Method and system for monitoring electricity consumption abnormity based on intelligent fusion terminal
By constructing an electricity consumption database and combining time-series difference and discreteness calculations, suspected abnormal times are screened out. By using the isolated forest algorithm and the power feature value of clusters, the problem of high false alarm rate in traditional methods is solved, and accurate identification and low false alarm rate monitoring of electricity theft are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-04
- Publication Date
- 2026-04-14
AI Technical Summary
Traditional isolated forest algorithms are prone to misjudging electricity theft and sudden power outages in electricity consumption anomaly monitoring, resulting in a high false alarm rate and difficulty in accurately identifying electricity theft patterns.
By acquiring multi-dimensional electricity consumption data, an electricity consumption database is constructed and clustered. By combining time-series difference and dispersion calculation, suspected abnormal moments are screened out. Anomaly detection is performed using the isolated forest algorithm. By combining the power feature value and dispersion of the clusters, electricity theft is accurately identified.
It significantly reduces the false alarm rate, improves the accuracy and robustness of electricity theft identification, and can dynamically adjust the identification strategy in real time to accurately monitor abnormal electricity consumption.
Smart Images
Figure CN121860811A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electricity consumption anomaly monitoring. In particular, it relates to a method and system for monitoring electricity consumption anomalies based on an intelligent fusion terminal. Background Technology
[0002] With the improvement of my country's power system and the acceleration of urbanization, the number of electricity consumers and the scale of concentrated electricity consumption are constantly increasing. Correspondingly, the difficulty of manually monitoring circuits is also increasing rapidly. Therefore, intelligent integrated terminals are needed for real-time monitoring of large-scale concentrated electricity consumption areas. In order to improve the accuracy of intelligent integrated terminals in detecting electricity theft, it is necessary to intelligently analyze and judge user electricity consumption data to reduce false alarms.
[0003] When using the traditional Isolation Forest algorithm to monitor abnormal electricity consumption, outliers are marked as anomalies to detect unusual user behavior. However, during a user's electricity usage, activities such as electricity theft and sudden power outages can make the data appear outliers. Therefore, the Isolation Forest algorithm will detect both electricity theft and sudden power outages as abnormal. Since sudden power outages are normal user behavior, the traditional Isolation Forest algorithm can lead to false alarms in the electricity anomaly monitoring system. Summary of the Invention
[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.
[0005] In the first aspect, the method for monitoring abnormal electricity consumption based on intelligent fusion terminals includes: acquiring multi-dimensional electricity consumption data of users at any sampling time; filtering all multi-dimensional electricity consumption data within a preset sampling period to obtain suspected abnormal times; constructing a user's electricity consumption database; clustering the electricity consumption database to obtain several clusters; for any cluster, using the power mean of all multi-dimensional electricity consumption data as the power feature value of the cluster; filtering the clusters to obtain the cluster to which the multi-dimensional electricity consumption data belongs at a target time, where the target time is any suspected abnormal time; using the power feature value of the cluster to which the data belongs as the power feature value of the target time; iterating through each suspected abnormal time to obtain the power feature value; taking suspected abnormal times less than a preset power feature threshold as sub-times; calculating the abnormality of the sub-times; and completing the monitoring of abnormal electricity consumption based on the abnormality.
[0006] Preferably, obtaining the suspected abnormal time includes: taking the previous time adjacent to the target time as the adjacent time, calculating the difference between the multi-dimensional electricity consumption data of the adjacent time and the multi-dimensional electricity consumption data of the target time as the multi-dimensional difference value of the target time, using an isolated forest to perform anomaly detection on all multi-dimensional difference values of the preset sampling period to obtain the anomaly score of any sampling time, and taking the sampling time with the anomaly score greater than the preset anomaly score threshold as the suspected abnormal time.
[0007] Preferably, the construction of the user's electricity consumption database includes: taking the sampling time when the abnormal score is not greater than a preset abnormal score threshold as the normal time, and constructing the user's electricity consumption database from the multi-dimensional electricity consumption data of the normal time.
[0008] Preferably, the step of filtering clusters to obtain the cluster to which the multi-dimensional electricity consumption data at the target time belongs includes: for any cluster, obtaining the multi-dimensional electricity consumption data of the cluster center; taking any dimension as the target dimension, calculating the variance of the electricity consumption data at all sampling times in the cluster at the target dimension; calculating the square of the first difference between the electricity consumption data at the target time at the target dimension and the electricity consumption data at the cluster center at the target dimension; taking the ratio of the square of the first difference to the variance as the first ratio, iterating through all dimensions to obtain the first ratio, taking the square root of the sum of all the first ratios, and taking the result of inverse normalization of the square root as the similarity; iterating through the multi-dimensional electricity consumption data at the target time and obtaining the similarity between each cluster, and selecting the maximum similarity as the cluster to which the multi-dimensional electricity consumption data at the target time belongs.
[0009] Preferably, the calculation of the anomaly of a sub-time point includes: for any sub-time point, constructing a first window on the time axis with the sub-time point as the starting point and a preset size as the window length, and calculating the dispersion of the first window; randomly selecting a preset number of sampling times as reference times in the cluster to which the sub-time point belongs, constructing a second window for any reference time point based on the construction method of the first window, and obtaining the dispersion of any second window point based on the calculation method of the dispersion of the first window point; filtering to obtain the minimum dispersion among all the dispersion of the second windows, and taking the ratio of the minimum dispersion to the dispersion of the first window point as the anomaly of the sub-time point.
[0010] Preferably, the calculation of the dispersion of the first window includes: calculating the mean power value at all sampling times in the first window, calculating the square of the second difference between the power value at any sampling time in the first window and the mean power value, taking the square root of the average cumulative value of the squares of the second differences at all sampling times as the numerator, and using the ratio of the numerator to the mean power value as the dispersion of the first window.
[0011] Preferably, the step of monitoring electricity consumption anomalies based on anomalies includes: in response to an anomaly not exceeding a preset anomaly threshold, adding the multi-dimensional electricity consumption data at the sub-time to the electricity consumption database and continuing monitoring; in response to an anomaly exceeding the preset anomaly threshold, the multi-dimensional electricity consumption data at the sub-time is considered a true anomaly value, and generating and sending an alarm signal.
[0012] Secondly, a system for monitoring abnormal electricity consumption based on a smart fusion terminal 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 method for monitoring abnormal electricity consumption based on a smart fusion terminal as described in any one of the claims is implemented.
[0013] The present invention has the following effects: This invention analyzes the time-series characteristics of electricity consumption data to effectively capture abrupt signals caused by electricity theft. By analyzing power fluctuation characteristics, it identifies electricity theft patterns that are difficult to detect using traditional methods. Through clustering and anomaly analysis, combined with time-series difference and dispersion calculations, it accurately distinguishes between normal electricity consumption and electricity theft. It can still sensitively detect electricity theft switching behavior even in low power ranges, significantly reducing the false alarm rate, and can dynamically adjust the identification strategy in real time. Ultimately, this not only improves the accuracy and robustness of electricity theft identification but also contributes to precise and efficient monitoring of abnormal electricity consumption. Attached Figure Description
[0014] Figure 1 This is a flowchart of a method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to an embodiment of the present invention. Detailed Implementation
[0015] 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.
[0016] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0017] Reference Figure 1 The method for monitoring abnormal power consumption based on intelligent fusion terminals includes steps S1-S3, as follows: S1: Obtain multi-dimensional electricity consumption data of the user at any sampling time, and filter all multi-dimensional electricity consumption data within the preset sampling period to obtain suspected abnormal times.
[0018] It should be noted that electricity theft by users typically manifests as bypassing electricity metering through illegal means (such as illegally connecting lines before the meter, damaging the metering device, or installing switching control switches), transferring part of the load to off-meter power supply, and flexibly switching between theft and normal electricity consumption. It is worth noting that user electricity consumption data exhibits significant time-series characteristics, showing obvious daily periodicity and high autocorrelation. For example, daily electricity load is usually regularly distributed during morning and evening peak hours, daytime stable periods, and nighttime off-peak hours. However, traditional anomaly detection methods (such as the Isolation Forest algorithm) have limitations when processing this type of data: they only determine anomalies based on the degree of isolation of data points, making it difficult to effectively distinguish between persistent abnormal electricity consumption patterns caused by actual electricity theft and sudden normal electricity consumption fluctuations caused by temporary equipment start-ups and shutdowns, holidays, or special events.
[0019] Therefore, this invention uses the abnormal data initially identified by the isolated forest algorithm as suspected abnormal samples, and then combines the time series characteristics, periodic patterns and contextual information of electricity consumption behavior for secondary screening, so as to improve the accuracy and robustness of electricity theft identification, reduce the false alarm rate, and thus provide reliable support for power companies to conduct accurate audits.
[0020] In one embodiment, multi-dimensional electricity consumption data of the user at any sampling time is collected. For example, the multi-dimensional data includes current, voltage, power and electricity consumption. The number of dimensions can be increased or decreased by those skilled in the art according to the actual situation.
[0021] The preceding time point adjacent to the target time point is taken as the adjacent time point. The difference between the multi-dimensional electricity consumption data of the adjacent time point and the multi-dimensional electricity consumption data of the target time point is calculated as the multi-dimensional difference value of the target time point. Anomaly detection is performed on all multi-dimensional difference values of the preset sampling period using an isolation forest to obtain the anomaly score of any sampling time point. Sampling times with anomaly scores greater than a preset anomaly score threshold are considered as suspected anomaly times. For example, the preset anomaly score threshold is set to 0.9.
[0022] It needs to be explained that if a user engages in electricity theft and switches between normal and theft modes by installing a switching switch, the electricity consumption data recorded by the meter will exhibit distinct phased characteristics: under normal usage, electricity consumption typically follows historical load patterns and is relatively high; however, once switched to theft mode, some load is bypassed by the metering device, causing a significant drop in the recorded electricity consumption data. Furthermore, because the external power supply is unaffected by metering, the data during this phase often shows abnormally low values and abnormal stability (i.e., minimal fluctuations). Therefore, when a user switches from normal to theft mode, their electricity consumption sequence will experience a significant abrupt change or step change at the switching moment. To address this characteristic, this invention proposes differential processing of temporally adjacent electricity consumption data points (i.e., calculating the difference in electricity consumption at adjacent times) to amplify the abrupt signal caused by the theft switching behavior; subsequently, the differential result is fused with the original electricity consumption data, historical periodic averages, time-period characteristics, and other multi-dimensional features to construct a multi-dimensional time-series feature vector, and anomaly detection is performed based on this vector. This method can more sensitively capture sudden changes in electricity consumption patterns caused by switching operations, and effectively identify intermittent, switching electricity theft behaviors that are difficult to detect by traditional methods.
[0023] S2: Construct a user's electricity consumption database, cluster the electricity consumption database to obtain several clusters, and for any cluster, use the power mean of all multi-dimensional electricity consumption data as the power feature value of the cluster.
[0024] In one embodiment, according to step S1, the sampling time when the abnormal score is not greater than a preset abnormal score threshold is taken as the normal time, and the multi-dimensional electricity consumption data of the normal time is used to construct the user's electricity consumption database.
[0025] An iterative self-organizing clustering algorithm was used to cluster an electricity consumption database to obtain several clusters. This algorithm is an unsupervised clustering algorithm that dynamically adjusts the number of clusters and cluster centers through an iterative process, thereby achieving self-organized data classification. Clustering using this algorithm to obtain clusters is an existing technique and will not be elaborated upon here.
[0026] For any cluster, the power mean of all multi-dimensional electricity consumption data is used as the power feature value of the cluster. This power feature value needs to be normalized so that the value range of the power feature value of each cluster falls between 0 and 1.
[0027] S3: Filter the clusters to obtain the clusters to which the multi-dimensional electricity consumption data belongs at the target time belongs. The target time is any suspected abnormal time. Use the power feature value of the cluster to which it belongs as the power feature value of the target time. Iterate through each suspected abnormal time to obtain the power feature value. Take the suspected abnormal time that is less than the preset power feature threshold as the sub-time. Calculate the abnormality of the sub-time. Complete the electricity consumption anomaly monitoring based on the abnormality.
[0028] It should be noted that when a user engages in electricity theft, because some loads are bypassed from the meter, the recorded electricity consumption data usually shows a significant and continuous decrease in power, meaning the power is significantly lower than normal and fluctuates little. Based on this pattern, after initially identifying suspected abnormal moments, this invention does not directly determine the multi-dimensional electricity consumption data of the suspected abnormal moments as electricity theft. For each suspected abnormal moment, its similarity to each cluster is calculated, and the cluster with the highest similarity is selected as the corresponding cluster; then, the power characteristic value of the corresponding cluster is used as the power characteristic value of that suspected abnormal moment.
[0029] If the power characteristic value at the suspected abnormal moment is less than the preset power characteristic threshold, it indicates that there is indeed an anomaly at that point, and further verification is needed to determine whether it is electricity theft. Conversely, if the power characteristic value at the suspected abnormal moment is not less than the preset power characteristic threshold (i.e., the user should be in a high-load state during that period), then the multi-dimensional electricity consumption data at the suspected abnormal moment is normal, and there is no need to proceed with the subsequent complex electricity theft analysis process.
[0030] In one embodiment, for any cluster, multi-dimensional electricity consumption data of the cluster center is obtained; taking any dimension as the target dimension, the variance of the electricity consumption data of all sampling times in the cluster at the target dimension is calculated; the square of the first difference between the electricity consumption data of the target time at the target dimension and the electricity consumption data of the cluster center at the target dimension is calculated; the ratio of the square of the first difference to the variance is taken as the first ratio, and the first ratio of all dimensions is obtained by iterating through them; the square root of the sum of all the first ratios is taken, and the result of inverse normalization of the square root is taken as the similarity; the similarity between the multi-dimensional electricity consumption data at the target time and each cluster is obtained by iterating through them, and the maximum similarity is selected as the cluster to which the multi-dimensional electricity consumption data at the target time belongs.
[0031] The power feature value of the cluster to which it belongs is used as the power feature value of the target time. The power feature value of each suspected abnormal time is obtained by traversing the cluster. Suspected abnormal times that are less than the preset power feature threshold are taken as sub-times. For example, the preset power feature threshold is set to 0.5.
[0032] For any sub-time point, a first window is constructed on the time axis, starting from the sub-time point and with a preset window length. The dispersion of the first window is calculated. A preset number of sample times are randomly selected from the cluster to which the sub-time point belongs as reference times. A second window for any reference time is constructed using the same method as the first window, and the dispersion of any second window is obtained using the same method as the first window. The minimum dispersion among all the dispersion values of the second windows is selected, and the ratio of the minimum dispersion value to the dispersion value of the first window is taken as the outlier of the sub-time point. For example, the window size of the first and second windows is set to 20, and the number of reference times is set to 10.
[0033] The calculation of the dispersion of the first window includes: calculating the mean power value at all sampling times in the first window; calculating the square of the second difference between the power value at any sampling time in the first window and the mean power value; taking the square root of the average cumulative value of the squares of the second differences at all sampling times as the numerator; and using the ratio of the numerator to the mean power value as the dispersion of the first window.
[0034] It should be noted that during normal electricity consumption, even in low-power states, the power consumption curves of normally on basic appliances such as refrigerators, routers, and security equipment will still exhibit a certain degree of natural fluctuation due to periodic start-stop or load fluctuations, resulting in some dispersion over time. In contrast, when a user engages in electricity theft (such as transferring part of the load off the meter via a switching switch), the meter only records the remaining, typically very low and stable, load within the meter, leading to abnormally smooth and almost unfluctuated electricity consumption data with a significantly lower dispersion than under normal low-load conditions. Based on this key difference, this invention selects sub-time points and further calculates the dispersion of the first window; simultaneously, it selects multiple reference times from the relevant cluster and calculates the dispersion of the second window for each reference time, taking the minimum value as the lower limit of dispersion under normal low-load conditions. If this minimum dispersion is still much greater than the dispersion of the sub-time point, it indicates that the electricity consumption behavior is highly inconsistent with the normal electricity fluctuation pattern of users, thus being identified as a genuine anomaly, and further inferring that the user is engaging in electricity theft.
[0035] It's important to explain that while both off-peak electricity consumption and electricity theft occur at relatively low power levels, the power consumption curve still shows significant fluctuations during off-peak hours due to the starting and stopping of household appliances. Electricity theft, essentially reducing electricity consumption on the meter to lower bills, also suffers from this problem because the starting and stopping of appliances is less noticeable on the curve. Consequently, electricity theft is reflected on the power consumption curve as a sharp drop in power consumption, followed by a smoother curve after the initial spike, without significant fluctuations compared to normal low-power consumption.
[0036] If the anomaly is not greater than the preset anomaly threshold, the multi-dimensional electricity consumption data at the sub-time point is added to the electricity consumption database and monitoring continues; if the anomaly is greater than the preset anomaly threshold, the multi-dimensional electricity consumption data at the sub-time point is considered a true anomaly, and an alarm signal is generated and sent.
[0037] The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the method for monitoring abnormal power consumption based on an intelligent fusion terminal according to the first aspect of the present invention.
[0038] 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.
[0039] 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 method for monitoring abnormal electricity consumption based on an intelligent fusion terminal, characterized in that, include: Acquire multi-dimensional electricity consumption data of users at any sampling time, and filter all multi-dimensional electricity consumption data within the preset sampling period to obtain suspected abnormal times; Construct a user electricity consumption database, cluster the electricity consumption database to obtain several clusters, and for any cluster, use the power mean of all multi-dimensional electricity consumption data as the power feature value of the cluster. Clusters are filtered to obtain the clusters to which the multi-dimensional electricity consumption data belongs at the target time. The target time is any suspected abnormal time. The power feature value of the cluster to which it belongs is used as the power feature value of the target time. The power feature value of each suspected abnormal time is obtained by traversing. Suspected abnormal times with power feature values less than a preset threshold are taken as sub-times. The abnormality of the sub-times is calculated. Electricity consumption anomaly monitoring is completed based on the abnormality.
2. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 1, characterized in that, The moments at which suspected anomalies are obtained include: The previous time adjacent to the target time is taken as the adjacent time. The difference between the multi-dimensional electricity consumption data of the adjacent time and the multi-dimensional electricity consumption data of the target time is calculated as the multi-dimensional difference value of the target time. Anomaly detection is performed on all multi-dimensional difference values of the preset sampling period using isolated forest to obtain the anomaly score of any sampling time. The sampling time with the anomaly score greater than the preset anomaly score threshold is taken as the suspected anomaly time.
3. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 2, characterized in that, The construction of the user's electricity consumption database includes: The sampling time when the abnormal score is no greater than the preset abnormal score threshold is taken as the normal time, and the multi-dimensional electricity consumption data of the normal time is used to build the user's electricity consumption database.
4. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 1, characterized in that, The clusters to which the multi-dimensional electricity consumption data at the target time belongs, obtained by filtering the clusters, include: For any cluster, obtain multi-dimensional electricity consumption data of the cluster center; Using any dimension as the target dimension, calculate the variance of electricity consumption data at all sampling times in the cluster along the target dimension; Calculate the squared first difference between the electricity consumption data at the target time in the target dimension and the electricity consumption data of the cluster centers in the target dimension; The ratio of the square of the first difference to the variance is taken as the first ratio. The first ratio of all dimensions is obtained by iterating through them. The sum of all the first ratios is squared to obtain the square root value. The square root value is then inversely normalized to be used as the similarity. The similarity between the multi-dimensional electricity consumption data at the target time and each cluster is obtained through iteration. The cluster with the highest similarity is selected as the cluster to which the multi-dimensional electricity consumption data at the target time belongs.
5. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 1, characterized in that, The anomalies of the computation sub-time include: For any sub-time point, starting from the sub-time point and with a preset size as the window length, construct a first window on the time axis and calculate the degree of dispersion of the first window; A preset number of sampling times are randomly selected from the cluster to which the sub-time belongs as reference times. The second window for any reference time is constructed in the same way based on the construction method of the first window, and the discreteness of any second window is obtained in the same way based on the calculation method of the discreteness of the first window. The minimum dispersion value among all the dispersion values of the second window is obtained by filtering, and the ratio of the minimum dispersion value to the dispersion value of the first window is taken as the outlier of the sub-time.
6. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 5, characterized in that, The degree of dispersion of the first window is calculated as follows: Calculate the mean power value at all sampling times in the first window. Calculate the square of the second difference between the power value at any sampling time in the first window and the mean power value. Take the square root of the average sum of the squares of the second differences at all sampling times as the numerator. The ratio of the numerator to the mean power value is taken as the degree of dispersion of the first window.
7. The method for monitoring abnormal electricity consumption based on an intelligent fusion terminal according to claim 1, characterized in that, The process of monitoring abnormal electricity consumption based on anomalies includes: If the anomaly is not greater than the preset anomaly threshold, the multi-dimensional electricity consumption data at the sub-time point will be added to the electricity consumption database for continued monitoring. When an anomaly exceeds a preset anomaly threshold, and the multi-dimensional electricity consumption data at the sub-time is considered a true anomaly, an alarm signal is generated and sent.
8. A system for monitoring abnormal electricity consumption based on an intelligent fusion terminal, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the method for monitoring abnormal power consumption based on a smart fusion terminal according to any one of claims 1-7.