An edge-computing-based meter terminal group fault detection method and system

Abstract

The present application relates to the field of fault detection, in particular to a kind of electric meter terminal group fault detection method and system based on edge computing, comprising: obtaining the current data sequence of each electric meter terminal accessed to the same phase in the preset time period, any current data sequence is regarded as target sequence, the window sequence corresponding to the maximum value of stability index is intercepted as the candidate sequence of target sequence, and the candidate sequence of each current data sequence is obtained by traversal;The clustering cluster corresponding to each sequence set is obtained, any clustering cluster is regarded as target cluster, the weighted group similarity of each candidate sequence in target cluster is calculated, and the candidate sequence with the weighted group similarity less than the preset threshold is regarded as potential abnormal sequence and carries out fault alarm.The present application can improve the accuracy of electric meter terminal group fault detection.

Description

Technical Field

[0001] This invention relates to the field of fault detection. More specifically, this invention relates to a method and system for fault detection of meter terminal groups based on edge computing. Background Technology

[0002] The meter terminals are the key interfaces connecting the electricity meter to external lines (such as incoming and outgoing lines, current transformers, etc.). An electricity meter typically has multiple meter terminals, which together form a meter terminal group. Meter terminal groups are prone to problems such as poor contact and overload. If these problems are not detected in time, they may lead to safety accidents such as insulation aging and fires, resulting in serious losses.

[0003] There are two commonly used fault detection technologies: one is to use a uniform and fixed anomaly judgment threshold, but the electricity load patterns of users corresponding to different meter terminals are different, and this method is difficult to adapt to such differences; the other is to judge anomalies by analyzing the consistency of electricity consumption data of different users, but the electricity consumption behavior of different users varies greatly, and the global comparison method is prone to high false positive and false negative rates. In short, both existing fault detection technologies have the problem of low detection accuracy. Summary of the Invention

[0004] The main objective of this application is to propose a method and system for detecting faults in meter terminal groups based on edge computing, aiming to improve the accuracy of fault detection in meter terminal groups.

[0005] To achieve the above objectives, an embodiment of the first aspect of this application proposes a fault detection method for meter terminal groups based on edge computing. The method includes: acquiring current data sequences of each meter terminal connected to the same phase within a preset time period; using any current data sequence as a target sequence; setting a collection window on the target sequence and anchoring the right boundary of the collection window to the end of the target sequence; expanding the collection window to the left according to a preset length interval; calculating the stability index of the window sequence corresponding to each length value within the length interval; selecting the window sequence corresponding to the maximum value of the stability index as a candidate sequence of the target sequence; and iterating through and acquiring the current data sequence of each current data sequence. Candidate sequences: Based on sequence length, all candidate sequences are classified into multiple sequence sets. Any sequence set is used as the target set. A clustering algorithm is performed on all candidate sequences in the target set, and the dynamic weight of each candidate sequence in the target set as the cluster center is calculated in each clustering iteration. The iteration terminates to obtain the clusters corresponding to the target set. The clusters corresponding to each sequence set are obtained by traversing. Any cluster is used as the target cluster. The weighted intra-group similarity of each candidate sequence in the target cluster is calculated based on the dynamic weight of the last iteration. Candidate sequences with intra-group similarity less than the similarity threshold are considered potential abnormal sequences. The potential abnormal sequences of each cluster are obtained by traversing and fault alarms are generated.

[0006] In some embodiments, calculating the stability index of the window sequence corresponding to each length value within the length interval of the acquisition window includes: taking the window sequence corresponding to any length value within the length interval as a first sequence, calculating the coefficient of variation of the first sequence, dividing the first sequence equally according to a preset unit to obtain multiple subsequences, calculating the similarity between each pair of adjacent subsequences and calculating the mean of all similarities, calculating the stability index of the first sequence based on the coefficient of variation and the mean of all similarities; and traversing to obtain the stability index of the window sequence corresponding to each length value.

[0007] In some embodiments, the method further includes: calculating the information entropy of each candidate sequence in the target set; taking any candidate sequence in the target set as a second sequence; taking the ratio of the information entropy of the second sequence to the mean of all information entropies as an influence factor; calculating the downsampling ratio of the second sequence based on the influence factor and the stability index of the second sequence; traversing to obtain the downsampling ratio of each candidate sequence in the target set; and performing downsampling processing on the candidate sequences according to the downsampling ratio.

[0008] In some embodiments, calculating the dynamic weight of each candidate sequence in the target set as a cluster center in each clustering iteration includes: randomly selecting a preset number of candidate sequences in the target set as initial cluster centers for the first clustering iteration; taking any candidate sequence in the target set as a third sequence; calculating the initial distance between the third sequence and each initial cluster center; calculating the scale parameter of the third sequence in the first clustering iteration; calculating the dynamic weight of the third sequence as a cluster center in the next clustering iteration based on the stability index of the third sequence, the ratio of the minimum initial distance to the scale parameter; and iterating through the target set to obtain the dynamic weight of each candidate sequence as a cluster center in each clustering iteration.

[0009] In some embodiments, calculating the scaling parameter of the third sequence in the initial clustering iteration includes: taking the cluster to which the third sequence belongs in the initial clustering iteration as the reference cluster, calculating the standard deviation of the stability index of all sequences in the reference cluster; obtaining the median of the distance between all sequences in the reference cluster and the cluster center of the reference cluster; and using the product of the standard deviation and the median as the scaling parameter of the third sequence in the initial clustering iteration.

[0010] In some embodiments, calculating the weighted intra-group similarity of each candidate sequence in the target cluster includes: calculating the confidence score of each candidate sequence in the target cluster; taking any sequence in the target cluster as the fourth sequence and any sequence in the target cluster other than the fourth sequence as the fifth sequence; multiplying the confidence score of the fourth sequence, the confidence score of the fifth sequence, and the normalized sequence distance between the fourth and fifth sequences to obtain the weighted distance between the fourth and fifth sequences; iterating through the target cluster to obtain the weighted distance between the fourth sequence and each sequence in the target cluster other than the fourth sequence; summing all weighted distances and normalizing the summed result to obtain the weighted intra-group similarity of the fourth sequence; and iterating through the target cluster to obtain the weighted intra-group similarity of each candidate sequence.

[0011] In some embodiments, calculating the confidence of each candidate sequence in the target cluster includes: calculating the target distance between each candidate sequence in the target cluster and the cluster center of the target cluster, multiplying the target distance of the fourth sequence by the ratio of the target distance of the fourth sequence to the mean of all the target distances, the stability index of the fourth sequence, and the dynamic weight corresponding to the last iteration of the fourth sequence to obtain the confidence of the fourth sequence; and traversing to obtain the confidence of each candidate sequence in the target cluster.

[0012] An embodiment of the second aspect of this application proposes a fault detection system for meter terminal groups based on edge computing. The system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-described fault detection method for meter terminal groups based on edge computing is implemented.

[0013] The beneficial effects of this invention are:

[0014] This invention first acquires the current data sequences of each meter terminal in the same phase, establishing a commonality analysis basis for the invention. Next, a stability index is introduced, adaptively extracting candidate sequences with the goal of maximizing the stability index. This ensures that the candidate sequences reflect the most stable operating characteristics of the meter terminals, avoiding feature distortion caused by a fixed window length and providing a reliable data foundation for subsequent analysis. Second, sequences with the same stable operating duration are grouped into the same sequence set based on sequence length. Then, sequences with the most stable characteristics and the same stable duration are clustered, further dividing the sequences based on current sequence morphology (user electricity consumption behavior). Dynamic weights are introduced during clustering, allowing sequences with high stability and strong representativeness to play a greater role in clustering. Finally, weighted intra-group similarity is calculated using dynamic weights to measure the degree of deviation of the sequence from the normal pattern within the cluster. Sequences with large deviations are identified as potentially abnormal sequences, triggering alarms to remind staff to promptly investigate and prevent faults. Compared to simple threshold judgment and simple consistency analysis, this invention considers the differences in user electricity consumption corresponding to different meter terminals, improving the accuracy of fault detection for meter terminal groups. Attached Figure Description

[0015] Figure 1 This is a flowchart of steps S1-S3 in an edge computing-based method for detecting faults in a meter terminal group according to an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] Reference Figure 1 A method for fault detection of meter terminal groups based on edge computing includes steps S1-S3, as follows:

[0019] Step S1: Obtain the current data sequence of each meter terminal connected to the same phase within a preset time period. Take any current data sequence as the target sequence, set a collection window on the target sequence and anchor the right boundary of the collection window to the end of the target sequence. Expand the collection window to the left according to the preset length interval, and calculate the stability index of the window sequence corresponding to each length value within the length interval. Extract the window sequence corresponding to the maximum value of the stability index as the candidate sequence of the target sequence. Iterate through and obtain the candidate sequence of each current data sequence.

[0020] It should be noted that shops and offices on the same floor or in the same area are often connected to the same phase; similar production equipment (such as machine tools and assembly lines) in industrial plants are also often connected to the same phase; and the terminals of meters connected to the same phase form a meter terminal group.

[0021] It should be noted that the preset time period is Heaven. The current data sequence of each meter terminal is used express.

[0022] It should be noted that the end of the target sequence is the latest current data, representing the latest power consumption status. The right boundary of the acquisition window is anchored at the end of the target sequence to always focus on the latest power consumption status, so as to ensure the timeliness and relevance of fault detection.

[0023] It should be noted that the upper limit of the length interval is the preset time period, i.e., 30 days. To ensure the normal operation of the stability index calculation, the lower limit of the length interval is 3 days. The length interval is [3, 30]. Due to the natural daily cycle of user electricity consumption (e.g., low electricity consumption during the day and high electricity consumption at night, and fixed operating periods for industrial equipment), the length value of the length interval is in days and is rounded to an integer. The length value is... express, At 3 o'clock, the length of the window sequence is 3 days.

[0024] In some embodiments, calculating the stability index of the window sequence corresponding to each length value within the length interval of the acquisition window includes: taking the window sequence corresponding to any length value within the length interval as a first sequence, calculating the coefficient of variation of the first sequence, dividing the first sequence equally according to a preset unit to obtain multiple subsequences, calculating the similarity between each pair of adjacent subsequences and calculating the mean of all similarities, calculating the stability index of the first sequence based on the coefficient of variation and the mean of all similarities; and traversing to obtain the stability index of the window sequence corresponding to each length value.

[0025] It should be noted that the preset unit is days. For example, if the length of the window sequence is 3 days, the window sequence is divided into 3 subsequences by days. The similarity between any two adjacent subsequences is calculated, and the two similarities are averaged. Then, the stability index of the window sequence is calculated by combining the coefficient of variation of the 3-day window sequence.

[0026] For example, the formula for calculating the stability index is as follows:

[0027]

[0028] In the formula middle, For the first The length value corresponding to each meter terminal is The stability index of the first sequence; For An exponential function with the base as its base.

[0029] In the formula middle, For the first The length value corresponding to each meter terminal is The coefficient of variation (CV) of the first sequence is calculated using existing techniques and will not be described in detail here.

[0030] In the formula middle, This is the exponential function value of the coefficient of variation of the first sequence. Coefficient of variation. Able to reflect the first The degree of fluctuation of current data in the first sequence (window sequence) corresponding to each meter terminal. The larger the coefficient of variation, the more drastic the fluctuation of current data. Therefore, the power consumption stability of the user corresponding to the meter terminal is poor under this scale.

[0031] In the formula middle, For the first The length value corresponding to each meter terminal is The Pearson correlation coefficient between adjacent subsequences of the first sequence is used to calculate the similarity between adjacent subsequences. This is the mean of all similarities.

[0032] Since the Pearson correlation coefficient ranges from -1 to 1, in order to ensure that the stability index is positive, the following is used in the calculation process: The mean is normalized. The more regular the electricity consumption behavior of the branch corresponding to the meter terminal, the larger the Pearson correlation coefficient, and the better the electricity consumption stability of the meter terminal.

[0033] It should be further explained that the stability index comprehensively reflects the intensity of current fluctuations within the window and the similarity between adjacent daily electricity consumption curves. If the window is too short, noise will dominate and stability will be low; if the window is too long, local fluctuations will be smoothed out and correlation will decrease. Therefore, the window corresponding to the highest stability index achieves a balance between temporal resolution and statistical stability. This lays the foundation for subsequent clustering and fault detection in this invention.

[0034] Step S2: Classify all candidate sequences according to sequence length to obtain multiple sequence sets. Take any sequence set as the target set, perform a clustering algorithm on all candidate sequences in the target set, and calculate the dynamic weight of each candidate sequence in the target set as the cluster center in each clustering iteration. When the iteration terminates, the cluster corresponding to the target set is obtained. Iterate through each sequence set to obtain the cluster corresponding to each sequence set.

[0035] It should be further noted that the present invention uses This represents the sequence length of a candidate sequence (the window sequence corresponding to the maximum value of the stability index). All candidate sequences in a sequence set have the same length.

[0036] In some embodiments, the method further includes: calculating the information entropy of each candidate sequence in the target set, taking any candidate sequence in the target set as a second sequence, taking the ratio of the information entropy of the second sequence to the mean of all information entropies as an influence factor, calculating the downsampling ratio of the second sequence based on the influence factor and the stability index of the second sequence, traversing to obtain the downsampling ratio of each candidate sequence in the target set, and performing downsampling processing on the candidate sequences according to the downsampling ratio.

[0037] It should be noted that, in order to reduce the complexity of subsequent analysis and calculation and further improve the sensitivity to key sections of changes in the operating status of the meter terminals, this invention performs downsampling processing on all candidate sequences of the target set before performing clustering algorithm on all candidate sequences of the target set.

[0038] For example, the formula for calculating the downsampling ratio is as follows:

[0039]

[0040] In the formula middle, For the first The downsampling ratio of the candidate sequence corresponding to each meter terminal.

[0041] In the formula middle, For the first The stability index of the candidate sequence corresponding to each meter terminal is determined by the formula. It can be seen that, The smaller the value, the more drastic the fluctuations in the current data of the corresponding candidate sequence. The larger the value, the more sampling points are retained for subsequent analysis.

[0042] In the formula middle, For the first The information entropy of the candidate sequence corresponding to each meter terminal reflects the uncertainty and complexity of the candidate sequence; For the first The mean of the information entropy of all candidate sequences in the sequence set to which the candidate sequence corresponding to each meter terminal belongs reflects the average level of current data of all meter terminals at the same time scale. The impact factor is the highest level of influence; a larger impact factor reflects the highest level of influence. The candidate sequence corresponding to each meter terminal is more complex than other candidate sequences of the same length, and more sampling points need to be retained in subsequent analysis.

[0043] It should be further explained that the stability index of the quantified meter terminal current data sequence within different length windows is adaptively selected, and the window with the highest stability index is selected to suppress the influence of noise while maintaining time sensitivity. The introduction of features based on stability index to downsample the original data sequence can effectively reduce the amount of data within the window without significantly losing the integrity of data features. This allows subsequent analysis to focus more on key sections reflecting changes in the terminal operating status, thereby improving the sensitivity to identify slight fluctuations, poor contact, and phase-to-phase current deviations in the terminal group.

[0044] In some embodiments, calculating the dynamic weight of each candidate sequence in the target set as a cluster center in each clustering iteration includes: randomly selecting a preset number of candidate sequences in the target set as initial cluster centers for the first clustering iteration; taking any candidate sequence in the target set as a third sequence; calculating the initial distance between the third sequence and each initial cluster center; calculating the scale parameter of the third sequence in the first clustering iteration; calculating the dynamic weight of the third sequence as a cluster center in the next clustering iteration based on the stability index of the third sequence, the ratio of the minimum initial distance to the scale parameter; and iterating through the target set to obtain the dynamic weight of each candidate sequence as a cluster center in each clustering iteration.

[0045] It should be noted that the preset quantity is determined by the elbow method.

[0046] For example, the formula for calculating the dynamic weight of a candidate sequence as a cluster center is as follows:

[0047]

[0048] In the formula middle, For the first The candidate sequence corresponding to each meter terminal is in the first The dynamic weights that determine which cluster centers are selected in each clustering iteration are higher than those that indicate a greater likelihood of being selected as a new cluster center.

[0049] In the formula middle, For the first The stability index of the candidate sequence corresponding to the first meter terminal; the higher the stability, the better. The candidate sequences exhibit smaller fluctuations in electricity consumption and more stable trends, making them more reliable as cluster centers for subsequent intra-group consistency calculations.

[0050] In the formula middle, For An exponential function with the base as its base.

[0051] In the formula middle, For the first Candidate sequences corresponding to each meter terminal.

[0052] In the formula middle, for In the The index of the cluster center to which the cluster belongs in the next clustering iteration; For the first The index in the next clustering iteration is Candidate sequences corresponding to cluster centers.

[0053] In the formula middle, express and The DTW (Dynamic Time Warping) distance between them, i.e., the first... The candidate sequence corresponding to each meter terminal is in the first The closer the distance between the cluster center of the corresponding cluster in the first clustering iteration, the better. The closer a candidate sequence is to the cluster center, the greater the likelihood that it will be selected as the new cluster center.

[0054] In the formula middle, For the first The scaling parameter in the next clustering iteration controls the sensitivity of the minimum DTW distance to the decay of dynamic weights. Fixing the parameter can easily lead to overfitting noise in small clusters, ignoring boundary points in large clusters, and amplifying interference from unstable sequences. Therefore, it needs to be adaptively adjusted during the iteration process. .

[0055] In some embodiments, calculating the scaling parameter of the third sequence in the initial clustering iteration includes: taking the cluster to which the third sequence belongs in the initial clustering iteration as the reference cluster, calculating the standard deviation of the stability index of all sequences in the reference cluster; obtaining the median distance between all sequences in the reference cluster and the cluster center of the reference cluster; and using the product of the standard deviation and the median as the scaling parameter of the third sequence in the initial clustering iteration.

[0056] For example, The calculation formula is as follows:

[0057]

[0058] In the formula middle, For the first The scaling parameter in the next clustering iteration.

[0059] In the formula middle, for In the The index of the cluster center to which the cluster belongs in the next clustering iteration. For the first The index in the next clustering iteration is The clusters corresponding to the cluster centers, for The Middle Stability metrics for candidate sequences for The standard deviation of all corresponding stability indices; the larger the standard deviation, the stronger the stability. The stability of the candidate sequences varies greatly, so the formula... The slower the DTW distance decays the dynamic weights, the more the influence of points farther from the current center point is amplified, promoting the cluster center to converge towards more stable distant points; conversely, the smaller the standard deviation, the more the weights are concentrated on nearby points.

[0060] In the formula middle, It is a median function. for The Middle Candidate sequences, For the first The index in the next clustering iteration is The candidate sequences corresponding to the cluster centers, for and The distance between them, through Take the median of all distances, and then apply the median to the formula. In Normalization.

[0061] It should be noted that during the clustering iteration process, the dynamic weights of all candidate sequences relative to their respective clusters are calculated, new cluster centers are selected, non-cluster centers are re-clustered, and the above steps are repeated until the cluster centers no longer change, the clustering ends, and the clustering results are output. Subsequent analysis is then performed based on the clustering results.

[0062] It should be further explained that this invention uses the DTW algorithm to calculate the shortest path distance between the candidate sequence corresponding to the meter terminal and the center of its cluster, and combines the stability index of the candidate sequence corresponding to the meter terminal to improve the method of iterative cluster center in the traditional k-means clustering algorithm, so that the cluster center converges to the stable point, which can improve the consistency within the group and reduce the interference of irregular meter terminals.

[0063] It should be noted that after the clustering iteration is completed, the clusters corresponding to each sequence set are obtained, and the clusters corresponding to each sequence set are all... indivual.

[0064] Step S3: Take any cluster as the target cluster, calculate the weighted intra-group similarity of each candidate sequence in the target cluster based on the dynamic weight of the last iteration, and take the sequence with intra-group similarity less than the similarity threshold as the potential abnormal sequence. Iterate through each cluster to obtain the potential abnormal sequence and issue a fault alarm.

[0065] In some embodiments, calculating the weighted intra-group similarity of each candidate sequence in the target cluster includes: calculating the confidence score of each candidate sequence in the target cluster; taking any sequence in the target cluster as the fourth sequence and any sequence in the target cluster other than the fourth sequence as the fifth sequence; multiplying the confidence score of the fourth sequence, the confidence score of the fifth sequence, and the normalized sequence distance between the fourth and fifth sequences to obtain the weighted distance between the fourth and fifth sequences; iterating through the target cluster to obtain the weighted distance between the fourth sequence and each sequence in the target cluster other than the fourth sequence; summing all weighted distances and normalizing the summed result to obtain the weighted intra-group similarity of the fourth sequence; and iterating through the target cluster to obtain the weighted intra-group similarity of each candidate sequence.

[0066] In some embodiments, calculating the confidence of each candidate sequence in the target cluster includes: calculating the target distance between each candidate sequence in the target cluster and the cluster center of the target cluster, multiplying the target distance of the fourth sequence by the ratio of the target distance of the fourth sequence to the mean of all target distances, the stability index of the fourth sequence, and the dynamic weight corresponding to the last iteration of the fourth sequence to obtain the confidence of the fourth sequence; and traversing to obtain the confidence of each candidate sequence in the target cluster.

[0067] For example, the confidence level is calculated using the following formula:

[0068]

[0069] In the formula middle, For the first The confidence level of the candidate sequence corresponding to each meter terminal.

[0070] In the formula middle, No. The stability index of the candidate sequence corresponding to each meter terminal is used. The higher the stability, the more regular the electricity consumption behavior of the corresponding meter terminal is. The similarity is calculated with other candidate sequences in the same cluster to quantify the degree of anomaly of the corresponding meter terminal. The confidence weight is strengthened to avoid interference from the current data of meter terminals with irregular electricity consumption behavior and improve the reliability of anomaly identification.

[0071] In the formula middle, For the first The candidate sequence corresponding to each meter terminal is selected as the dynamic weight of the cluster center in the last clustering iteration. The larger the weight, the higher the stability of the corresponding candidate sequence and the closer it is to the cluster center, and the higher the confidence level. For the first Candidate sequences corresponding to each meter terminal. for The index of the cluster center to which the cluster belongs in the last clustering iteration; The index in the last clustering iteration is Candidate sequences corresponding to cluster centers. express and The DTW distance between them, that is The distance between it and the cluster center of its own cluster (cluster) in the last clustering iteration. The index in the last clustering iteration is The clusters corresponding to the cluster centers, i.e. The cluster to which it belongs. for The Middle 10 candidate sequences. for and DTW distance. Used to represent The mean of all corresponding DTW distances.

[0072] Able to reflect the first The closer a candidate sequence corresponding to a meter terminal is to its cluster center (cluster center), the higher the confidence level.

[0073] For example, the formula for calculating weighted intra-group similarity is as follows:

[0074]

[0075] In the formula middle, For the first The weighted intra-group similarity of candidate sequences corresponding to each meter terminal represents the degree of consistency between the meter terminal and other meter terminals in the same group (within the cluster), and is used to assess whether it deviates from the overall trend of the group.

[0076] In the formula middle, For the first The confidence level of the candidate sequence corresponding to each meter terminal. The larger the value, the higher the stability of the corresponding meter terminal current data, the smaller the current fluctuation and the more stable the trend, and the higher the confidence weight. The index in the last clustering iteration is The clusters corresponding to the cluster centers. for The Middle The confidence level of each candidate sequence. Reflecting the The importance of each candidate sequence in the assessment of intragroup similarity.

[0077] In the formula middle, For the first The candidate sequence corresponding to the first meter terminal and the first The distance between candidate sequences.

[0078] For normalization function, This is the normalized distance. The shorter the distance, the smaller the difference in current change patterns between the two candidate sequences, and the higher the similarity.

[0079] In the formula middle, For the first The candidate sequence corresponding to the first meter terminal and the first Weighted distance between candidate sequences; formula The numerator is the sum of all weighted distances, which is normalized using the sum of the denominators.

[0080] It should be further explained that by introducing a confidence weighting mechanism, the stability differences of candidate sequences within the same cluster are fully utilized, giving higher confidence meter terminals a higher weight in similarity calculation. This reduces the interference of abnormal or unstable meter terminals on the results, enabling more accurate capture of deviations in single-terminal current characteristics within meter terminal groups. This improves detection accuracy and provides a reliable basis for real-time alarms and maintenance decisions of subsequent edge computing nodes.

[0081] It should be noted that the similarity threshold is a hyperparameter and can be selected based on actual conditions or experience. Sequences with similarity less than the similarity threshold within the weighted group are considered potential abnormal sequences. The meter terminal information corresponding to these potential abnormal sequences is uploaded to the control center so that staff can promptly troubleshoot the fault.

[0082] This invention also provides a fault detection system for meter terminal groups based on edge computing. The system includes a processor and a memory, the memory storing computer program instructions. When the computer program instructions are executed by the processor, they implement a fault detection method for meter terminal groups based on edge computing according to the first aspect of this invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface; their configuration and functions are known in the art and will not be described further here.

[0083] It should be noted that the preferred embodiments of this application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of this application. For those skilled in the art, various modifications and improvements can be made without departing from the concept of the invention, and these all fall within the protection scope of the invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. An edge computing based method for detecting a terminal group fault of an electric meter, characterized in that, include: Acquire the current data sequence of each meter terminal connected to the same phase within a preset time period. Take any current data sequence as the target sequence, set a collection window on the target sequence and anchor the right boundary of the collection window to the end of the target sequence. Expand the collection window to the left according to the preset length interval, and calculate the stability index of the window sequence corresponding to each length value within the length interval. Extract the window sequence corresponding to the maximum value of the stability index as the candidate sequence of the target sequence. Iterate to obtain the candidate sequence of each current data sequence. Multiple sequence sets are obtained by classifying all candidate sequences according to their length. Any sequence set is used as the target set. A clustering algorithm is performed on all candidate sequences in the target set, and the dynamic weight of each candidate sequence in the target set as the cluster center is calculated in each clustering iteration. The iteration terminates to obtain the cluster corresponding to the target set. The cluster corresponding to each sequence set is obtained by traversing. Take any cluster as the target cluster, calculate the weighted intra-group similarity of each candidate sequence in the target cluster based on the dynamic weight of the last iteration, and take the candidate sequences with intra-group similarity less than the similarity threshold as potential abnormal sequences. Iterate through each cluster to obtain potential abnormal sequences and issue fault alarms. The stability indices for calculating the window sequence corresponding to each length value within the length interval of the acquisition window include: Take the window sequence corresponding to any length value within the length range as the first sequence, calculate the coefficient of variation of the first sequence, divide the first sequence into multiple subsequences according to the preset unit, calculate the similarity between each pair of adjacent subsequences and calculate the mean of all similarities, and calculate the stability index of the first sequence based on the coefficient of variation and the mean of all similarities. Iterate through the window sequence corresponding to each length value to obtain the stability index; The dynamic weights for each candidate sequence in the target set to be selected as a cluster center in each clustering iteration include: A predetermined number of candidate sequences are randomly selected from the target set as initial cluster centers for the first clustering iteration. Any candidate sequence in the target set is taken as the third sequence. The initial distance between the third sequence and each initial cluster center is calculated. The scale parameter of the third sequence in the first clustering iteration is calculated. The dynamic weight of the third sequence being selected as a cluster center in the next clustering iteration is calculated based on the stability index of the third sequence, the ratio of the minimum initial distance to the scale parameter. The dynamic weights of each candidate sequence in the target set that are selected as cluster centers in each clustering iteration are obtained by traversing the set.

2. The edge computing based terminal group fault detection method of claim 1, wherein, Also includes: Calculate the information entropy of each candidate sequence in the target set, take any candidate sequence in the target set as the second sequence, take the ratio of the information entropy of the second sequence to the mean of all information entropies as the influence factor, calculate the downsampling ratio of the second sequence based on the influence factor and the stability index of the second sequence, and iterate to obtain the downsampling ratio of each candidate sequence in the target set. The candidate sequences are downsampled according to the downsampling ratio.

3. The edge computing based terminal group fault detection method of claim 1, wherein, Calculating the scaling parameters of the third sequence in the initial clustering iteration includes: The cluster to which the third sequence belongs in the first clustering iteration is used as the reference cluster, and the standard deviation of the stability index of all sequences in the reference cluster is calculated. Obtain the median distance between all sequences in the reference cluster and the cluster center of the reference cluster; The product of the standard deviation and the median is used as the scaling parameter for the third sequence in the first clustering iteration.

4. The method for fault detection of meter terminal groups based on edge computing according to claim 1, characterized in that, Calculating the weighted intra-group similarity of each candidate sequence in the target cluster includes: Calculate the confidence score of each candidate sequence in the target cluster. Take any sequence in the target cluster as the fourth sequence and any sequence in the target cluster other than the fourth sequence as the fifth sequence. Multiply the confidence scores of the fourth and fifth sequences by the normalized sequence distance between the fourth and fifth sequences to obtain the weighted distance between the fourth and fifth sequences. Iterate through the fourth sequence to obtain the weighted distance between the fourth sequence and each sequence in the target cluster other than the fourth sequence. Accumulate all weighted distances and normalize the accumulated result to obtain the weighted intra-group similarity of the fourth sequence. Iterate through the target cluster to obtain the weighted intra-group similarity of each candidate sequence.

5. The method for fault detection of meter terminal groups based on edge computing according to claim 4, characterized in that, Calculating the confidence score for each candidate sequence in the target cluster includes: Calculate the target distance between each candidate sequence in the target cluster and the cluster center of the target cluster, and multiply the result by the ratio of the target distance of the fourth sequence to the mean of all the target distances, the stability index of the fourth sequence, and the dynamic weight corresponding to the last iteration of the fourth sequence to obtain the confidence level of the fourth sequence. Iterate through the target cluster to obtain the confidence score of each candidate sequence.

6. A fault detection system for meter terminal groups based on edge computing, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a method for detecting faults in meter terminal groups based on edge computing according to any one of claims 1-5.

Images