Electric energy meter time-sharing load monitoring method and system based on multi-dimensional electricity utilization characteristics

Abstract

The invention relates to the field of load monitoring, in particular to an electric energy meter time-sharing load monitoring method and system based on multi-dimensional electricity utilization characteristics, and the method comprises the steps: collecting and preprocessing the electrical time sequence data of a user electric energy meter; extracting multi-dimensional statistical characteristics of current and voltage of the users under a plurality of analysis windows, and clustering to obtain a cluster and membership degree of each user; for the target user, respectively calculating a first anomaly degree of the target user and a historical sequence thereof and a second anomaly degree of the target user and a sequence of other users in the same period in the cluster to which the target user belongs; according to the membership degree of the user under each analysis window, carrying out weighted fusion on the second anomaly degree to obtain a third anomaly degree; and calculating a comprehensive abnormality based on the first abnormality and the third abnormality, and determining whether the load is abnormal. According to the invention, through dual reference of individual history and a dynamic group and multi-scale credible fusion, the accuracy, the adaptive ability and the reliability of load anomaly monitoring are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of load monitoring. More specifically, this invention relates to a time-of-use load monitoring method and system for electricity meters based on multi-dimensional electricity consumption characteristics. Background Technology

[0002] With the advancement of smart grid construction, smart meters have been deployed on a large scale, enabling real-time collection of multi-dimensional time-series data such as voltage, current, power, and power factor. This has changed the traditional extensive model of monthly meter reading and provided a data foundation for refined load analysis.

[0003] However, most existing load analysis methods are still based on single or limited dimensions such as cumulative electricity consumption, failing to fully explore and integrate the deep electricity consumption characteristics contained in the high-dimensional, time-series data provided by smart meters. This results in an insufficiently refined and comprehensive characterization of user load characteristics, behavior patterns, and abnormal states.

[0004] Therefore, how to effectively extract composite features that accurately reflect the inherent characteristics and changing patterns of load from massive, multi-dimensional electricity consumption data to support advanced applications such as distribution network optimization, demand-side management, and abnormal electricity consumption identification has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problem of how to deeply mine and construct composite features that can accurately and comprehensively characterize user load characteristics from multi-dimensional time-series data collected by smart meters, this invention provides solutions in the following aspects.

[0006] In the first aspect, the time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics includes: Obtain electrical time-series parameters collected from the electricity meters of each user within the target area, and extract multi-dimensional electricity consumption feature vectors for each user under multiple preset time-length analysis windows; Cluster analysis is performed on all users based on the multidimensional electricity consumption feature vectors under a single time length analysis window to obtain clusters corresponding to each electricity consumption habit; and the membership degree of each user relative to its own cluster is calculated under each time length. Select any user as the target user and extract the real-time load sequence of the target user; within the same analysis window of the same time length, calculate the first anomaly between the real-time load sequence of the target user and the historical load sequence of the target user in the same period, and the second anomaly between the real-time load sequence of the target user and the load sequence of other users in the same period within the cluster to which the target user belongs. Based on the membership degree of the target user at each time length, the second anomaly degree under each analysis window is weighted and fused to obtain the third anomaly degree; Based on the first and third anomalies, the comprehensive anomaly of the target user is calculated, and the current load anomaly status of the target user is determined according to the comprehensive anomaly.

[0007] Preferably, the electrical timing parameters include at least one of current, voltage, and load at each sampling time.

[0008] Preferably, the time length includes a first length and a second length, wherein the first length is less than the second length; For the first length, construct an analysis window, where the endpoint of the analysis window is the current time and the length of the analysis window is the first length; The second length performs the same operation as the first length.

[0009] Preferably, the multidimensional electricity consumption feature vector includes: For each user, a set of identical statistical characteristics are calculated for their current and voltage time-series parameters under each analysis window. The statistical characteristic sets of current and voltage for each user under any analysis window are combined to form a comprehensive multi-dimensional electricity consumption characteristic vector.

[0010] Preferably, obtaining the first anomaly degree includes: Calculate the Pearson correlation coefficient between the real-time load sequence of the target user and the historical load sequence of the target user for the same period; the length of the load sequence for the same period is the same as the length of the real-time load sequence of the target user, and the ends of the lengths are the same; The first anomaly degree is obtained by using the Pearson correlation coefficient as the exponent of an exponential function and performing exponential calculation.

[0011] Preferably, obtaining the second anomaly degree includes: Identify the cluster to which the target user belongs and count the number of users in that cluster; Calculate the rate of change distribution of the real-time load sequence of the target user and the rate of change distribution of the real-time load sequences of other users within the cluster, respectively. For each user within the cluster, calculate the KL divergence between the rate of change distribution of its real-time load sequence and that of the target user's real-time load sequence. The second outlier is obtained by averaging the normalized KL divergence values ​​of all users within the cluster.

[0012] Preferably, obtaining the third anomaly degree includes: The weight of the analysis window corresponding to each time length is calculated based on the membership degree of the target user at each time length. The weight of the analysis window is multiplied by its corresponding second anomaly degree. Then, the results of multiplying all the analysis windows are added together to obtain the third anomaly degree.

[0013] Preferably, obtaining the overall anomaly degree includes: The composite anomaly is the product of the third anomaly and the mean of the first anomalies across all analysis windows.

[0014] Preferably, determining the current load anomaly status of the target user based on the real-time comprehensive anomaly level includes: When the overall anomaly level is greater than the preset safety threshold, the current target user load is determined to be abnormal, and an alarm is immediately issued; when the overall anomaly level is less than or equal to the preset safety threshold, the current target user load is determined to be in a normal state.

[0015] Secondly, a time-of-use load monitoring system for electricity meters based on multi-dimensional electricity consumption characteristics includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics described in any one of the claims is implemented.

[0016] The beneficial effects of this invention are: This invention effectively improves the accuracy, reliability, and context awareness of load anomaly monitoring by integrating multi-dimensional electrical characteristics across multiple time scales and innovatively introducing a dual reference and dynamic weighting mechanism of "individual history - similar groups." This method can simultaneously identify both short-term sudden anomalies and long-term behavioral drift; its modular design also enhances the system's interpretability and scalability, thus providing a solid data analysis foundation for advanced power applications such as load forecasting and energy efficiency management. Attached Figure Description

[0017] Figure 1 This is a flowchart of steps S1-S4 in the time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics according to an embodiment of the present invention.

[0018] Figure 2 This is a structural block diagram of the time-of-use load monitoring system for electricity meters based on multi-dimensional electricity consumption characteristics, according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0020] For ease of description, the physical area covered by all electricity meters powered by the same transformer in the same area is marked as the target area.

[0021] Reference Figure 1 The time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics includes steps S1-S4, as detailed below: S1: Obtain the electrical parameters collected by the electricity meters of each user in the target area.

[0022] In one embodiment, electrical timing parameters uploaded in real time by each user's electricity meter within the target area are collected through a communication network. The parameters at each sampling time include current, voltage, and power (load).

[0023] The raw parameters collected above are preprocessed, including data smoothing, linear interpolation to fill missing values, and normalization to eliminate dimensions.

[0024] After the above processing, the obtained electrical timing parameters provide a high-quality and reliable data foundation for subsequent load monitoring.

[0025] S2: Based on electrical parameters, extract multi-dimensional electricity consumption characteristics of each user under multiple preset time length analysis windows. Perform cluster analysis on all users for the multi-dimensional electricity consumption characteristics under a single time length analysis window to obtain clusters corresponding to each electricity consumption habit, and calculate the membership degree of each user relative to its own cluster under each time length.

[0026] A user's electricity consumption behavior is not a static label, but a complex function that dynamically changes with the time of day, the date of the week, and even the seasons. A residential user may exhibit a low-energy consumption pattern during the day on a weekday, while switching to a high-energy-consumption home mode at night; a commercial user exhibits an active business model during business hours. Therefore, using a single, fixed time window (such as the entire day) to characterize users inevitably loses information about the multi-scale, periodicity, and context-dependent nature of their behavior. Furthermore, the definition of a "normal" electricity consumption pattern is relative; it exists among user groups with similar lifestyles or production rhythms. Simply setting a global threshold to judge abnormalities is too coarse and cannot adapt to diverse electricity consumption scenarios.

[0027] Therefore, it is necessary to construct a multi-timescale feature system and conduct independent cluster analysis at each scale to dynamically characterize the user's behavior at different times, providing a multi-level and contextualized reference benchmark for subsequent anomaly monitoring.

[0028] In one embodiment, multiple time lengths are preset, such as a first length of 1 hour (to capture short-term transients and events), a second length of 24 hours (to depict intraday cyclical patterns), and a third length of 168 hours, i.e., one week (to reflect weekly cyclical habits).

[0029] The first, second, and third lengths mentioned above all use the current time as the endpoint, and extract electrical timing parameters of the corresponding lengths to form their respective analysis windows.

[0030] Next, we will perform the following operations using one of the analysis windows as an example (repeat the same operations for the other analysis windows): Within this analysis window, the same set of statistical characteristics are independently calculated for each user's current and voltage sequences (i.e., the sequences consisting of all sampling times within the analysis window), including mean, variance or standard deviation, kurtosis, skewness, and peak percentage (i.e., the proportion of time during which current / voltage is at a relatively high value within the analysis window).

[0031] The five characteristic values ​​of the current sequence and the five characteristic values ​​of the voltage sequence are further combined to form a 10-dimensional multidimensional electricity consumption characteristic vector.

[0032] Then, using the multidimensional electricity consumption feature vectors of all users within the target area as input, an unsupervised clustering algorithm such as K-means clustering is employed. Based on the spatial distance between the multidimensional electricity consumption feature vectors, the algorithm automatically divides all users into several clusters. Users in each cluster exhibit highly similar electricity consumption patterns within the analysis window, thus being defined as representing a typical electricity consumption habit.

[0033] After the above clustering is completed, for each cluster, the cluster center is calculated, and then the Euclidean distance from the multidimensional feature vector of each user in the cluster to the cluster center is calculated. The maximum value of the Euclidean distance from all users in the cluster to the cluster center is found. The ratio of the Euclidean distance of each user to the cluster center to the maximum value of the Euclidean distance is calculated. Finally, the negative of the ratio is calculated and normalized to obtain the membership degree of each user relative to its cluster in this analysis window.

[0034] S3: Select any user as the target user and extract the real-time load sequence of the target user; under the same time length analysis window, calculate the first anomaly degree between the real-time load sequence of the target user and the historical load sequence of the target user in the same period, and the second anomaly degree between the real-time load sequence of the target user and the load sequence of other users in the same period within the cluster to which the target user belongs. Based on the membership degree of the target user in each time length, the second anomaly degree under each analysis window is weighted and fused to obtain the third anomaly degree.

[0035] In load monitoring, abnormal conditions need to be examined from two interrelated but independent perspectives: first, vertically, whether there is a break or drift compared to its own historical normal (e.g., equipment failure causing a baseline drop, or electricity theft introducing new load); second, horizontally, whether it seems out of place compared to "neighbors" with similar living habits (e.g., a surge in nighttime electricity consumption in a single household while other similar households are quiet).

[0036] Simply examining a user's own history cannot uncover a new, but already spreading, abnormal pattern within a group; comparing only with other groups may misjudge a user who has consistently maintained unique but legitimate electricity usage habits. Therefore, a parallel, dual-channel anomaly detection mechanism must be established. Furthermore, a user's behavioral patterns at different time scales, such as hourly, daily, and weekly, contribute differently to the overall normality of their behavior.

[0037] A user whose electricity consumption fluctuates on an hourly scale but is normal on a daily scale represents a different type and severity of anomaly than a user whose electricity consumption is stable on an hourly scale but deviates significantly from historical patterns on a daily scale.

[0038] This necessitates a mechanism that can integrate users' deviations from their respective groups across different time scales, and this integration should reflect the stability of users' patterns at each scale.

[0039] In one embodiment, taking an analysis window as an example, any user is selected as the target user, and the Pearson correlation coefficient between the target user's real-time load sequence and the target user's historical load sequence for the same period is calculated.

[0040] If the length of the analysis window is the first length, in order to eliminate the influence of the daily cycle, the above-mentioned historical period can be defined as taking the same time period of yesterday. The second and third lengths are operated in the same way, so that the length of the historical period load sequence is the same as the length of the target user's real-time load sequence, and the end time of the length is the same.

[0041] Using the negative of the Pearson correlation coefficient as the exponent of the exponential function, we perform exponential calculations to obtain the first anomaly degree between the target user's real-time load sequence and its historical load sequence for the same period. This quantifies the degree to which the target user's current load behavior deviates from its historical norm. The first anomaly degree then satisfies the following relationship: In the formula, For the first degree of anomaly mentioned above, The Pearson correlation coefficient is the relationship between the real-time load sequence of the target user and the historical load sequence of the target user for the same period.

[0042] The larger the Pearson correlation coefficient between the two sequences, the more similar the two sequences are in terms of trend and morphology, and the lower the first anomaly. The smaller the Pearson correlation coefficient between the two sequences, the less correlated the two sequences are, and the higher the first anomaly.

[0043] In addition to calculating the Pearson correlation coefficient as described above, the Euclidean distance between the real-time load sequence of the target user and the historical load sequence of the target user (calculated using the L2 norm) can also be calculated. The calculated Euclidean distance is then normalized using the norm function to obtain the first anomaly degree between the real-time load sequence of the target user and the historical load sequence of the target user.

[0044] In one embodiment, the cluster to which the target user belongs is determined, and the number of users in the cluster is counted. The rate of change distribution of the real-time load sequence of the target user and the rate of change distribution of the real-time load sequences of other users in the cluster are calculated respectively. For each user in the cluster, the Kullback-Leibler divergence (relative entropy) between the rate of change distribution of the real-time load sequence of the target user and the real-time load sequence of the target user is calculated. The normalized values ​​of the Kullback-Leibler divergences of all users in the cluster are averaged to obtain the second anomaly degree between the real-time load sequence of the target user and the concurrent load sequences of other users in the cluster to which the target user belongs, thereby quantifying the degree of deviation of the target user's current dynamic behavior from the average dynamic pattern of the entire similar group.

[0045] Specifically, for the real-time load sequence of the target user, the difference or percentage change between adjacent time points in the sequence is calculated to obtain the rate of change sequence. Then, through histogram statistics or kernel density statistics, the rate of change sequence is fitted to a probability distribution, thereby obtaining the rate of change distribution of the real-time load sequence of the target user. Similarly, the rate of change distribution of the real-time load sequences of other users within the same cluster can be obtained.

[0046] The larger the KL divergence mentioned above, the less similar the target user's behavioral dynamics are to those of other users within the cluster.

[0047] After obtaining the second anomaly of a user relative to the concurrent load sequences of other users within the same cluster under multiple analysis windows, it is necessary to further integrate them into a comprehensive population anomaly index.

[0048] In one embodiment, the credibility of group deviations from different time scales is weighted based on the stability differences of users' own multi-scale behavioral patterns, i.e., the membership degree calculated above.

[0049] First, divide the membership degree of each analysis window by the sum of the membership degrees of all analysis windows to obtain the weight of each analysis window. Then, multiply the weight of each analysis window by its corresponding second anomaly degree. Finally, add the results of the multiplication of all analysis windows to obtain the third anomaly degree.

[0050] The above calculations show that the more typical (higher membership) a user's electricity consumption pattern is in an analysis window, the greater the weight of the group deviation signal (second anomaly) calculated in that analysis window will be in the final comprehensive group anomaly assessment.

[0051] In summary, the calculated first anomaly is used to monitor the user's own behavior patterns, the second anomaly is used to compare whether the user's behavior is synchronized with that of similar users in real time, and the calculation of the third anomaly organically integrates the group consistency characteristics from different analysis windows such as hours, days, and weeks into a unified indicator that represents the overall group deviation.

[0052] S4: Based on the first and third anomalies, calculate the comprehensive anomaly of the target user, and determine the current load anomaly status of the target user based on the comprehensive anomaly.

[0053] Through the above S3, two indicators were obtained that characterize anomalies from different dimensions: the first anomaly degree (individual historical deviation) and the third anomaly degree (multi-scale comprehensive group deviation). However, in real-world scenarios, a user may experience a significant change in their daily electricity consumption pattern compared to their history due to the purchase of a new appliance, i.e., a high first anomaly degree, but their electricity consumption curve remains similar to that of similar households that have also purchased new appliances, i.e., a low third anomaly degree. Conversely, a user who steals electricity may carefully maintain a total amount similar to their historical level, i.e., a low first anomaly degree, but their specific electricity consumption dynamics during different time periods are drastically different from those of similar households, i.e., a high third anomaly degree.

[0054] Therefore, combining evidence from these two dimensions leads to a more robust and interference-resistant final judgment.

[0055] In one embodiment, the mean of the first anomaly of the target user across all analysis windows is calculated, which represents the overall level of the user's "individual historical deviation".

[0056] Furthermore, the product of the third anomaly and the mean of the first anomalies under all analysis windows is taken as the comprehensive anomaly of the target user's real-time load sequence.

[0057] Furthermore, the calculated comprehensive anomaly degree is compared with a preset safety threshold. When the comprehensive anomaly degree is greater than the preset safety threshold, the current target user's load status is determined to be abnormal, and an alarm is immediately issued. The abnormal user information, abnormal time, anomaly degree value, and possible related characteristics (such as which scale is significantly abnormal) are notified to the power grid operators through methods such as monitoring screen pop-ups, sending text messages, pushing APP messages, or generating work orders. When the comprehensive anomaly degree is less than or equal to the preset safety threshold, the current target user's load is determined to be in a normal state. The system records the monitoring results and continues to perform cyclical monitoring for the next user or the next time period.

[0058] The aforementioned safety thresholds can be obtained through statistical analysis of historical normal data, such as taking the 99.5th percentile of the overall anomaly rate of normal samples, or they can be set by domain experts based on an acceptable false alarm rate.

[0059] This invention also provides a time-of-use load monitoring system for electricity meters based on multi-dimensional electricity consumption characteristics. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics according to the first aspect of the present invention is implemented.

[0060] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0061] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics, characterized in that, include: Obtain electrical time-series parameters collected from the electricity meters of each user within the target area, and extract multi-dimensional electricity consumption feature vectors for each user under multiple preset time-length analysis windows; Cluster analysis is performed on all users based on the multidimensional electricity consumption feature vectors under a single time length analysis window to obtain clusters corresponding to each electricity consumption habit; and the membership degree of each user relative to its own cluster is calculated under each time length. Select any user as the target user and extract the real-time load sequence of the target user; within the same analysis window of the same time length, calculate the first anomaly between the real-time load sequence of the target user and the historical load sequence of the target user in the same period, and the second anomaly between the real-time load sequence of the target user and the load sequence of other users in the same period within the cluster to which the target user belongs. Based on the membership degree of the target user at each time length, the second anomaly degree under each analysis window is weighted and fused to obtain the third anomaly degree; Based on the first and third anomalies, the comprehensive anomaly of the target user is calculated, and the current load anomaly status of the target user is determined according to the comprehensive anomaly.

2. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 1, characterized in that, The electrical timing parameters include at least one of the current, voltage, and load at each sampling time.

3. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 2, characterized in that, The time length includes a first length and a second length, wherein the first length is less than the second length; For the first length, construct an analysis window, where the endpoint of the analysis window is the current time and the length of the analysis window is the first length; The second length performs the same operation as the first length.

4. The time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics according to claim 3, characterized in that, The multidimensional electricity consumption feature vector includes: For each user, a set of identical statistical characteristics are calculated for their current and voltage time-series parameters under each analysis window. The statistical characteristic sets of current and voltage for each user under any analysis window are combined to form a comprehensive multi-dimensional electricity consumption characteristic vector.

5. The time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics according to claim 4, characterized in that, The acquisition of the first anomaly degree includes: Calculate the Pearson correlation coefficient between the real-time load sequence of the target user and the historical load sequence of the target user for the same period; the length of the load sequence for the same period is the same as the length of the real-time load sequence of the target user, and the ends of the lengths are the same; The first anomaly degree is obtained by using the Pearson correlation coefficient as the exponent of an exponential function and performing exponential calculation.

6. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 1, characterized in that, The acquisition of the second anomaly degree includes: Identify the cluster to which the target user belongs and count the number of users in that cluster; Calculate the rate of change distribution of the real-time load sequence of the target user and the rate of change distribution of the real-time load sequences of other users within the cluster, respectively. For each user within the cluster, calculate the KL divergence between the rate of change distribution of its real-time load sequence and that of the target user's real-time load sequence. The second outlier is obtained by averaging the normalized KL divergence values ​​of all users within the cluster.

7. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 1, characterized in that, The acquisition of the third anomaly degree includes: The weight of the analysis window corresponding to each time length is calculated based on the membership degree of the target user at each time length. The weight of the analysis window is multiplied by its corresponding second anomaly degree. Then, the results of multiplying all the analysis windows are added together to obtain the third anomaly degree.

8. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 1, characterized in that, The acquisition of the comprehensive anomaly degree includes: The composite anomaly is the product of the third anomaly and the mean of the first anomalies across all analysis windows.

9. The method for time-of-use load monitoring of electricity meters based on multi-dimensional electricity consumption characteristics according to claim 1, characterized in that, The current load anomaly status of a target user is determined based on real-time comprehensive anomaly assessment, including: When the overall anomaly level is greater than the preset safety threshold, the current target user load is determined to be abnormal, and an alarm is immediately issued; when the overall anomaly level is less than or equal to the preset safety threshold, the current target user load is determined to be in a normal state.

10. A time-of-use load monitoring system for electricity meters based on multi-dimensional electricity consumption characteristics, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions, which, when executed by the processor, implement the time-of-use load monitoring method for electricity meters based on multi-dimensional electricity consumption characteristics according to any one of claims 1-9.

Images