communicate with each other via a communication bus 405. The memories 404 store computer programs, including program instructions. The processors 401 execute the program instructions stored in the memories 404. Specifically, the processors 401 are configured to invoke the program instructions to perform the functions of each module / unit in the above-described device embodiments, for example... Figure 3 The functions of the data acquisition module 31, feature vector determination module 32, anomaly score determination module 33, anomaly feature vector determination module 34, and target anomaly pattern determination module 35 are shown.

[0169] It should be understood that, in the embodiments of this application, the processor 401 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0170] Input device 402 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 403 may include a display (LCD, etc.), a speaker, etc.

[0171] The memory 404 may include read-only memory and random access memory, and provides instructions and data to the processor 401. A portion of the memory 404 may also include non-volatile random access memory. For example, the memory 404 may also store information such as multi-dimensional electrical parameters of the low-voltage lines in the distribution area, the feature vectors corresponding to the multi-dimensional electrical parameters, and the anomaly scores corresponding to the feature vectors.

[0172] In a specific implementation, the processor 401, the input device 402, and the output device 403 described in the embodiments of the present application can perform the implementation manners of the abnormality identification method based on the intelligent fusion terminal provided by the embodiments of the present application, and can also perform the implementation manners of the electronic device described in the embodiments of the present application, which will not be described here.

[0173] In another embodiment of the present application, a computer readable storage medium is provided, which stores a computer program. The computer program includes program instructions, which are executed by a processor to implement all or part of the processes of the above-mentioned embodiments. The computer program can also instruct related hardware to complete the implementation by the computer program. The computer program can be stored in a computer readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0174] The computer readable storage medium can be an internal storage unit of the electronic device of any of the above-mentioned embodiments, such as a hard disk or a memory of the electronic device. The computer readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer readable storage medium can include both the internal storage unit and the external storage device of the electronic device. The computer readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0175] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device and the unit described above can refer to the corresponding process in the foregoing method embodiment, and will not be repeated here.

[0176] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device and the unit described above can refer to the corresponding process in the foregoing method embodiment, and will not be repeated here.

[0177] In several embodiments provided in the present application, it should be understood that the disclosed electronic device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the module / unit is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules, units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces or modules, and can also be electrical, mechanical or other form of connection.

[0178] The module / unit described as a separate component can be or can not be physically separated, and the component displayed as a module / unit can be or can not be a physical module / unit, that is, can be located in one place, or can be distributed to a plurality of network modules / units. Part or all of the modules / units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0179] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present separately, or two or more modules can be integrated in one unit. The integrated module / unit can be realized in the form of hardware or software functional module / unit.

[0180] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any skilled person in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An anomaly identification method based on an intelligent fusion terminal, characterized in that, include: The system acquires multi-dimensional electrical parameters under the current operating conditions. These multi-dimensional electrical parameters include first sampling data and second sampling data. The first sampling data and the second sampling data are data collected from the low-voltage lines in the transformer area under the current operating conditions. The first sampling data includes steady-state voltage data, steady-state current data, power data, and harmonic data. The second sampling data includes instantaneous current values ​​and instantaneous voltage values. Based on the multi-dimensional electrical parameters and through a pre-trained dual-channel feature learning network model, the feature vectors corresponding to the multi-dimensional electrical parameters are determined. The feature vectors are used to characterize the operating status of the low-voltage lines in the distribution area. The feature vector is subjected to a predetermined number of forward propagation processes based on a pre-trained anomaly scoring model to obtain an anomaly score array. The standard deviation of each anomaly score in the anomaly score array is taken as the uncertainty score, and the arithmetic mean of each anomaly score in the anomaly score array is taken as the anomaly score of the feature vector. The uncertainty score is used to characterize the reliability of the anomaly score array output by the anomaly scoring model, and the anomaly score is used to characterize the difference between the multidimensional electrical parameters and the normal multidimensional electrical parameters corresponding to the normal mode. If the abnormal score is higher than a preset score threshold and the uncertainty score is lower than a preset standard score, then the feature vector is taken as an abnormal feature vector. The target anomaly pattern of the low-voltage line in the transformer area is determined based on the anomaly feature vector and the anomaly feature library, wherein the anomaly feature library is used to characterize the correspondence between the anomaly feature vector and the anomaly pattern.

2. The anomaly identification method based on intelligent fusion terminal as described in claim 1, characterized in that, The step of determining the feature vectors corresponding to the multi-dimensional electrical parameters based on the multi-dimensional electrical parameters and through a pre-trained dual-channel feature learning network model includes: The first sampled data is input into the first channel of a pre-trained dual-channel feature learning network model to obtain the first feature vector. The second sampled data is input into the second channel of a pre-trained dual-channel feature learning network model to obtain the second feature vector. The first feature vector and the second feature vector are concatenated and fused to obtain the feature vector corresponding to the multi-dimensional electrical parameters.

3. The anomaly identification method based on intelligent fusion terminal as described in claim 1, characterized in that, The step of determining the target anomaly pattern of the low-voltage line in the transformer area based on the anomaly feature vector and the anomaly feature database includes: Calculate the cosine similarity value between the abnormal feature vector and each feature vector in the abnormal feature library, and determine the cosine similarity value group; Select the two cosine similarity values ​​with the highest cosine similarity values ​​from the cosine similarity value group. The two cosine similarity values ​​with the highest cosine similarity values ​​include a first similarity value and a second similarity value, wherein the first similarity value is greater than the second similarity value. If it is determined that the absolute value of the difference between the two cosine similarity values ​​is higher than the preset similarity threshold, then the abnormal pattern corresponding to the first similarity value is taken as the target abnormal pattern of the low-voltage line in the transformer area. If the absolute value of the difference between the two cosine similarity values ​​is not higher than the preset similarity threshold, then the cosine similarity values ​​in the cosine similarity value group are arranged in descending order, and the abnormal patterns corresponding to the first preset number of cosine similarity values ​​are determined. A candidate abnormal pattern set is formed according to each abnormal pattern, and the target abnormal pattern of the low-voltage line in the transformer area is determined according to the candidate abnormal pattern set.

4. The anomaly identification method based on intelligent fusion terminal as described in claim 3, characterized in that, Determining the target anomaly pattern of the low-voltage line in the distribution area based on the candidate anomaly pattern set includes: Based on the feature vectors corresponding to each anomaly pattern in the candidate anomaly pattern set, a feature vector set is determined, which is a set composed of the feature vectors corresponding to all anomaly patterns in the candidate anomaly pattern set. The feature vectors in the feature vector set are clustered according to the clustering algorithm to obtain the clusters corresponding to each feature vector. Each feature vector is mapped to the clusters corresponding to the candidate anomaly pattern set to obtain anomaly pattern groups with clustering labels. For each group of abnormal patterns, determine the abnormal representative pattern of the abnormal patterns in that group to obtain the abnormal representative pattern corresponding to each group. Obtain the intra-group consistency score of each abnormal pattern in the abnormal pattern group, and the inter-group separation score of each abnormal pattern in the abnormal pattern group. Based on the intragroup consistency score, the intergroup separation score, and the confidence formula, a confidence value is determined for each representative anomaly pattern. The confidence formula is as follows: Where c represents the confidence value of each anomaly pattern. The score represents the consistency score within the group. The intergroup separation score. The weights corresponding to the intragroup consistency scores. The weights corresponding to the between-group separation scores; The target anomaly mode of the low-voltage line in the distribution area is determined based on the confidence value and the preset confidence threshold.

5. The anomaly identification method based on intelligent fusion terminal as described in claim 4, characterized in that, The acquisition of the intra-group consistency score for each abnormal pattern group within the abnormal pattern group, and the inter-group separation score for each abnormal pattern group within the abnormal pattern group, includes: The average distance within each abnormal pattern is determined based on the distance between each feature vector in each abnormal pattern and the feature vector of the representative abnormal pattern. The average distance is converted into an intragroup consistency score according to the consistency transformation formula, which is: ,in, The score represents the consistency score within the group. Scaling factor This represents the normalized average distance within the group. Determine the minimum distance between the feature vector of the representative anomalous pattern in each group of anomalous patterns and the feature vectors of other representative anomalous patterns; The minimum distance is converted into a between-group separation score according to the separation conversion formula, which is: ,in, The intergroup separation score. This is the separation scaling factor. It represents the minimum distance between the feature vectors of the anomaly representative pattern and the feature vectors of other anomaly representative patterns.

6. The anomaly identification method based on intelligent fusion terminal as described in claim 1, characterized in that, The method further includes: If the abnormal score is not higher than a preset score threshold, then the abnormal score is converted into a health index according to the health index conversion formula, which is as follows: ,in, H For health index, h These are abnormal scores. The preset score threshold; Obtain the historical health index sequence and its time series of the low-voltage lines in the transformer area within a preset historical time period, wherein the time series corresponds one-to-one with the health index of the historical time period; The time series is converted into time values ​​to form a time value series; Based on the historical health index sequence, the time value sequence, and the linear regression equation, determine the linear relationship between the health index and the time value; The slope is determined based on the linear relationship described above; The health level of the low-voltage lines in the distribution area is determined based on the health index and the slope.

7. The anomaly identification method based on an intelligent fusion terminal as described in claim 6, characterized in that, The process of determining the health level of the low-voltage lines in the distribution area based on the health index and the slope includes: If the health index is greater than a preset first health threshold and the slope is greater than a preset first slope threshold, then the health level is determined to be the first health level. If the health index is less than or equal to a preset first health threshold and greater than a preset second health threshold, or if the slope is less than or equal to a preset second slope threshold, then the health level is determined to be the second health level. If the health index is less than or equal to a preset second health threshold, the health level is determined to be the third health level, wherein the health status of the first health level, the second health level, and the third health level decreases sequentially.

8. An anomaly identification device based on an intelligent fusion terminal, characterized in that, include: The data acquisition module is used to acquire multi-dimensional electrical parameters under the current operating conditions. The multi-dimensional electrical parameters include first sampling data and second sampling data. The first sampling data and the second sampling data are data of the low-voltage line in the transformer area collected under the current operating conditions. The first sampling data includes steady-state voltage data, steady-state current data, power data and harmonic data. The second sampling data includes instantaneous current value and instantaneous voltage value. The feature vector determination module is used to determine the feature vectors corresponding to the multi-dimensional electrical parameters based on the multi-dimensional electrical parameters and through a pre-trained dual-channel feature learning network model. The feature vectors are used to characterize the operating status of the low-voltage lines in the distribution area. An anomaly score determination module is used to perform forward propagation processing on the feature vector a preset number of times according to a pre-trained anomaly scoring model to obtain an anomaly score array. The standard deviation of each anomaly score in the anomaly score array is used as the uncertainty score, and the arithmetic mean of each anomaly score in the anomaly score array is used as the anomaly score of the feature vector. The uncertainty score is used to characterize the reliability of the anomaly score group output by the anomaly scoring model, and the anomaly score is used to characterize the difference between the multi-dimensional electrical parameters and the normal multi-dimensional electrical parameters corresponding to the normal mode. The abnormal feature vector determination module is used to identify the feature vector as an abnormal feature vector when the abnormal score is higher than a preset score threshold and the uncertainty score is lower than a preset standard score. The target anomaly pattern determination module is used to determine the target anomaly pattern of the low-voltage line in the distribution area based on the anomaly feature vector and the anomaly feature library, wherein the anomaly feature library is used to characterize the correspondence between the anomaly feature vector and the anomaly pattern.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images