Low-voltage area acquisition frame length dynamic adaptation method, device and terminal based on cluster analysis

Abstract

The application discloses a low-voltage transformer area acquisition frame length dynamic adaptation method and device based on cluster analysis and a terminal, relates to the low-voltage transformer area operation and maintenance technical field, and the scheme provided by the application reduces the cluster model update consumption through the flow incremental clustering mechanism, enables single sample processing to only involve the statistical quantity recursion of a single or a small number of clusters, does not need full recalculation, can continuously track the transformer area characteristic change in a shorter time period, always maintains the timeliness of the hierarchical model, avoids the model deviation accumulation problem caused by the traditional static snapshot method, has the ability of real-time adaptation to the dynamic change of the transformer area load, improves the timeliness and accuracy of the acquisition data, and thus provides a solid foundation for ensuring the accuracy of subsequent line loss calculation.

Description

Technical Field

[0001] This application relates to the field of low-voltage distribution area operation and maintenance technology, and in particular to a method, device and terminal for dynamic adaptation of low-voltage distribution area acquisition frame length based on cluster analysis. Background Technology

[0002] The low-voltage distribution area electricity consumption information collection system is a core infrastructure for power grid companies to achieve refined management of the distribution network and support marketing operations, widely covering public transformer areas, dedicated transformer users, and residential users. The collection terminal, by setting a reasonable collection frame length (i.e., data collection cycle interval), must ensure data integrity and real-time performance while balancing communication channel capacity and terminal power consumption. If the frame length is configured too long (collection frequency too low), it will miss short-term, drastic fluctuations such as sudden changes in distributed photovoltaic output and concentrated charging of electric vehicles, resulting in the collected data failing to reflect the real-time operating status of the distribution area in a timely and detailed manner. If the frame length is configured too short (collection frequency too high), although single-point sampling is dense, the massive amount of data will crowd out the limited communication channel capacity, easily causing data packet loss and transmission delays, thus compromising data integrity and real-time performance.

[0003] The current frame length configuration methods generally adopt an empirical frame length configuration strategy based on transformer area type or periodically use historical batch clustering methods to perform offline clustering on characteristics such as load fluctuation rate, peak ratio, frequency of abnormal events, three-phase imbalance, and penetration rate of new loads in transformer areas. The transformer areas are divided into several demand levels and matched with corresponding frame lengths. The specific process is as follows: maintenance personnel export the historical operation data of all transformer areas from the electricity information collection system every quarter (usually containing daily features of several months or even a year), use classic clustering algorithms such as K-means for offline processing, obtain cluster centers and level divisions, and then issue frame length configuration instructions in batches after manual review.

[0004] However, with the advancement of the construction of new power systems, a large number of new loads such as electric vehicle charging piles and distributed photovoltaics are connected to low-voltage distribution areas, causing the load characteristics of distribution areas to exhibit significant spatiotemporal dynamic evolution characteristics. Under this environment, the effective adaptation cycle of clustering models is constantly shortened, which leads to an increase in the deviation between the classification results and actual needs, and ultimately affects the timeliness of data collection and the accuracy of line loss calculation. Summary of the Invention

[0005] This application provides a method, device, and terminal for dynamic adaptation of data acquisition frame length in low-voltage distribution areas based on cluster analysis, which aims to improve the timeliness and accuracy of data acquisition by adapting to dynamic changes in the load of distribution areas.

[0006] To achieve the above-mentioned objectives, the first aspect of this application provides a method for dynamic adaptation of low-voltage distribution area acquisition frame length, comprising:

[0007] Collect real-time operational data from the transformer substation;

[0008] Based on the real-time operation data of the transformer area, feature extraction is performed in combination with a preset statistical period to obtain the transformer area operation feature vector for the statistical period.

[0009] Based on the historical clusters contained in the historical cluster set, the candidate cluster corresponding to the minimum distance is determined by the Mahalanobis distance between the operating feature vector of the transformer area and the historical cluster. If the Mahalanobis distance between the operating feature vector of the transformer area and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, the operating feature vector of the transformer area is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the operating feature vector of the transformer area. The historical cluster set contains multiple historical clusters, and the historical clusters are obtained by clustering based on the historical transformer area operation data according to different acquisition frame length requirement levels by the clustering model.

[0010] In response to the frame length configuration command, clustering operations are performed using the clustering model based on the updated historical cluster set, and the target acquisition frame length requirement level is determined according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

[0011] Preferably, the adaptive Mahalanobis distance threshold is calculated as follows:

[0012] ;

[0013] In the formula, The threshold for determining the Mahalanobis distance of the k-th cluster is... For degrees of freedom Significance level The corresponding chi-square distribution critical value, where d is the feature vector dimension and c is the preset small sample correction coefficient. This represents the current cumulative number of samples in the k-th cluster.

[0014] Preferably, after constructing a new historical cluster based on the operating feature vector of the transformer area, the method further includes:

[0015] When the number of historical clusters exceeds a preset upper limit, calculate the normalized Euclidean distance and probability distribution similarity between any pair of historical clusters.

[0016] Based on the normalized Euclidean distance and probability distribution similarity of the historical cluster pairs, and according to the preset judgment conditions, if the judgment conditions are met, the historical cluster pairs are merged to obtain merged historical clusters. The cluster feature vector of the merged historical cluster is obtained by weighted fusion based on the cluster feature vectors of the two historical clusters before merging, combined with weight coefficients. The weight coefficients are determined according to the sample size of the two historical clusters before merging. The acquisition frame length requirement level corresponding to the merged historical cluster inherits the higher level of the two historical clusters before merging.

[0017] Preferably, it further includes:

[0018] A dynamic sliding time window is constructed using a smooth iterative formula based on distribution entropy;

[0019] Based on the dynamic sliding time window, the set of operating feature vectors of the transformer area within the current range of the dynamic sliding time window is determined, and the set of operating feature vectors of the transformer area is clustered to obtain a temporary cluster set.

[0020] Based on the temporary cluster set, calculate the centroid offset and dispersion rate of the temporary cluster set, wherein the centroid offset is the average Euclidean distance between the temporary cluster center and the reference cluster center in the current window, and the dispersion rate is the average ratio of the average sample variance of each cluster in the current window to the reference variance.

[0021] The centroid offset and the rate of change of dispersion are weighted and fused to obtain the comprehensive drift score of the current dynamic sliding time window;

[0022] According to the preset comprehensive drift score threshold and the upper limit of the number of windows exceeding the limit, when it is detected that the comprehensive drift score of multiple consecutive windows exceeds the comprehensive drift score threshold and the number of consecutive windows exceeding the limit reaches the upper limit of the number of windows exceeding the limit, a concept drift alarm message is output.

[0023] Preferably, when a concept drift alarm message is detected, adjusting the acquisition frame length configuration parameters of the low-voltage substation according to the target acquisition frame length requirement level includes:

[0024] Based on the target acquisition frame length requirement level and the current acquisition frame length requirement level of the low-voltage substation, a frame length requirement level sequence is determined according to the preset acquisition frame length requirement level sorting information. The frame length requirement level sequence consists of each acquisition frame length requirement level from the current acquisition frame length requirement level to the target acquisition frame length requirement level.

[0025] Based on the frame length requirement level sequence and combined with the preset adjustment cycle interval configuration information, the acquisition frame length configuration parameters of the low-voltage station area are adjusted sequentially in each adjustment cycle according to the next acquisition frame length requirement level until the acquisition frame length configuration parameters reach the standard corresponding to the target acquisition frame length requirement level.

[0026] Preferably, after adjusting the acquisition frame length configuration parameters of the low-voltage substation according to the target acquisition frame length requirement level, the method further includes:

[0027] Monitor the acquisition success rate and channel load rate of the data acquisition equipment in the monitoring area;

[0028] Based on the comparison results of the acquisition success rate, the channel load rate and the preset abnormal status judgment conditions, when the acquisition success rate or the channel load rate meets the abnormal status judgment conditions, the acquisition frame length configuration parameter of the low-voltage station area is reverted to the acquisition frame length requirement level before adjustment.

[0029] Preferably, based on the historical cluster set, the candidate clusters corresponding to the minimum distance between the operating feature vector of the transformer area and each historical cluster are determined, including:

[0030] Based on the historical cluster set, the Euclidean distance between the operating feature vector of the transformer area and each historical cluster is calculated. Based on the Euclidean distance and combined with a preset initial screening distance threshold, each historical cluster is initially screened to obtain preliminary candidate clusters.

[0031] Based on the preliminary candidate clusters, the Mahalanobis distance between the operating feature vector of the transformer area and each preliminary candidate cluster is calculated, and the preliminary candidate cluster corresponding to the minimum Mahalanobis distance is taken as the final candidate cluster.

[0032] Preferably, feature extraction is performed based on the real-time operation data of the transformer substation and a preset statistical period to obtain the transformer substation operation feature vector for the statistical period, including:

[0033] Based on the real-time operation data of the transformer area, feature extraction is performed in combination with a preset statistical period to obtain the operation characteristics of the transformer area in the statistical period. The operation characteristics of the transformer area include: load fluctuation rate, peak percentage, frequency of abnormal events, three-phase imbalance and new load penetration rate.

[0034] The z-score normalization process is applied to the operating characteristics of each transformer area to obtain the operating characteristic vector of the transformer area.

[0035] The second aspect of this application provides a low-voltage distribution area acquisition frame length dynamic adaptation device, comprising:

[0036] The transformer area data acquisition unit is used to collect real-time operating data of the transformer area;

[0037] The feature vector extraction unit is used to extract features based on the real-time operation data of the transformer area and in combination with a preset statistical period to obtain the transformer area operation feature vector for the statistical period.

[0038] The streaming incremental clustering update unit is used to determine the candidate cluster corresponding to the minimum distance between the transformer area operation feature vector and the historical clusters contained in the historical cluster set. If the Mahalanobis distance between the transformer area operation feature vector and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, the transformer area operation feature vector is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the transformer area operation feature vector. The historical cluster set contains multiple historical clusters, and the historical clusters are obtained by the clustering model based on the historical transformer area operation data according to different acquisition frame length requirement levels.

[0039] The frame length configuration instruction response unit is used to respond to the frame length configuration instruction, perform clustering operations based on the updated historical cluster set through the clustering model, and determine the target acquisition frame length requirement level according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

[0040] A third aspect of this application provides a low-voltage distribution area acquisition frame length dynamic adaptation terminal, including: a memory and a processor;

[0041] The memory is used to store program code, which corresponds to the low-voltage station area acquisition frame length dynamic adaptation method provided in the first aspect of this application.

[0042] The processor is used to read and execute the program code to implement the low-voltage distribution area acquisition frame length dynamic adaptation method.

[0043] As can be seen from the above technical solutions, this application has the following advantages:

[0044] The solution provided in this application utilizes a streaming incremental clustering mechanism. Based on this mechanism, single-sample processing involves only the statistical recursion of a single or a small number of clusters, eliminating the need for full recalculation. This reduces model update time from hours to seconds, enabling continuous tracking of transformer area characteristics on a daily basis. It maintains the timeliness of the hierarchical model, avoids the model bias accumulation problem caused by traditional static snapshot methods, and has the ability to adapt to dynamic changes in transformer area load in real time. This improves the timeliness and accuracy of collected data, thus providing a solid foundation for ensuring the accuracy of subsequent line loss calculations. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of the core process of an embodiment of a low-voltage substation acquisition frame length dynamic adaptation method based on cluster analysis provided in this application.

[0047] Figure 2 This is a schematic diagram illustrating the determination and operation process of streaming online incremental clustering and cluster merging in an embodiment of a low-voltage substation acquisition frame length dynamic adaptation method based on cluster analysis provided in this application.

[0048] Figure 3 This is a schematic diagram illustrating the adaptive sliding window length adjustment, two-dimensional drift detection, and progressive frame length adjustment in an embodiment of a low-voltage substation acquisition frame length dynamic adaptation method based on cluster analysis provided in this application.

[0049] Figure 4 This is a schematic diagram of the core architecture of an embodiment of a low-voltage substation acquisition frame length dynamic adaptation device based on cluster analysis provided in this application.

[0050] Figure 5 This is a schematic diagram of the core architecture of a low-voltage distribution area acquisition frame length dynamic adaptation terminal embodiment based on cluster analysis provided in this application. Detailed Implementation

[0051] Research has revealed that the continuous connection of new loads to low-voltage distribution areas has led to significant spatiotemporal dynamic evolution of load characteristics. This evolution stems from the combined effects of the inherent operating patterns of these new loads and external disturbances such as meteorological conditions and regional economic fluctuations, resulting in a continuous trend-driven drift of core characteristic parameters of the distribution area. Traditional offline clustering methods employ a "static snapshot" update mechanism, freezing historical data distribution only at discrete time points, completely failing to detect the gradual distribution changes that occur between two update windows. The fundamental contradiction between the continuous dynamic evolution of data and the fixed nature of the model causes the cluster centers and cluster boundaries trained based on outdated historical data to continuously accumulate and amplify the deviation from the actual power consumption patterns of the current distribution area. This significantly shortens the effective adaptation cycle of the clustering model, and the grading results obtained using the "outdated" clustering model increasingly deviate from actual needs, ultimately affecting the timeliness of data collection and the accuracy of line loss calculations.

[0052] In view of this, this application provides a method, device and terminal for dynamic adaptation of low-voltage distribution area acquisition frame length based on cluster analysis, which is used to realize the ability to adapt to dynamic changes in distribution area load and improve the timeliness and accuracy of acquisition data.

[0053] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] First, a detailed description of an embodiment of a low-voltage transformer area acquisition frame length dynamic adaptation method based on cluster analysis provided in this application is as follows:

[0055] Please see Figure 1 This application provides a method for dynamic adaptation of low-voltage distribution area acquisition frame length, including:

[0056] Step 101: Collect real-time operating data of the transformer area;

[0057] Step 102: Based on the real-time operation data of the transformer area, feature extraction is performed in combination with the preset statistical period to obtain the transformer area operation feature vector for the statistical period.

[0058] Step 103: Based on each historical cluster contained in the historical cluster set, determine the candidate cluster corresponding to the minimum distance by using the Mahalanobis distance between the operating feature vector of the transformer area and the historical cluster. If the Mahalanobis distance between the operating feature vector of the transformer area and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, then the operating feature vector of the transformer area is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the operating feature vector of the transformer area.

[0059] Step 104: In response to the frame length configuration command, clustering operation is performed using a clustering model based on the updated historical cluster set, and the target acquisition frame length requirement level is determined according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

[0060] For ease of understanding, the following explains some key terms in this embodiment:

[0061] Real-time operating data of a low-voltage distribution area refers to the instantaneous or short-cycle operating status data such as power consumption, voltage, current, and power factor acquired by various sensors and data acquisition devices during actual operation. This data reflects the current power load characteristics and operating status of the distribution area.

[0062] The statistical period refers to the time window used to aggregate and analyze real-time operational data of a transformer substation. For example, it can be set to one day, one week, or one month, and the data within this time window will be used to extract the operational characteristics of the substation.

[0063] The transformer substation operation feature vector is a multi-dimensional numerical representation formed by extracting and quantifying features from the real-time operation data of the substation within a statistical period. This vector can comprehensively reflect the key operational characteristics of the substation within a specific statistical period, such as load fluctuations, peak distribution, and abnormal events.

[0064] A historical cluster set refers to a collection of multiple historical clusters. This set stores the clustering results of the clustering model at different time points based on historical station operation data, representing different acquisition frame length requirement levels. The construction process can be seen in the following example: The first batch of historical data is clustered offline using the k-means++ algorithm or other clustering algorithms to generate an initial historical cluster set (the initial number of clusters is generally set to 5, corresponding to five requirement levels: extremely high, high, medium, low, and extremely low). Each cluster stores three types of summary statistics: sample count, feature mean vector (as the cluster center), and feature variance vector. A corresponding acquisition frame length strategy is bound to each cluster. Furthermore, a grid spatial index structure can be established simultaneously: the normalized feature space is evenly divided into 10 intervals per dimension, forming 10^d grid cells (d is the feature dimension). A list of cluster numbers contained in each grid is recorded to accelerate the initial screening of candidate clusters for new samples.

[0065] Historical clusters refer to the categories obtained by clustering models based on historical transformer station operation data and according to different acquisition frame length requirement levels. Each historical cluster represents a type of transformer station with similar operating characteristics and the same frame length requirement level.

[0066] Mahalanobis distance is a distance metric that measures the similarity between two sample points. It takes into account the covariance of the data distribution and can effectively handle the correlation between features of different dimensions, thus more accurately reflecting the distance between sample points and cluster centers.

[0067] A candidate cluster is a historical cluster selected during streaming incremental clustering updates by calculating the Mahalanobis distance between the current transformer area's operating feature vector and each historical cluster in the historical cluster set, and then choosing the historical cluster with the minimum distance. This candidate cluster represents the category to which the current transformer area's operating feature vector is most likely to belong.

[0068] The adaptive Mahalanobis distance threshold is a dynamic threshold used to determine whether the operational feature vector of a transformer substation should be included in a candidate cluster. This threshold can be adjusted according to the internal data distribution and sample size of the candidate cluster to adapt to the dynamic changes in the operational characteristics of the substation.

[0069] The cluster feature vector refers to the central vector of each historical cluster, which represents the average or typical characteristics of the operating feature vectors of all transformer areas within that cluster. When a new transformer area's operating feature vector is incorporated into the cluster, the cluster feature vector is updated to reflect changes in the data distribution within the cluster.

[0070] Clustering models are mathematical models or algorithms used to classify or group the operational feature vectors of transformer substations. These models can identify substations with similar characteristics and group them into the same category, thereby enabling hierarchical management of substations.

[0071] The data acquisition frame length requirement level refers to dividing a distribution area into different data acquisition frequency levels based on its operational characteristics and service needs. For example, it can be divided into high, medium, and low levels, each corresponding to a different data acquisition frame length.

[0072] A frame length configuration command is a command issued by system or maintenance personnel to trigger an adjustment of the frame length for low-voltage distribution area data acquisition. This command typically includes the adjustment target or triggering conditions.

[0073] The target acquisition frame length requirement level refers to the ideal acquisition frame length requirement level that the low-voltage substation area should be adjusted to, as determined by the clustering model calculation results.

[0074] The frame length configuration parameter refers to the specific setting in the low-voltage distribution area acquisition equipment used to control the data acquisition cycle interval. Adjusting this parameter directly affects the data acquisition frequency.

[0075] More specifically, the low-voltage distribution area acquisition frame length dynamic adaptation method provided in this embodiment has the following implementation logic:

[0076] First, please refer to Figure 2 Real-time operational data from the low-voltage distribution area is collected. This data can be periodically uploaded to a data center or cloud platform by smart meters, data acquisition terminals, or other monitoring devices installed in the low-voltage distribution area. For example, basic electrical parameters such as voltage, current, active power, and reactive power are collected. This raw data forms the basis for subsequent analysis and feature extraction.

[0077] Next, based on the collected real-time operational data of the transformer substations, feature extraction is performed in conjunction with a preset statistical period to obtain the substation operational feature vector for that statistical period. For example, a fixed statistical period, such as daily, can be set, and then statistics such as load, abnormal event records, and three-phase imbalance can be calculated for that day. These statistics can be directly combined into a feature vector to characterize the operational status of the transformer substation within that statistical period.

[0078] Based on this, and using the historical clusters contained in the historical cluster set, the candidate cluster corresponding to the minimum distance is determined by calculating the Mahalanobis distance between the current transformer area's operating feature vector and each historical cluster. For example, each historical cluster in the historical cluster set can be traversed, the Mahalanobis distance between the current transformer area's operating feature vector and the cluster center of each historical cluster can be calculated, and then the historical cluster with the minimum Mahalanobis distance can be selected as the candidate cluster. This historical cluster set can be pre-constructed during system initialization by performing a one-time cluster analysis on a large amount of historical transformer area operating data, where each historical cluster represents a type of transformer area with a specific acquisition frame length requirement level.

[0079] If the Mahalanobis distance between the current operating feature vector of a transformer area and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, then the operating feature vector of the transformer area is incorporated into the candidate cluster, and updated according to the cluster feature vector of the candidate cluster. For example, when an operating feature vector of a transformer area is determined to belong to a certain candidate cluster, the vector is added to the dataset of that cluster, and the cluster feature vectors of that cluster (e.g., feature mean vector and feature variance vector) are updated by simple averaging or weighted averaging to reflect the impact of the newly added sample. The adaptive Mahalanobis distance threshold can be a preset fixed value or set based on experience.

[0080] For example, regarding the update of cluster feature vectors, when the running feature vector of a certain area is determined to be a sample within the cluster, the feature mean vector and variance vector of the cluster are updated using a recursive formula with a cluster age decay factor, and the number of samples is updated simultaneously. The formula is as follows:

[0081] ;

[0082] in, The time decay coefficient, For the creation time of the k-th cluster, Let be the feature vector of the new sample arriving at time t. Let be the number of samples in the k-th cluster at time t-1. Used to refer to cluster eigenvectors, namely the eigenmean vector and the eigenvariance vector. and These represent the cluster feature vectors of the k-th cluster at time t-1 and time t, respectively.

[0083] Otherwise, if the Mahalanobis distance between the operating feature vector of a transformer area and the candidate cluster is not less than the adaptive Mahalanobis distance threshold, a new historical cluster is constructed based on the operating feature vector of that transformer area. For example, when the operating feature vector of a transformer area is significantly different from all existing historical clusters, indicating that it does not belong to any existing category, it can be considered as the emergence of a new transformer area operating mode, and a new historical cluster is initialized centered on this vector. The required frame length level for this new cluster can be initially set based on the degree of matching between its feature vector and the existing levels.

[0084] Finally, in response to the frame length configuration command, based on the updated historical cluster set, clustering operations are performed using the clustering model. Based on the clustering results, the target acquisition frame length requirement level is determined, and the acquisition frame length configuration parameters of the low-voltage distribution area are adjusted accordingly. For example, upon receiving a frame length configuration command, the system can rerun the clustering model to perform an overall analysis of the currently updated historical cluster set, or reclassify each distribution area, thereby determining a target acquisition frame length requirement level most suitable for each area's current operating characteristics. Subsequently, based on this target level, the corresponding acquisition frame length configuration parameters are sent to the acquisition equipment in the low-voltage distribution area, completing the frame length adjustment.

[0085] It is understandable that the above frame length configuration command can be triggered either by a timer program or by manual input.

[0086] This method effectively addresses the spatiotemporal evolution of transformer substation load characteristics caused by new load access by collecting real-time operational data and dynamically updating the clustering model. Through a streaming incremental clustering mechanism, the model continuously senses changes in transformer substation operating modes, avoiding the model expiration problem caused by the "static snapshot" updates of traditional offline clustering. This achieves dynamic and accurate adaptation of the low-voltage transformer substation acquisition frame length, ensuring the timeliness and completeness of the acquired data, thereby improving the accuracy of line loss calculation.

[0087] In some of the above embodiments, this application proposes to determine whether the operating feature vector of a transformer area is incorporated into a candidate cluster by using an adaptive Mahalanobis distance threshold. Based on this, this application further proposes the following formula for calculating the adaptive Mahalanobis distance threshold:

[0088] ;

[0089] In the formula, The threshold for determining the Mahalanobis distance of the k-th cluster is... For degrees of freedom Significance level The corresponding chi-square distribution critical value, where d is the feature vector dimension and c is the preset small sample correction coefficient. This represents the current cumulative number of samples in the k-th cluster.

[0090] Wherein, the adaptive Mahalanobis distance threshold It is a key criterion used to determine whether newly introduced transformer area operating feature vectors should be merged into existing candidate clusters. Unlike a fixed threshold, this threshold can be dynamically adjusted based on the statistical characteristics of the clusters, thereby improving the accuracy and robustness of clustering decisions.

[0091] The calculation formula This represents the chi-square distribution critical value corresponding to the degrees of freedom d and the significance level α. Here, d corresponds to the dimension of the operating feature vector of the transformer area, i.e., the number of features. The significance level α is a preset probability value, usually taken as 0.05 or 0.01, used to control the strictness of the judgment. This provides a statistical benchmark for the Mahalanobis distance threshold, ensuring that, given the data dimensions and confidence level, determining whether a new sample belongs to an existing cluster is statistically significant. Its specific value can be obtained by consulting a chi-square distribution table or through statistical calculation.

[0092] The calculation formula (1+c / N) k ) is a small-sample correction factor used to dynamically adjust the threshold. Here, c is the small-sample correction coefficient, a configurable parameter used to adjust the strength of the correction. N k Let N be the current cumulative sample count of the k-th cluster, representing the number of data points already included in that cluster. When the number of samples in the cluster is N... k When the value is small, this correction factor will increase significantly, making the adaptive Mahalanobis distance threshold... The larger the sample size N, the more relaxed the conditions for incorporating new samples into the cluster, preventing premature inclusion of new samples in a new cluster due to insufficient sample size. Conversely, when the number of samples N in a cluster increases... k When the value is large, the correction factor approaches 1, making the threshold more stringent and ensuring the clarity and stability of the cluster boundaries.

[0093] By introducing an adaptive Mahalanobis distance threshold, this application overcomes the limitations of traditional fixed thresholds in streaming incremental clustering. Specifically, this threshold combines a chi-square distribution critical value. and based on the number of cluster samples N k Small sample correction factor (1+c / N) k Chi-square distribution critical value This provides a statistical basis for threshold setting, ensuring the scientific validity and reliability of the judgment given a significance level α and eigenvector dimension d. Simultaneously, a small sample correction factor (1+c / N) is used. k This effectively solves the instability problem that may occur in Mahalanobis distance calculation for small sample clusters. When the number of samples in a cluster is N kWhen the value is small, the correction factor increases the threshold, making it easier for new samples to be incorporated into existing small clusters. This avoids frequently creating new clusters due to insufficient samples, thus improving the robustness of clustering. Conversely, when the number of samples in a cluster N is large... k When the threshold is large, the correction factor approaches 1, making the threshold more stringent and ensuring the clarity of the cluster boundaries. This adaptive threshold mechanism allows the clustering process to better adapt to the dynamic changes in the operating data of the transformer substation and the statistical characteristics of different clusters, thereby improving the accuracy and stability of streaming incremental clustering. This provides a more reliable basis for determining the subsequent target acquisition frame length requirements, and ultimately achieves more accurate and effective dynamic adaptation of low-voltage transformer substation acquisition frame lengths.

[0094] In some of the above embodiments, this application proposes to dynamically update the historical cluster set based on the operating feature vector of the transformer area using streaming incremental clustering, and to construct new historical clusters when the conditions of existing clusters are not met. Furthermore, this application proposes that after constructing new historical clusters based on the operating feature vector of the transformer area, the following steps are also included:

[0095] When the number of historical clusters exceeds a preset upper limit, calculate the normalized Euclidean distance and probability distribution similarity between any pair of historical clusters.

[0096] Based on the normalized Euclidean distance and probability distribution similarity of the historical cluster pairs, and according to the preset judgment conditions, if the judgment conditions are met, the historical cluster pairs are merged to obtain the merged historical clusters.

[0097] The cluster feature vector of the merged historical cluster is obtained by weighted fusion based on the cluster feature vectors of the two historical clusters before the merger and weight coefficients. The weight coefficients are determined according to the sample size of the two historical clusters before the merger, and the acquisition frame length requirement level corresponding to the merged historical cluster inherits the higher level of the two historical clusters before the merger.

[0098] For details, please refer to Figure 2 When the number of historical clusters exceeds a preset upper limit, the system will trigger a cluster merging mechanism. This preset upper limit can be flexibly configured based on the actual application scenario, system computing resources, and requirements for clustering accuracy; for example, it can be set to 10, 15, or more. This upper limit is designed to balance the complexity of the model with the ability to provide a refined representation of the operating status of the transformer substations.

[0099] To assess the similarity between historical cluster pairs, this application employs two complementary metrics: normalized Euclidean distance and probability distribution similarity. Normalized Euclidean distance measures the geometric proximity of the centers (i.e., cluster feature vectors) of two clusters in the feature space. Normalization ensures comparability of distance values ​​across different feature dimensions. Probability distribution similarity, from a statistical perspective, reflects the degree of overlap or difference in sample distributions within two clusters. Specifically, probability distribution similarity can be characterized using confidence ellipsoidal overlap or probability density-based KL divergence. Confidence ellipsoidal overlap quantifies similarity by calculating the degree of overlap between the confidence regions of the two clusters in the feature space; higher overlap indicates closer sample distributions. Probability density-based KL divergence, on the other hand, characterizes similarity by measuring the information difference between two probability distributions; smaller KL divergence values ​​indicate greater similarity between the two distributions.

[0100] After calculating the normalized Euclidean distance and probability distribution similarity of historical cluster pairs, the system decides whether to merge them based on preset criteria. These criteria can be a composite logic, for example, requiring the normalized Euclidean distance to be less than a certain threshold and the probability distribution similarity (e.g., confidence ellipsoid overlap) to be greater than a certain threshold, or the probability distribution similarity (e.g., KL divergence) to be less than a certain threshold. By comprehensively considering these two metrics, the system ensures that only clusters that are highly similar in both geometric location and statistical distribution are merged, thus avoiding information distortion caused by unreasonable cluster merging.

[0101] When the merging conditions are met, two historical clusters will be merged into a new historical cluster. The cluster feature vector of the merged historical cluster is not simply averaged, but rather a weighted fusion based on the cluster feature vectors of the two historical clusters before the merge. The weighting coefficients are determined based on the sample size of the two historical clusters before the merge; that is, the larger the sample size of the cluster, the greater its contribution to the feature vector of the merged new cluster. This weighted fusion method ensures that the feature vector of the merged cluster can more accurately reflect the overall characteristics of all the samples it contains. Furthermore, the acquisition frame length requirement level corresponding to the merged historical cluster will inherit the higher level from the two historical clusters before the merge. For example, if one cluster corresponds to the "high-frequency acquisition" level and the other to the "medium-frequency acquisition" level, the merged cluster will inherit the "high-frequency acquisition" level. This strategy aims to ensure that the simplification of the model during the cluster merging process does not reduce the precision requirements for data acquisition in the distribution area, thereby maintaining the accurate perception capability of the distribution area's operating status.

[0102] Through the above technical solution, this application can actively identify and merge similar historical clusters when the number of historical clusters reaches a preset upper limit during the streaming incremental clustering process. This effectively avoids the infinite expansion of the historical cluster set, reduces the computational complexity and storage overhead of the clustering model, and thus improves the system's operating efficiency and resource utilization. Simultaneously, by comprehensively considering normalized Euclidean distance and probability distribution similarity to determine the cluster merging conditions, it ensures that only statistically highly similar clusters are merged, thereby maintaining the accuracy of the clustering results and the ability to effectively distinguish the operating status of the transformer substations. The weighted fusion of cluster feature vectors and the strategy of inheriting higher acquisition frame length requirements further ensure that the merged clusters accurately reflect the operating characteristics of the transformer substations they represent and maintain the responsiveness to high-frequency data acquisition needs, avoiding a decrease in data acquisition accuracy due to cluster merging, thus ensuring the robustness and effectiveness of dynamic adaptation of low-voltage transformer substation acquisition frame lengths.

[0103] In some of the above embodiments, this application proposes a low-voltage distribution area acquisition frame length dynamic adaptation method that relies on historical cluster sets to determine the target acquisition frame length requirement level. However, over time, the operating mode and load characteristics of low-voltage distribution areas may change significantly, causing the historical cluster set to fail to accurately reflect the actual operating status of the current distribution area, thereby affecting the accuracy and effectiveness of frame length adaptation. To address this, this application further proposes a concept drift detection mechanism, specifically including:

[0104] A dynamic sliding time window is constructed using a smooth iterative formula based on distribution entropy.

[0105] It should be noted that the following can be consulted: Figure 3 This dynamic sliding time window is designed to adaptively capture local characteristic changes in the data stream. Its size and sliding step can be adjusted according to the rate and pattern of change in the data stream, thereby more sensitively capturing potential concept drift. The window length is adjusted using a smooth iterative formula based on distribution entropy, as follows:

[0106] ;

[0107] In the formula, The window length for the current period. To adjust the step size coefficient, This represents the distribution entropy value of the cluster labels within the current window. The preset stable distribution reference entropy value is used; the window length is limited to between 24h and 168h.

[0108] The value of β (adjustment step size coefficient) is a hyperparameter that is flexibly configured according to the load change characteristics of the distribution area in actual deployment. Its function is essentially to control the sensitivity of the sliding window length to changes in distributed entropy, i.e., to balance "response speed" and "noise resistance stability." In engineering practice, it is typically set between 0.01 and 0.2: a smaller value results in a smoother window change, suitable for stable distribution areas with small load fluctuations and slow changes; a larger value results in a more agile window response to entropy changes, suitable for distribution areas with high penetration rates of new loads and drastic load characteristic changes. It is generally recommended to use the intermediate value of 0.08 as the default configuration for initial deployment, and then fine-tune it based on actual detection results after 2-4 weeks of operation. If the window length fluctuates frequently and large amounts, resulting in frequent false alarms, then appropriately decrease β; if the window responds slowly to significant load changes and has many missed alarms, then appropriately increase β. Fine-tuning to a reasonable range through short-term observation can meet engineering requirements.

[0109] Based on the dynamic sliding time window, the system determines the set of transformer area operation feature vectors within the current dynamic sliding time window range and performs clustering processing on the set of transformer area operation feature vectors to obtain a temporary cluster set. This step aims to obtain the data features within the current time window and perform local clustering to reflect the local structure of the transformer area operation data at the current moment. Once the range of the dynamic sliding time window is determined, the system collects all newly arrived transformer area operation feature vectors within the window to form a feature vector set. Subsequently, the aforementioned clustering model or other clustering algorithms such as K-means, DBSCAN, and hierarchical clustering can be used to process this set to obtain a set of temporary clusters, which represent the main patterns or states of the data within the current window.

[0110] Based on the temporary cluster set, the centroid offset and dispersion rate of the temporary cluster set are calculated. The centroid offset is the average Euclidean distance between the temporary cluster center and the reference cluster center in the current window, and the dispersion rate is the average ratio of the average sample variance of each cluster in the current window to the reference variance. The centroid offset and dispersion rate are key indicators for measuring the difference between the current data distribution and the reference data distribution, used to quantify the degree of concept drift. The centroid offset reflects the change in the location of the data center, while the dispersion rate reflects the change in the density or dispersion of the data distribution. The reference cluster center and reference variance can be derived from the parameters of the corresponding clusters in the historical cluster set, or from the parameters of the clustering results within a previous stable time period.

[0111] The centroid offset and the rate of change of dispersion are weighted and fused to obtain the comprehensive drift score for the current dynamic sliding time window. Weighting and fusing two independent drift metrics allows for a more comprehensive and robust assessment of the overall degree of concept drift, avoiding the limitations of a single metric. For example, the centroid offset and the rate of change of dispersion can be linearly weighted and summed using preset weighting coefficients. These weights can be adjusted based on the sensitivity to changes in center position and distribution shape in actual application scenarios.

[0112] Based on preset comprehensive drift score thresholds and an upper limit on the number of windows exceeding the limit, when multiple consecutive windows are detected to have comprehensive drift scores exceeding the threshold, and the number of consecutively exceeding windows reaches the upper limit, a concept drift alarm message is output. For example, if the scores of three consecutive windows exceed the threshold of 0.7, concept drift is determined to have occurred. The detection process is based solely on cluster summary statistics and does not re-cluster. This mechanism is used to determine whether concept drift has actually occurred and to issue an alarm in a timely manner. By setting thresholds and an upper limit on the number of consecutively exceeding windows, false judgments caused by short-term fluctuations or noise can be effectively avoided, improving the accuracy and stability of drift detection. The system continuously monitors the comprehensive drift score calculated for each dynamic sliding time window. When preset conditions are met, concept drift is determined to have occurred, and an alarm message is immediately output.

[0113] Through the above technical solution, this application can adaptively capture local characteristic changes in low-voltage distribution area operating data and comprehensively quantify the difference between the current data distribution and the baseline distribution. By obtaining a comprehensive drift score through weighted fusion and combining it with a judgment mechanism for the number of consecutive out-of-limit windows, misjudgments caused by short-term fluctuations are effectively avoided, improving the accuracy and robustness of concept drift detection. Once concept drift is detected, the system can promptly issue an alarm, prompting the need to update the historical cluster set or retrain the clustering model. This ensures the long-term effectiveness and accuracy of the dynamic adaptation method for low-voltage distribution area acquisition frame length, avoids a decline in adaptation effect due to changes in data distribution, and guarantees the real-time nature and rationality of acquisition frame length adjustment.

[0114] In some of the above embodiments, although the scheme proposed in this application can dynamically determine the target acquisition frame length requirement level based on the operating characteristics of the distribution area, and adjust the acquisition frame length configuration parameters of the low-voltage distribution area and the concept drift detection mechanism for clustering accordingly, when concept drift is triggered, directly adjusting the acquisition frame length from the current level to the target level may cause momentary instability in the data acquisition system or communication link due to excessive frame length change, or even data loss or transmission errors, affecting the reliability and stability of the system. Therefore, this application further proposes detailed steps for adjusting the acquisition frame length configuration parameters of the low-voltage distribution area according to the target acquisition frame length requirement level, specifically including:

[0115] Based on the target acquisition frame length requirement level and the current acquisition frame length requirement level of the low-voltage substation, the frame length requirement level sequence is determined according to the preset acquisition frame length requirement level sorting information; wherein, the frame length requirement level sequence consists of each acquisition frame length requirement level from the current acquisition frame length requirement level to the target acquisition frame length requirement level.

[0116] Based on the frame length requirement level sequence and combined with the preset adjustment cycle interval configuration information, the acquisition frame length configuration parameters of the low-voltage station area are adjusted sequentially in each adjustment cycle according to the next acquisition frame length requirement level until the acquisition frame length configuration parameters reach the standard corresponding to the target acquisition frame length requirement level.

[0117] It should be noted that when determining the frame length requirement level sequence, the system maintains a predefined sorting information for the acquisition frame length requirement levels. This sorting information can be an ordered list, such as an arrangement from high frequency to low frequency or from low frequency to high frequency. After the system obtains the acquisition frame length requirement level for the current low-voltage substation and the target acquisition frame length requirement level determined through clustering operations, it searches for the positions of these two levels in this sorting information. Subsequently, the system extracts all levels between the current level and the target level, including the current level and the target level themselves, thus constructing a complete frame length requirement level sequence. For example, if the current level is "Medium," the target level is "Extra High," and the sorting information is "Extra High, High, Medium, Low, Extra Low," assuming the adjustment period is set to 24 hours, the current level is allowed to adjust to the target level by a maximum of one level within each period (i.e., from "Medium" to "High"). After adjustment, a 24-hour cooling timer is started, during which reverse adjustment is prohibited. After the cooling timer ends, another level is adjusted (i.e., from "High" to "Extra High"). Through multiple iterations, all transformer areas gradually approach the final classification result, avoiding abrupt changes.

[0118] The above technical solution breaks down the frame length adjustment process into multiple small steps and introduces adjustment period intervals, enabling the frame length configuration parameters of low-voltage distribution areas to smoothly and gradually transition from the current level to the target level. This phased adjustment method effectively avoids problems such as system instability, data transmission interruption, or data loss that may be caused by sudden changes in frame length, significantly improving the operational reliability and data integrity of the acquisition system. Simultaneously, the preset sorting information and adjustment period intervals ensure the orderliness and controllability of the adjustment process, allowing the system to maintain a continuous and stable operating state when adapting to new acquisition requirements, thereby better supporting subsequent data analysis and business applications.

[0119] In some of the above embodiments, this application proposes a dynamic adaptation method for the acquisition frame length of low-voltage distribution areas. This method can dynamically adjust the acquisition frame length configuration parameters based on the distribution area's operational feature vector and historical cluster set. However, in practical applications, although the adjustment is based on data analysis and model prediction, due to instantaneous changes in the communication environment, fluctuations in equipment status, or limitations of model prediction, the newly adjusted acquisition frame length configuration parameters may still lead to poor actual operating performance of the distribution area's acquisition equipment, such as a decrease in acquisition success rate or excessively high channel load rate, thereby affecting data integrity and system stability.

[0120] To address the aforementioned issues, this application further proposes that, after adjusting the acquisition frame length configuration parameters of the low-voltage substation according to the target acquisition frame length requirement level, the following additional steps are included:

[0121] Monitor the acquisition success rate and channel load rate of the data acquisition equipment in the monitoring area.

[0122] The data acquisition success rate refers to the proportion of data acquisition devices in a specific time period that successfully complete data acquisition and upload. It can be calculated by comparing the number of acquisition requests sent by the acquisition devices with the number of successfully received responses and uploaded data. For example, after each acquisition cycle, the acquisition devices report the execution results of the acquisition task to the higher-level system, including success or failure status. The system then summarizes and calculates the success rate based on this information. Channel load rate refers to the degree to which a communication channel is occupied by data transmission. It can be obtained by monitoring the ratio of the actual data transmission volume per unit time to the channel's maximum transmission capacity. For example, the communication module can record the size and time of the data packets it sends and receives, and the system calculates the channel bandwidth usage based on this. This monitoring data can be reported by the acquisition devices themselves or statistically analyzed by the higher-level communication management unit.

[0123] Based on this, according to the comparison results of the acquisition success rate, the channel load rate, and the preset anomaly judgment conditions, when the acquisition success rate or the channel load rate meets the anomaly judgment conditions, the acquisition frame length configuration parameters of the low-voltage substation are rolled back to the acquisition frame length requirement level before the adjustment. The preset anomaly judgment conditions can be set based on historical operating experience, system design requirements, or expert knowledge. For example, they can be set as specific numerical thresholds such as an acquisition success rate below 90% or a channel load rate above 80%, or they can be an anomaly range dynamically calculated based on statistical methods (such as moving average or standard deviation). The system compares the real-time monitored acquisition success rate and channel load rate with these preset judgment conditions. When any condition is met, the system will trigger a rollback operation, restoring the current acquisition frame length configuration parameters of the low-voltage substation to the state before the most recent adjustment. This typically requires the system to save the current frame length requirement level before each frame length adjustment so that the saved level can be loaded and applied when needed.

[0124] By introducing a real-time monitoring mechanism for the acquisition success rate and channel load rate of the data acquisition equipment in the distribution area, and setting conditions for judging abnormal status, this application can promptly detect the negative impacts that may be caused by the adjustment of the acquisition frame length. When a decrease in acquisition performance (such as acquisition success rate) or excessively high communication resources (such as channel load rate) is detected, the system can respond quickly and revert the acquisition frame length configuration parameters to the stable state before the adjustment. This effectively avoids problems such as system instability, incomplete data acquisition, or communication congestion caused by improper or unsuitable frame length adjustments, thereby enhancing the robustness and reliability of the dynamic adaptation method for low-voltage distribution area acquisition frame length, and ensuring the quality of data acquisition and the continuous stable operation of the system.

[0125] In some of the above embodiments, when determining the candidate cluster corresponding to the minimum distance based on the historical cluster set, this application needs to calculate the Mahalanobis distance between the operating feature vector of the transformer area and each historical cluster. However, the calculation of Mahalanobis distance involves the inversion of the covariance matrix. When the number of historical clusters is large or the feature vector dimension is high, directly calculating the Mahalanobis distance for all historical clusters will bring a large computational burden and affect the real-time response efficiency of the system. To address this, this application further proposes a method for dynamic adaptation of low-voltage transformer area acquisition frame length based on the historical cluster set, which determines the candidate cluster corresponding to the minimum distance by using the Mahalanobis distance between the operating feature vector of the transformer area and each historical cluster. This includes:

[0126] Based on the historical cluster set, the Euclidean distance between the operating feature vector of the transformer area and each historical cluster is calculated. Based on the Euclidean distance and combined with the preset initial screening distance threshold, each historical cluster is initially screened to obtain preliminary candidate clusters. Based on the preliminary candidate clusters, the Mahalanobis distance between the operating feature vector of the transformer area and each preliminary candidate cluster is calculated. The preliminary candidate cluster corresponding to the minimum Mahalanobis distance is taken as the final candidate cluster.

[0127] Specifically, when determining candidate clusters, the system first calculates the Euclidean distance between the current transformer area's operating feature vector and each historical cluster, based on the historical cluster set. Euclidean distance, a commonly used method for measuring the "straight-line distance" between two points in multidimensional space, is relatively simple to calculate and has low computational cost, without involving complex matrix operations. Its role is to provide a fast, preliminary similarity measure for identifying historical clusters that are geometrically close to the current transformer area's operating feature vector. For example, given a transformer area's operating feature vector and the cluster feature vectors of historical clusters, the system iterates through each dimension of the feature vectors. The system then iterates through each historical cluster in the historical cluster set, calculating the Euclidean distance between the current transformer area's operating feature vector and the cluster feature vector of that cluster.

[0128] Subsequently, based on the Euclidean distance and a preset initial screening distance threshold, each historical cluster is preliminarily screened to obtain preliminary candidate clusters. The purpose of this preliminary screening is to quickly eliminate obviously irrelevant historical clusters before performing the computationally expensive Mahalanobis distance calculation, thereby effectively narrowing the scope of subsequent Mahalanobis distance calculations. The preset initial screening distance threshold is a key parameter that can be set according to actual application scenarios and experience. It is used to define which historical clusters are sufficiently close in terms of Euclidean distance and deserves further refined consideration. After calculating the Euclidean distance between the running feature vector of the platform and all historical clusters, the system compares these distances with the preset initial screening distance threshold. Only historical clusters whose Euclidean distance is less than or equal to this threshold are retained, forming a smaller set, i.e., preliminary candidate clusters.

[0129] Based on this, the Mahalanobis distance between the operating feature vector of the transformer area and each preliminary candidate cluster is calculated according to the preliminary candidate clusters. The preliminary candidate cluster with the minimum Mahalanobis distance is selected as the final candidate cluster. Mahalanobis distance can take into account the covariance information of the data distribution, thus more accurately measuring the similarity between two sample points, especially when there is a correlation between feature dimensions, its measurement results are more reliable. After the initial screening significantly reduces the number of historical clusters, the Mahalanobis distance is calculated on the smaller number of preliminary candidate clusters, which ensures that the final selected candidate clusters are not only geometrically close, but also highly similar in statistical distribution. For each cluster j in the preliminary candidate cluster set, the system calculates the Mahalanobis distance between the operating feature vector x of the transformer area and the cluster center of cluster j. After calculating the Mahalanobis distances of all preliminary candidate clusters, the system selects the cluster with the smallest Mahalanobis distance as the final candidate cluster.

[0130] By introducing Euclidean distance for initial screening using the above technical solution, the number of historical clusters requiring Mahalanobis distance calculation can be significantly reduced. Euclidean distance calculation is simple and fast, effectively eliminating a large number of irrelevant clusters, thus reducing overall computational complexity. Based on this, Mahalanobis distance calculation is then performed on the smaller number of initial candidate clusters, ensuring the accuracy of the final candidate cluster selection while significantly improving the algorithm's efficiency and real-time performance. This two-stage distance calculation strategy effectively solves the computational burden problem caused by directly calculating the Mahalanobis distance of all clusters, while ensuring clustering accuracy. This allows the low-voltage distribution area acquisition frame length dynamic adaptation method to respond more quickly and efficiently to changes in the distribution area's operating status.

[0131] In some embodiments described above in this application, feature extraction is proposed based on real-time operating data of the transformer substation and a preset statistical period to obtain a transformer substation operating feature vector for the statistical period. Building upon this, this application further proposes that obtaining a transformer substation operating feature vector for the statistical period based on real-time operating data of the transformer substation and a preset statistical period includes: extracting features based on real-time operating data of the transformer substation and a preset statistical period to obtain operating features for the transformer substation for the statistical period, wherein the operating features include: load fluctuation rate, peak percentage, frequency of abnormal events, three-phase imbalance, and new load penetration rate; and performing z-score normalization on each operating feature of the transformer substation to obtain the operating feature vector.

[0132] Specifically, the selection of the aforementioned operating characteristics of the distribution area is crucial for accurately depicting the operating status of the low-voltage distribution area. Load fluctuation rate refers to the degree of load change relative to the average load within a specific statistical period. It effectively reflects the stability and activity of the distribution area's load. For example, it can be obtained by calculating the relationship between the maximum, minimum, and average load values ​​within the statistical period. Peak load percentage characterizes the intensity or duration of peak load in the distribution area within the statistical period. For example, it can be calculated as the ratio of peak load to average load, or the proportion of peak load duration to the total statistical time, which helps identify the peak electricity consumption characteristics of the distribution area. Abnormal event frequency refers to the cumulative number of various abnormal events occurring in the distribution area within the statistical period, such as voltage exceeding limits, current overload, and circuit breaker tripping. It directly reflects the reliability and stability of the distribution area's operation and is usually obtained through statistical analysis of alarm logs or event records. Three-phase imbalance measures the degree of imbalance in the three-phase current or voltage of the distribution area. For example, it can be calculated as the negative sequence component or imbalance rate of the three-phase current or voltage, which is of great significance for assessing the power supply quality and equipment health of the distribution area. The penetration rate of new loads reflects the proportion of new electrical equipment such as electric vehicle charging piles and distributed photovoltaic grid connection in the total load of the distribution area. For example, it can be obtained by statistically analyzing the number, capacity or electricity consumption of new load equipment. This helps to understand the evolution trend and future development potential of the load structure of the distribution area.

[0133] To eliminate differences in the dimensions and numerical ranges of operational features among different transformer substations and ensure that each feature has the same weight when constructing the feature vector, this application performs z-score normalization on the operational features of each substation. This process effectively prevents certain features with larger values ​​from dominating subsequent clustering analysis, thereby improving the accuracy and robustness of the clustering algorithm. After normalization, these feature values ​​collectively constitute the substation operational feature vector, which serves as the input to the subsequent streaming incremental clustering model to accurately represent the operational status of the substation.

[0134] Through the aforementioned technical solution, this application comprehensively and multidimensionally captures the operating status of low-voltage distribution areas by selecting representative operating characteristics such as load fluctuation rate, peak percentage, frequency of abnormal events, three-phase imbalance, and penetration rate of new loads. This includes load characteristics, power quality, equipment health status, and load structure change trends. The combination of these features has strong discriminative and representative capabilities, effectively reflecting subtle changes in the operating mode of the distribution area. Simultaneously, by performing z-score normalization on these features, the differences in dimensions and numerical ranges between different features are eliminated, ensuring that all features have the same weight when constructing feature vectors. This avoids certain features with larger values ​​dominating the clustering results, thereby significantly improving the accuracy and robustness of the subsequent streaming incremental clustering algorithm. Ultimately, the obtained distribution area operating feature vector can more accurately and stably represent the actual operating status of low-voltage distribution areas, providing high-quality input data for subsequent dynamic adaptation of acquisition frame lengths. This makes frame length adjustment decisions more precise and timely, effectively improving data acquisition efficiency and system adaptability.

[0135] The above is a detailed description of an embodiment of a low-voltage distribution area acquisition frame length dynamic adaptation method provided in this application. The following is a detailed description of an embodiment of a low-voltage distribution area acquisition frame length dynamic adaptation device provided in this application.

[0136] Please see Figure 4 This application provides a low-voltage distribution area acquisition frame length dynamic adaptation device, comprising:

[0137] The transformer area data acquisition unit 201 is used to collect real-time operating data of the transformer area.

[0138] The feature vector extraction unit 202 is used to extract features based on the real-time operation data of the transformer area and in combination with a preset statistical period to obtain the transformer area operation feature vector of the statistical period.

[0139] The streaming incremental clustering update unit 203 is used to determine the candidate cluster corresponding to the minimum distance between the transformer area operation feature vector and the historical cluster based on each historical cluster contained in the historical cluster set. If the Mahalanobis distance between the transformer area operation feature vector and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, the transformer area operation feature vector is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the transformer area operation feature vector. The historical cluster set contains multiple historical clusters, and the historical clusters are obtained by the clustering model based on the historical transformer area operation data according to different acquisition frame length requirement levels.

[0140] The frame length configuration instruction response unit 204 is used to respond to the frame length configuration instruction, perform clustering operation through the clustering model based on the updated historical cluster set, and determine the target acquisition frame length requirement level according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

[0141] Through the above technical solutions, the device achieves continuous tracking of the dynamic evolution of load characteristics in the distribution area, effectively solving the model adaptation failure problem caused by the excessively long update cycle of offline clustering methods. The streaming incremental clustering mechanism enables the clustering model to respond in real time to the gradual drift of characteristics such as load volatility and peak percentage caused by the access of new loads such as distributed photovoltaics and electric vehicle charging piles, avoiding the contradiction of "data constantly changing while the model remains unchanged" in traditional methods. The introduction of adaptive Mahalanobis distance threshold further improves the accuracy of cluster affiliation determination, enabling the clustering model to better adapt to the spatiotemporal dynamic evolution characteristics of new power systems, thereby ensuring the integrity and real-time nature of the collected data and providing a reliable data foundation for refined line loss calculation.

[0142] In other implementations, such as Figure 5 As shown, this application proposes a low-voltage distribution area acquisition frame length dynamic adaptation terminal, including a memory 33 and a processor 31, which can be connected via a communication bus 34. The memory 33 is used to store program code, which corresponds to the aforementioned low-voltage distribution area acquisition frame length dynamic adaptation method. The processor 31 is used to read and execute the program code to implement the low-voltage distribution area acquisition frame length dynamic adaptation method.

[0143] This embodiment combines the memory and processor in a collaborative manner. The memory stores the program code corresponding to the dynamic adaptation method of the low-voltage distribution area acquisition frame length, and the processor reads and executes the program code. This enables real-time analysis of the distribution area's operating characteristics and dynamic adjustment of the frame length configuration, resolving the contradiction between the dynamic evolution of the distribution area's load characteristics and the static update mechanism of the offline clustering method. This achieves the effects of extending the model adaptation cycle, improving the timeliness of data acquisition, and increasing the accuracy of line loss calculation.

[0144] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the terminals, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0145] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0146] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0147] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for dynamic adaptation of frame length in low-voltage distribution area acquisition, characterized in that, include: Collect real-time operational data from the transformer substation; Based on the real-time operation data of the transformer area, feature extraction is performed in combination with a preset statistical period to obtain the transformer area operation feature vector for the statistical period. Based on the historical clusters contained in the historical cluster set, the candidate cluster corresponding to the minimum distance is determined by the Mahalanobis distance between the operating feature vector of the transformer area and the historical cluster. If the Mahalanobis distance between the operating feature vector of the transformer area and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, the operating feature vector of the transformer area is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the operating feature vector of the transformer area. The historical cluster set contains multiple historical clusters, and the historical clusters are obtained by clustering based on the historical transformer area operation data according to different acquisition frame length requirement levels by the clustering model. In response to the frame length configuration command, clustering operations are performed using the clustering model based on the updated historical cluster set, and the target acquisition frame length requirement level is determined according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

2. The method for dynamic adaptation of low-voltage distribution area acquisition frame length according to claim 1, characterized in that, The formula for calculating the adaptive Mahalanobis distance threshold is: ； In the formula, The threshold for determining the Mahalanobis distance of the k-th cluster is... For degrees of freedom Significance level The corresponding chi-square distribution critical value, where d is the feature vector dimension and c is the preset small sample correction coefficient. This represents the current cumulative number of samples in the k-th cluster.

3. The method for dynamic adaptation of low-voltage distribution area acquisition frame length according to claim 1, characterized in that, After constructing a new historical cluster based on the operating feature vector of the transformer area, the following is also included: When the number of historical clusters exceeds a preset upper limit, calculate the normalized Euclidean distance and probability distribution similarity between any pair of historical clusters. Based on the normalized Euclidean distance and probability distribution similarity of the historical cluster pairs, and according to the preset judgment conditions, if the judgment conditions are met, the historical cluster pairs are merged to obtain merged historical clusters. The cluster feature vector of the merged historical cluster is obtained by weighted fusion based on the cluster feature vectors of the two historical clusters before merging, combined with weight coefficients. The weight coefficients are determined according to the sample size of the two historical clusters before merging. The acquisition frame length requirement level corresponding to the merged historical cluster inherits the higher level of the two historical clusters before merging.

4. The method for dynamic adaptation of low-voltage distribution area acquisition frame length according to claim 1, characterized in that, Also includes: A dynamic sliding time window is constructed using a smooth iterative formula based on distribution entropy; Based on the dynamic sliding time window, the set of operating feature vectors of the transformer area within the current range of the dynamic sliding time window is determined, and the set of operating feature vectors of the transformer area is clustered to obtain a temporary cluster set. Based on the temporary cluster set, calculate the centroid offset and dispersion rate of the temporary cluster set, wherein the centroid offset is the average Euclidean distance between the temporary cluster center and the reference cluster center in the current window, and the dispersion rate is the average ratio of the average sample variance of each cluster in the current window to the reference variance. The centroid offset and the rate of change of dispersion are weighted and fused to obtain the comprehensive drift score of the current dynamic sliding time window; According to the preset comprehensive drift score threshold and the upper limit of the number of windows exceeding the limit, when it is detected that the comprehensive drift score of multiple consecutive windows exceeds the comprehensive drift score threshold and the number of consecutive windows exceeding the limit reaches the upper limit of the number of windows exceeding the limit, a concept drift alarm message is output.

5. The low-voltage distribution area acquisition frame length dynamic adaptation method according to claim 4, characterized in that, When a concept drift alarm message is detected, the low-voltage substation's acquisition frame length configuration parameters are adjusted according to the target acquisition frame length requirement level, including: Based on the target acquisition frame length requirement level and the current acquisition frame length requirement level of the low-voltage substation, a frame length requirement level sequence is determined according to the preset acquisition frame length requirement level sorting information. The frame length requirement level sequence consists of each acquisition frame length requirement level from the current acquisition frame length requirement level to the target acquisition frame length requirement level. Based on the frame length requirement level sequence and combined with the preset adjustment cycle interval configuration information, the acquisition frame length configuration parameters of the low-voltage station area are adjusted sequentially in each adjustment cycle according to the next acquisition frame length requirement level until the acquisition frame length configuration parameters reach the standard corresponding to the target acquisition frame length requirement level.

6. The method for dynamic adaptation of low-voltage distribution area acquisition frame length according to claim 1, characterized in that, After adjusting the acquisition frame length configuration parameters of the low-voltage substation according to the target acquisition frame length requirement level, the following also includes: Monitor the acquisition success rate and channel load rate of the data acquisition equipment in the monitoring area; Based on the comparison results of the acquisition success rate, the channel load rate and the preset abnormal status judgment conditions, when the acquisition success rate or the channel load rate meets the abnormal status judgment conditions, the acquisition frame length configuration parameter of the low-voltage station area is reverted to the acquisition frame length requirement level before adjustment.

7. The method for dynamic adaptation of low-voltage distribution area acquisition frame length according to claim 1, characterized in that, Based on the historical cluster set, the candidate clusters corresponding to the minimum distances between the operating feature vectors of the transformer area and each historical cluster are determined, including: Based on the historical cluster set, the Euclidean distance between the operating feature vector of the transformer area and each historical cluster is calculated. Based on the Euclidean distance and combined with a preset initial screening distance threshold, each historical cluster is initially screened to obtain preliminary candidate clusters. Based on the preliminary candidate clusters, the Mahalanobis distance between the operating feature vector of the transformer area and each preliminary candidate cluster is calculated, and the preliminary candidate cluster corresponding to the minimum Mahalanobis distance is taken as the final candidate cluster.

8. A method for dynamic adaptation of low-voltage distribution area acquisition frame length according to any one of claims 1 to 7, characterized in that, Based on the real-time operation data of the transformer substation, and combined with a preset statistical period, feature extraction is performed to obtain the transformer substation operation feature vector for the statistical period, including: Based on the real-time operation data of the transformer area, feature extraction is performed in combination with a preset statistical period to obtain the operation characteristics of the transformer area in the statistical period. The operation characteristics of the transformer area include: load fluctuation rate, peak percentage, frequency of abnormal events, three-phase imbalance and new load penetration rate. The z-score normalization process is applied to the operating characteristics of each transformer area to obtain the operating characteristic vector of the transformer area.

9. A low-voltage distribution area acquisition frame length dynamic adaptation device, characterized in that, include: The transformer area data acquisition unit is used to collect real-time operating data of the transformer area; The feature vector extraction unit is used to extract features based on the real-time operation data of the transformer area and in combination with a preset statistical period to obtain the transformer area operation feature vector for the statistical period. The streaming incremental clustering update unit is used to determine the candidate cluster corresponding to the minimum distance between the transformer area operation feature vector and the historical clusters contained in the historical cluster set. If the Mahalanobis distance between the transformer area operation feature vector and the candidate cluster is less than the adaptive Mahalanobis distance threshold of the candidate cluster, the transformer area operation feature vector is merged into the candidate cluster, and the cluster feature vector of the candidate cluster is used. Otherwise, a new historical cluster is constructed based on the transformer area operation feature vector. The historical cluster set contains multiple historical clusters, and the historical clusters are obtained by the clustering model based on the historical transformer area operation data according to different acquisition frame length requirement levels. The frame length configuration instruction response unit is used to respond to the frame length configuration instruction, perform clustering operations based on the updated historical cluster set through the clustering model, and determine the target acquisition frame length requirement level according to the obtained clustering results, so as to adjust the acquisition frame length configuration parameters of the low-voltage station area according to the target acquisition frame length requirement level.

10. A low-voltage distribution area acquisition frame length dynamic adaptation terminal, characterized in that, include: Memory and processor; The memory is used to store program code, which corresponds to the low-voltage station area acquisition frame length dynamic adaptation method as described in any one of claims 1 to 8; The processor is used to read and execute the program code to implement the low-voltage distribution area acquisition frame length dynamic adaptation method.

Images