A meter box electric energy anomaly determination method

Abstract

The application provides a kind of metering box electric energy anomaly determination method, it is related to low-voltage distribution network monitoring technical field.The core is to collect the electrical data of transformer and each metering box synchronously, dynamically construct and real-time update topological correlation matrix, and extract multi-dimensional electrical characteristics.According to the energy conservation relation, the energy conservation deviation vector is calculated, and the pre-constructed abnormal physical pattern library is matched to accurately identify the abnormal type.The application breaks through the limitation of traditional time domain analysis, can effectively mine hidden abnormal signals, and significantly improves the accuracy of abnormal identification.Combining dynamic topology identification and frequency energy spectrum analysis, the misdiagnosis and missed diagnosis problems caused by topological information lag or weak abnormal characteristics are solved.The complex time domain deviation signal is converted to the frequency domain for pattern matching, which can distinguish different abnormal reasons from the physical mechanism level, and realize the leap from discovering abnormality to diagnosing type.

Description

Technical Field

[0001] This invention relates to the field of low-voltage power distribution network monitoring technology, specifically a method for determining power anomalies in metering boxes. Background Technology

[0002] With the deepening of smart grid construction, the data acquisition capabilities of low-voltage distribution substations have been significantly enhanced, and smart meters on the transformer outgoing side and user side have basically achieved full coverage. However, the intermediate link between the transformer and the end user—the metering box—has long lacked effective online monitoring methods due to its large number, wide distribution, and complex operating environment. The metering box houses multiple energy meters from the same power supply area, and the electrical characteristics of its incoming side can comprehensively reflect the electricity consumption patterns of users in that area as well as the operational health of the metering devices themselves. Therefore, conducting energy anomaly analysis at the metering box level is of great significance for ensuring the accuracy of metering in low-voltage distribution networks, preventing electricity theft, and improving operation and maintenance efficiency.

[0003] Currently, detection methods for power anomalies in low-voltage distribution networks mainly focus on two levels: one is based on the main meter data on the transformer outgoing side, analyzing abnormal fluctuations in line losses in the distribution area to discover potential electricity theft or metering faults; the other is based on the data of individual user meters, analyzing the abrupt changes in historical curves of voltage, current, and power to identify anomalies. However, existing methods have significant limitations. First, the analysis method based on the main meter line loss in the distribution area can only locate the problematic area, unable to accurately identify which metering box or user meter is malfunctioning, resulting in a huge workload for on-site investigation. Second, the analysis method based on individual meters often ignores the mutual influence between downstream meters and changes in the line topology. When the anomaly characteristics are weak or multiple anomaly sources overlap, it is difficult to accurately separate the true cause of the anomaly. Traditional analysis methods mostly rely solely on changes in voltage and current amplitudes, resulting in low detection accuracy when faced with complex, non-periodic anomalies.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for determining abnormal electrical energy in a metering box, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for determining abnormal power consumption in a metering box, comprising the following steps: Step 1: Collect voltage and current data from the transformer outgoing side and the metering box incoming side of the target low-voltage distribution area. Based on the voltage data, determine the phase difference between the transformer outgoing side and the metering box incoming side, and construct a real-time topology correlation matrix between the transformer and the metering box by combining the phase difference and voltage change relationship. Step 2: When a phase difference or voltage change is detected, update the real-time topology correlation matrix. Based on the real-time topology correlation matrix, obtain the electrical characteristics of the metering box, and calculate the difference sequence between the incoming power of each metering box and the sum of the power measured by the power meter in the metering box according to the energy conservation relationship, so as to form an energy conservation deviation vector. Step 3: Perform frequency domain analysis on the energy conservation deviation vector to obtain frequency domain energy characteristics, and match the frequency domain energy characteristics with a pre-built library of anomalous physical patterns to determine the anomaly type; Step 4: After matching and determining the anomaly type, determine the upstream and downstream nodes of the target metering box based on the real-time topological correlation matrix, eliminate the influence of upstream and downstream nodes on the energy conservation deviation, separate the anomaly contribution of the target metering box, and output the anomaly analysis results.

[0007] Furthermore, voltage and current sampling data from the transformer outgoing side and the incoming side of each metering box are acquired, and the voltage and current sampling data are synchronized and aligned in time. The aligned sampled data is divided into multiple feature statistical windows according to the preset time segmentation rules; Within each of the aforementioned feature statistics windows, voltage characteristic parameters for phase difference calculation are determined based on voltage sampling data. These voltage characteristic parameters include at least voltage amplitude parameters and phase angle parameters characterizing the phase.

[0008] Furthermore, the phase difference calculation and the generation of the real-time topological correlation matrix include the following steps: Within each feature statistics window, the voltage phase angle on the transformer outgoing side is used as the reference phase. The difference between the voltage phase angle on the incoming side of each metering box and the reference phase is calculated to obtain the phase difference of each metering box within the feature statistics window. The metering boxes are sorted according to the size of the phase difference, and the sorted metering boxes are divided into at least two phase difference partitions based on the preset phase difference partition threshold. Each phase difference partition corresponds to a different feeder branch on the transformer outgoing line side. The phase difference partition threshold is adaptively determined according to the statistical distribution of the phase difference. For metering boxes that fall into the same phase difference zone, their relative position order in the corresponding feeder branch is determined according to the magnitude of the voltage amplitude on the incoming side of the metering box. Based on the feeder branch affiliation relationship corresponding to the phase difference partition and the relative position order within the same partition, a real-time topology association matrix describing the connection relationship between the transformer outgoing side and each metering box is constructed. Changes in the time series of phase difference and / or voltage amplitude are detected, and when a change is detected that meets preset conditions, an update of the real-time topology correlation matrix is ​​triggered.

[0009] Furthermore, the real-time update of the topological association matrix includes the following steps: When a change detection triggers an update, the phase difference of the target metering bin that has changed within one or more feature statistics windows after the update is recalculated. The recalculated phase difference is compared with the phase difference partition threshold: if the recalculated phase difference crosses the partition boundary, it is determined that the feeder branch affiliation of the target metering box has changed, and the upper-level association information of the target metering box in the real-time topology association matrix is ​​updated. If the recalculated phase difference does not cross the partition boundary, the relative position order of the target metering box in its feeder branch in the real-time topology association matrix is ​​updated according to whether the voltage amplitude change of the target metering box meets the preset conditions, while keeping its feeder branch affiliation unchanged.

[0010] Furthermore, within one or more power frequency cycles corresponding to the feature statistics window, the voltage sampling data and current sampling data are decomposed in the frequency domain to obtain the fundamental component and the amplitude parameters corresponding to at least one set of preset harmonic components. The active power corresponding to the fundamental component and harmonic components is calculated, and the ratio of the active power of the harmonic components to the active power of the fundamental component is processed to obtain the harmonic content rate sequence. Based on the feeder branch attribution relationship determined by the real-time topology association matrix, the harmonic content rate sequences of each metering box within the same feeder branch are compared horizontally, and metering boxes whose harmonic content rate deviates from the statistical level of the feeder branch by more than the preset condition are identified as candidates for harmonic anomalies. Obtain the three-phase voltage amplitude and three-phase current amplitude, and calculate the negative sequence unbalance and zero sequence unbalance based on the symmetrical components; Within the feature statistics window, statistical analysis is performed on negative-order imbalance and zero-order imbalance to obtain feature vectors characterizing their dynamic evolution. The feature vectors include at least the change magnitude feature and the change persistence feature. Obtain the total active power time series on the incoming side of the metering box and the active power time series of each sub-meter in the metering box, and perform normalization processing. The correlation between the total active power time series and the active power time series of each sub-meter is calculated using a sliding time window to obtain the correlation sequence; Within each feature statistical window, statistical analysis is performed on the correlation sequences falling into that window to obtain at least two indicators for quantifying the synchronicity of load fluctuations. The indicators include at least the correlation mean indicator and the correlation dispersion indicator. The electrical characteristics include harmonic anomaly candidates, feature vectors characterizing the dynamic evolution of negative-sequence and zero-sequence unbalance, and indices used to quantify load fluctuation synchronicity.

[0011] Furthermore, the formation of the energy conservation deviation vector specifically includes: For each characteristic statistical window, the cumulative value of active power on the incoming side of the metering box within that characteristic statistical window is obtained as the total incoming energy; and the cumulative values ​​of active power of each energy meter in the metering box within that characteristic statistical window are summed to obtain the sum of the energy of each meter. Based on the metering box connection relationship and / or relative position within the branch determined by the real-time topology correlation matrix, the line loss energy used for energy correction is determined, and the line loss energy is deducted from the total incoming line energy as a fixed line loss energy. The energy difference between the total incoming energy, the energy measured by the sub-meter, and the fixed line loss energy is processed to obtain the energy conservation deviation value within the characteristic statistical window; The energy conservation deviation values ​​corresponding to multiple feature statistical windows are arranged in chronological order to form an energy conservation deviation vector, where the positive or negative sign of the deviation value is used to characterize the direction of the deviation.

[0012] Furthermore, the extraction of the frequency domain energy features includes: The energy conservation deviation vector is decomposed into multiple frequency bands by performing time-frequency transformation on the energy conservation deviation vector. Calculate the energy value of each frequency band within each characteristic statistical window, construct a frequency domain energy spectrum distribution map of energy variation with frequency, and store it in matrix form, where the matrix elements are used to characterize the energy value of the corresponding frequency band within the corresponding characteristic statistical window; The preset frequency range is divided into multiple specific frequency bands, and the energy values ​​within each specific frequency band are aggregated to obtain the frequency domain energy characteristics corresponding to each specific frequency band.

[0013] Furthermore, the construction and matching of the abnormal physical pattern library includes: The corresponding sample energy conservation deviation vectors are obtained in advance for various metering box anomaly types and converted into frequency domain energy spectrum distribution maps to extract the standard frequency domain energy features corresponding to each anomaly type, thereby establishing the correspondence between anomaly types and standard frequency domain energy features and forming an anomaly physical pattern library. The frequency domain energy characteristics of the metering box under test are extracted to form the feature vector to be measured; the standard feature vectors corresponding to each anomaly type are retrieved from the abnormal physical pattern library; Calculate the similarity between the feature vector to be tested and each standard feature vector, and determine the anomaly type that matches the measuring box to be tested based on the similarity results.

[0014] Furthermore, the topological location of the target metering box is determined in the real-time topological correlation matrix, and the upstream node set of the target metering box is determined by tracing upwards based on the real-time topological correlation matrix. The upstream node set includes nodes located on the power supply path from the power supply side to the target metering box; at the same time, the downstream node set of the target metering box is determined. For the same feature statistical window, extract the frequency domain energy features of the target metering box, and then extract the frequency domain energy features of each upstream node in the upstream node set in turn. When the abnormal feature value of an upstream node and the abnormal feature value of the target meter box meet the preset similarity conditions, the ratio of the abnormal feature value of the upstream node and the abnormal feature value of the target meter box is processed to obtain the contribution coefficient of the upstream node to the abnormality of the target meter box, and the contribution amount of the upstream node is obtained according to the contribution coefficient. The total upstream contribution is obtained by summing the contributions of each upstream node, and then the total upstream contribution is deducted from the abnormal characteristic value of the target metering box to obtain the net characteristic value after removing the influence of upstream nodes.

[0015] Furthermore, the anomaly analysis results, after eliminating the influence of downstream nodes, include: Obtain the electrical characteristics of each downstream energy meter in the downstream node set within the same feature statistics window, and calculate the correlation between the electrical characteristics of each downstream energy meter and the energy conservation deviation vector of the target metering box. When the correlation meets the preset conditions, the corresponding downstream electricity meter is identified as a suspected abnormal downstream node, and the influence of the suspected abnormal downstream node on the energy conservation deviation vector of the target meter box is estimated. The total downstream impact value is obtained by summing the impact of each suspected abnormal downstream node, and then the total downstream impact value is subtracted from the net feature value to obtain the feature value corresponding to the abnormal contribution of the target metering box itself. Output the topological location information of the target metering box, the anomaly type determination result, the total upstream contribution, and the total downstream impact value.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention utilizes a dynamic topology identification mechanism to perceive changes in the connection relationship between the metering box and the transformer in real time and automatically update the topology correlation matrix. This provides a precise electrical connection path basis for anomaly analysis, avoiding misdiagnosis or missed diagnosis due to lagging topology information. Furthermore, this invention integrates frequency domain energy spectrum analysis of multi-dimensional electrical characteristics and energy conservation deviation vectors. This allows for in-depth mining of hidden anomaly signals that are difficult to capture using traditional time-domain analysis. By matching with a pre-built anomaly physical pattern library, it achieves a leap from anomaly detection to type diagnosis, significantly improving the accuracy of anomaly identification. When an anomaly is determined in the metering box, the contributions of upstream and downstream nodes can be traced based on real-time topology relationships, and the anomaly portion of the metering box itself can be separated. This enables precise allocation of responsibility for the anomaly, effectively solving the problems of existing technologies in accurately locating the source of anomalies and separating the main cause in scenarios with multiple anomaly sources. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a schematic diagram comparing the harmonic content of the present invention; Figure 3 This is a schematic diagram of the dynamic evolution trajectory of the three-phase imbalance in this invention; Figure 4 This is a bar chart illustrating the load fluctuation synchronicity of the present invention; Figure 5 This is a schematic diagram of the load fluctuation synchronization of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figures 1-5The present invention provides a technical solution: A method for determining abnormal power consumption in a metering box, comprising the following steps: Step 1: Collect voltage and current data from the transformer outgoing side and the incoming side of each metering box in the target low-voltage distribution area. Based on the voltage data, determine the phase difference between the transformer outgoing side and the incoming side of each metering box, and construct a real-time topology correlation matrix between the transformer and each metering box by combining the phase difference and voltage change relationship.

[0021] In this embodiment, intelligent measurement units with synchronous phasor measurement functions are deployed on the transformer outgoing side and the incoming side of each metering box in the target low-voltage distribution substation. The core of these intelligent measurement units lies in their built-in high-precision timing modules, such as receiver modules based on the Global Positioning System (GPS) or the BeiDou Navigation Satellite System. Each intelligent measurement unit simultaneously collects instantaneous voltage and current values ​​according to the same set sampling start time and sampling frequency, and marks each collected instantaneous voltage and current value with a corresponding timestamp. The timestamp accuracy should be at least at the microsecond level to ensure the accuracy of subsequent power frequency period division and phase calculation. The sampling frequency setting must follow the Nyquist sampling theorem; considering the highest order of harmonics to be analyzed (e.g., at least the 19th harmonic), the sampling frequency should be at least 40 times the power frequency. For example, for a 50Hz power frequency system, if it is necessary to analyze the 19th harmonic (950Hz), the sampling frequency can be set to 2560Hz or 5120Hz, that is, 128 points or 256 points are sampled in each power frequency cycle (20ms) to ensure accurate waveform reconstruction.

[0022] In this embodiment, the instantaneous voltage value collected by the intelligent measurement unit on the transformer output side is first used to detect the voltage zero-crossing point of each power frequency cycle. For example, linear interpolation can be used to accurately determine the moment when the instantaneous value equals 0 when the voltage waveform transitions from the negative half-cycle to the positive half-cycle between sampling points. The time interval between two adjacent zero-crossing points in the same direction (e.g., the zero-crossing point from negative to positive) is one power frequency cycle (approximately 20ms for a 50Hz system). Then, using the first voltage zero-crossing point on the transformer output side as the reference time point for the entire system, the continuous time axis is divided into a series of continuous, non-overlapping time segments. Each such time segment is a characteristic statistical window. The length of the characteristic statistical window (i.e., the number of power frequency cycles it contains) is a configurable parameter, and its selection needs to strike a balance between time resolution and statistical stability. For example, the length of the characteristic statistical window can be set to 15 minutes, corresponding to 45,000 power frequency cycles of a 50Hz power frequency system. For residential areas with relatively stable load changes, a window length of 30 minutes or even 1 hour can be used to reduce the computational load for data storage and analysis. For industrial and commercial users or areas with rapidly fluctuating loads, the window length can be shortened to 5 minutes or 1 minute to improve the response capability to rapid changes and anomalies. Finally, based on the high-precision timestamp carried by each voltage and current instantaneous value sampling point, they are divided into the corresponding power frequency cycle and characteristic statistical window.

[0023] The root mean square (RMS) calculation is performed on the sequence of voltage sampling data to obtain the voltage amplitude for that power frequency cycle. The specific calculation method for the voltage amplitude is as follows: For a power frequency cycle containing N sampling points, let the sequence of voltage sampling data be... If n=1,2,…,N, then its root mean square value (i.e., voltage amplitude) The formula for calculating ) is: Where n is the index of the sampling point within the power frequency cycle, and N is the number of sampling points within one power frequency cycle, such as 128 or 256 mentioned above.

[0024] A Fourier transform is performed on the sequence of voltage sampling data under each power frequency cycle to extract the initial phase angle of the voltage fundamental phasor for the corresponding power frequency cycle.

[0025] In this embodiment, the Fourier transform is typically implemented using the Discrete Fourier Transform (DFT) or its fast algorithm (FFT). Through DFT, the voltage and current signals in the time domain can be decomposed into phasor forms of each frequency component, namely amplitude and initial phase angle. Taking voltage as an example, for a sampling sequence of one power frequency cycle... The complex form of its fundamental component (i.e., the component with a frequency of 50 Hz). It can be calculated using the following formula: in, The imaginary unit satisfies Calculated The argument is the initial phase angle of the fundamental voltage phasor. .

[0026] It should be noted that, since the result of the Fourier transform depends on the selected time window, in order to eliminate the spectral leakage effect and ensure the accuracy of the calculation, the sampling data window of each power frequency cycle must be strictly synchronized with the actual power frequency cycle. In this invention, it is precisely through the aforementioned power frequency cycle division based on the voltage zero-crossing point that it is ensured that each sampling sequence participating in the DFT calculation corresponds exactly to a complete power frequency cycle, thereby achieving synchronous sampling of the DFT.

[0027] Step 2: When a phase difference or voltage change is detected, update the real-time topology correlation matrix. Based on the real-time topology correlation matrix, obtain the electrical characteristics of the metering box, and calculate the difference sequence between the incoming power of each metering box and the sum of the power measured by the power meter in the metering box according to the energy conservation relationship, so as to form an energy conservation deviation vector.

[0028] In this embodiment, the phase difference calculation logic is as follows: For each power frequency cycle, the initial phase angle of the fundamental phasor of the transformer output voltage in that power frequency cycle is taken as the reference phase; and the difference between the initial phase angle of the fundamental phasor of the input voltage of each metering box and the reference phase is calculated in that power frequency cycle to obtain the phase difference of each metering box in that power frequency cycle.

[0029] Specifically, let the initial phase angle of the fundamental phasor of the transformer output voltage be in the k-th power frequency cycle. The initial phase angle of the fundamental phasor of the voltage on the input side of the i-th metering box is Then, under this power frequency cycle, the phase difference between the metering box and the transformer is... The calculation formula is: This difference is usually positive because current flowing through the line causes a voltage drop, resulting in the voltage phase at the end lagging behind that at the beginning. However, a small negative value may also appear for very short lines. In practical applications, its absolute value can be taken or it can be adjusted by adding... Normalize it to the principal value range (e.g., 0 to 1) in a way that allows it to be normalized. Within ) The physical significance of phase difference lies in the fact that it comprehensively reflects the combined effect of reactive power flow and line impedance, and is an effective indicator for measuring electrical distance.

[0030] The specific logic for dynamically generating the real-time topological association matrix is ​​as follows: The metering boxes are sorted based on their phase differences from smallest to largest. This is because, at any given moment, the metering box that is electrically closer to the transformer has a smaller phase difference between its voltage phase and the voltage phase on the transformer's output side. Therefore, by sorting the phase differences of all metering boxes (the average of the most recent several cycles can be selected to eliminate random errors) from smallest to largest, this sorting result initially reflects the electrical proximity of each metering box to the transformer.

[0031] Based on the physical characteristics of the power distribution line, a phase difference partitioning threshold is set, and the sorted metering boxes are divided into multiple phase difference partitions. Each partition corresponds to a feeder branch on the transformer outgoing side. The phase difference partitioning threshold is adaptively determined by statistical analysis of the phase difference data of each metering box.

[0032] The phase difference zoning threshold is the key boundary for distinguishing different feeder branches. Since each feeder branch has different line type, length, and load characteristics, the resulting phase difference variation range also differs. Therefore, a fixed threshold cannot be used; an adaptive method must be employed to determine it. There are several specific methods for adaptively determining the threshold; this embodiment provides the following two preferred schemes: Option 1: Cluster analysis-based method: First, phase difference data for all metering boxes within a feature statistics window are acquired, forming a one-dimensional dataset. Then, a density-based clustering algorithm (such as DBSCAN) is used to cluster them. The advantage of the DBSCAN algorithm is that it does not require pre-specifying the number of clusters, can identify clusters of arbitrary shapes, and effectively identify noise points. Its core parameters include the neighborhood radius and the minimum number of points. The neighborhood radius can be determined based on the statistical value of the phase difference measurement accuracy; for example, it can be set to 0.5 times the standard deviation of the phase differences of all metering boxes. The minimum number of points can be set based on the expected minimum number of metering boxes on a feeder branch; for example, it can be set to 2 or 3. After the clustering algorithm runs, it automatically groups metering boxes with similar phase difference values ​​into one class, with each class corresponding to a feeder branch. The boundary value of each cluster (e.g., the average of the maximum and minimum phase differences in the cluster, or with a protection margin) can be used as the phase difference partitioning threshold for that feeder branch.

[0033] The protection margin is introduced to avoid frequent jittering of metering boxes between different zones caused by small phase difference changes due to normal load fluctuations, thereby improving the stability of topology identification. The size of the protection margin can be determined based on the maximum fluctuation range of phase difference under steady-state conditions in historical data. For example, the daily maximum fluctuation range of phase difference for all metering boxes under normal operating conditions can be statistically analyzed, and the 95th percentile can be taken as the protection margin. Assume the initial value of the boundary of a certain zone obtained through clustering is... The protection margin was obtained through statistical analysis. Then the final upper boundary can be set as The lower boundary is set to In this way, as long as the phase difference of the metering box falls within this expanded interval, it is considered to still belong to that zone, effectively preventing misjudgments caused by minor fluctuations.

[0034] Method 2: Inflection point detection-based method: Arrange all metering boxes in ascending order of phase difference to form a sequence. Then calculate the difference in this sequence (subtracting the previous value from the next) to obtain the phase difference jump value between adjacent metering boxes. Under normal circumstances, the phase difference changes gradually within the same branch, with small jump values; however, between different branches, due to differences in line parameters and loads, there may be a relatively large jump value. By setting a jump threshold (e.g., 3 times the mean of all jump values, or the upper edge determined by a statistical box-plot method), the first jump point exceeding this threshold can be used as the boundary between the first and second partitions. This process is repeated to divide all partitions.

[0035] The rationale for using an adaptive method to determine the threshold is that it fully considers the differences in load characteristics and line parameters of different transformer areas and at different times, making the topology identification results more consistent with the actual situation on site and having stronger universality and robustness.

[0036] For metering boxes falling within the same phase difference zone, their electrical distances on the feeder branch are determined based on their voltage amplitude, from largest to smallest. On the same feeder branch, due to line impedance, the voltage gradually decreases along the power flow direction. Therefore, metering boxes with larger voltage amplitudes are located closer to the beginning of the branch (i.e., closer to the transformer side), while those with smaller voltage amplitudes are located closer to the end of the branch. Thus, for metering boxes within the same zone, sorting them by their voltage amplitude (which can be the average of the most recent cycles) from largest to smallest reflects their upstream and downstream relationship on the feeder branch.

[0037] Using the transformer outgoing line side as the root node, each feeder branch as a first-level branch, and each metering box as a leaf node, a real-time topology association matrix describing the connection relationship between the transformer and each metering box is constructed. The constructed real-time topology association matrix can be represented by various data structures, such as adjacency lists or tree structures. Preferably, this matrix is ​​a single... The matrix is ​​a set of subarrays, where M represents the number of metering boxes, numbered 1...M, and the transformer is numbered 0. Matrix elements... This indicates the connection relationship between nodes p and q. For example, if metering box m is a direct downstream node of metering box i (i.e., i is the parent node of m), then set... Meanwhile, for rapid retrieval, a list can be maintained to record the parent node (i.e., the nearest upstream metering box or transformer) and the list of child nodes (i.e., the metering boxes directly connected downstream) of each metering box. This dynamically constructed matrix provides accurate path information for tracing the impact of upstream and downstream nodes in subsequent steps.

[0038] Abrupt changes are detected in the time series of phase difference or voltage amplitude of the metering box. When an abrupt change is detected, it is determined that the phase difference or voltage amplitude of the metering box has changed abruptly.

[0039] The purpose of abrupt change detection is to sense changes in distribution network topology or line parameters in real time, such as planned load rerouting, line switching after a fault, or increased impedance due to line aging. Abrupt change detection can be implemented using various algorithms; this embodiment provides two specific methods: Method 1: Sliding Window Method Define a sliding window of length L1 and continuously calculate the mean of the phase difference (or voltage amplitude) data within the window. and standard deviation When the data of a new power frequency cycle When they arrive, if If the coefficient is zero, then a mutation has occurred. This is a sensitivity parameter that can be set according to actual needs. For example, if a higher detection sensitivity is required, it can be set to... To reduce false alarms, you can set... The choice of L1 needs to balance response speed and stability; for example, L1 can be set to 50 power frequency cycles.

[0040] Method 2: Cumulative Sum (CUSUM) Control Chart Method The CUSUM method is highly sensitive to the cumulative effect of small shifts and is suitable for detecting persistent shifts. The cumulative sum of positive shifts is defined. The cumulative sum of negative offsets Initial value For the data of the kth power frequency cycle The updated formula is as follows: in, These represent the cumulative positive and negative offsets of the k-th power frequency cycle, respectively. These represent the cumulative positive and negative offsets during the (k-1)th power frequency cycle, respectively. This represents the data for the k-th power frequency cycle. This is the target mean during normal operation, C is the allowable drift (usually half the standard deviation), and the threshold H is used to determine whether a sudden change has occurred. or When a mutation occurs, it is determined that a mutation has taken place. Parameters C and H can be optimized using Monte Carlo simulations based on historical data.

[0041] The rationale for detecting sudden changes lies in the fact that connectivity or line parameters in the physical world do not change frequently; once a sudden change occurs, it inevitably corresponds to some kind of event. By monitoring such sudden changes in real time, it is possible to ensure that the topology correlation matrix remains consistent with the actual operating mode, which is a prerequisite for achieving accurate anomaly analysis.

[0042] In this embodiment, the update logic of the real-time topological association matrix is ​​as follows: When a sudden change in the phase difference of the metering box is detected, the timestamp of the sudden change and the phase difference value before and after the sudden change are recorded.

[0043] Reacquire the phase difference data of the metering box for several consecutive power frequency cycles after the sudden change. If no sudden change occurs during these consecutive power frequency cycles, recalculate the phase difference of the metering box. The number of consecutive power frequency cycles can be set to a fixed number, such as 50 power frequency cycles (corresponding to 1 second), to ensure that the data enters a new steady state. At the same time, it is necessary to verify that no further sudden change is detected within these multiple cycles to ensure that the calculated average value is under the new steady state.

[0044] The recalculated phase difference is compared with the phase difference partition boundary value before the abrupt change: if the recalculated phase difference falls into another phase difference partition, it is determined that the physical connection relationship of the metering box has changed, and it is moved from the original feeder branch to the new feeder branch, and the parent node information of the metering box in the real-time topology association matrix is ​​updated; if the recalculated phase difference is still in the original phase difference partition, but the voltage amplitude change exceeds the preset threshold, it is determined that the line parameters of the feeder branch where the metering box is located have changed, and the relative position order of the metering box on the feeder branch in the real-time topology association matrix is ​​updated, but its parent node remains unchanged.

[0045] The key here is to distinguish between changes in physical connections and changes in line parameters, specifically: Changes in physical connections: A phase difference jump from one zone to another signifies a fundamental change in the electrical path of the metering box, for example, switching from feeder A to feeder B. In this case, the topology changes, requiring updates to its parent node and corresponding adjustments to the upstream and downstream order of all relevant metering boxes on both the old and new feeder branches.

[0046] Line parameter changes: A sudden change in phase difference, still within a zone, but with a significant drop (or rise) in voltage amplitude, may indicate a fault in the line between the meter box and the upstream node (e.g., poor contact leading to increased impedance) or that the line has been replaced (e.g., with thinner wires). In this case, the electrical distance of the meter box on the branch changes (it's farther from the power source), so its relative position order on the branch may need to be adjusted. For example, a meter box that was originally the second highest voltage may become lower than the following meter boxes after a fault, thus requiring a reordering. However, the overall connection relationship (parent node) of the branch remains unchanged.

[0047] The preset threshold for judging voltage amplitude changes can be a relative rate of change, for example, a voltage amplitude drop exceeding 5%. This threshold should be set in conjunction with the allowable voltage fluctuation range and measurement accuracy of the distribution area; for example, it can be set to 3%-5% of the rated voltage. This range falls in the middle to high end of the normal fluctuation range (±7%), meaning that this embodiment focuses on changes that significantly deviate from normal operating conditions but have not yet reached the level of a serious fault. For example, the voltage of a metering box drops from 98% to 94% of its rated value, a decrease of 4 percentage points. This is still within the national standard's allowable range of 94%-106%, but a 4% drop is enough to raise concern, potentially indicating increased line contact resistance or insulation aging. Through this refined processing, topology updates can not only sense changes in connection relationships but also changes in line health status, providing richer information for operation and maintenance.

[0048] In this embodiment, the fundamental and harmonic energy distribution characteristics are used as candidate objects for harmonic anomalies, specifically: A fast Fourier transform is performed on the sequence of voltage and current sampling data for each power frequency cycle to decompose the voltage and current amplitudes, including the fundamental frequency, third harmonic component, fifth harmonic component, and seventh harmonic component.

[0049] Here, the detailed calculation process of the Fast Fourier Transform (FFT) is similar to that in step 1, but it requires extracting information from multiple frequency points. For a sampling frequency of... The system, after FFT, yields a spectral resolution of [missing value]. By locating the base frequency The amplitude of each harmonic can be accurately extracted by using the spectral lines of 50Hz and its integer multiples (150Hz, 250Hz, 350Hz...). In this embodiment, the third, fifth, and seventh harmonics are used as examples, but in practical applications, it can be extended to higher harmonics (such as the 9th, 11th, etc.) as needed to capture richer harmonic features.

[0050] The active power of the fundamental wave and each harmonic component is calculated separately, and the percentage of the active power of each harmonic component to the active power of the fundamental wave is calculated to form a harmonic content rate sequence.

[0051] For the h-th harmonic, its active power The calculation formula is: in, and These are the voltage and current amplitudes of the h-th harmonic, respectively. and These are their initial phase angles.

[0052] Harmonic content It is a commonly used indicator for measuring harmonic levels, defined as the active power of the h-th harmonic and the active power of the fundamental frequency. percentage For each power frequency cycle, we can calculate , , These values, and their changes over time, constitute a harmonic content sequence.

[0053] Based on the feeder branch to which each metering box belongs, determined by the real-time topological correlation matrix, the harmonic content sequence of each metering box on the same feeder branch is compared horizontally, and metering boxes that deviate from the average level of the branch are extracted as candidates for harmonic anomalies.

[0054] The specific method for horizontal comparison is as follows: For a specific feeder branch, assume it has F metering boxes. For each characteristic statistical window, calculate the harmonic content of a certain harmonic in all metering boxes on that branch (e.g., (Average) and standard deviation Then, for the i-th metering box on this branch, the deviation of its harmonic content rate is calculated. : in, It is a very small positive number used to prevent division by zero errors. When the i-th metering box... The value exceeds the preset threshold. At that time, the metering box is marked as a candidate for harmonic anomaly.

[0055] threshold The determination can be achieved through statistical methods. For example, assuming that the harmonic content follows a normal distribution, then a setting can be made. For a 99.7% confidence interval, meter boxes deviating from the average level by three standard deviations are considered anomalies. Alternatively, box plot analysis of historical data can be used to identify meter boxes exceeding the upper edge (Q3 + 1.5 * IQR) as anomaly candidates; Q3 represents the third quartile, and IQR represents the interquartile range. The rationale for this horizontal comparison method is that users on the same feeder branch have similar electrical equipment and consumption behaviors, therefore their harmonic background levels should also be similar. If the harmonic level of a meter box deviates significantly from its neighbors, it is highly likely that there is abnormal equipment generating harmonics inside (such as inferior electrical appliances) or that the metering device itself is inaccurate in measuring harmonics.

[0056] The dynamic evolution of the three-phase imbalance is characterized by an eigenvector, specifically: Obtain the voltage and current amplitudes of phases A, B, and C; calculate the negative sequence current imbalance and zero sequence current imbalance for each power frequency cycle using the symmetrical component method.

[0057] The symmetrical component method decomposes a three-phase asymmetrical phasor system into three sets of symmetrical sequence component systems: positive sequence, negative sequence, and zero sequence. For current, taking phase A as the reference, its negative sequence current... and zero-sequence current The calculation formula is as follows: Wherein, rotation factor , These are the phasors (including amplitude and phase) of the current in each phase. It also represents the imaginary unit, but its application is to the power frequency phasor, which represents the actual physical quantities of the power system. Then, the negative sequence current imbalance... and zero-sequence current imbalance It is usually defined as the corresponding sequence component and the positive sequence current. The percentage, that is , .

[0058] Within each feature statistics window, record the change trajectory of negative sequence current imbalance and zero sequence current imbalance, including the rate of change, fluctuation amplitude, and duration.

[0059] Feature points in the trajectory of change are extracted, including the start time, peak time, and recovery time of the imbalance increase, to form a feature vector describing the dynamic evolution of the imbalance.

[0060] Treat the imbalance as a time series signal We can extract its dynamic evolution features. Assuming a feature statistics window... Within this context, we detected an imbalance event. First, we need to determine the starting point of this event. And the endpoint. This can be achieved by calculating the first derivative of the time series and setting a rate of change threshold. When the derivative continuously exceeds the rate of change threshold, the event is considered to have started; when the derivative falls back below the rate of change threshold and remains below it for a period of time, the event is considered to have ended.

[0061] The rate of change can be taken as the average rate of change of the imbalance in the initial stage (e.g., the first 5 cycles from the starting point), i.e. , It indicates the length of the power frequency cycle.

[0062] Fluctuation amplitude: The difference between the peak value and the value at the start of the event (or the average value before the event).

[0063] Duration: The total time from the starting point to the recovery point.

[0064] Feature points: the time values ​​or offsets relative to the window start point for the start time, peak time, and recovery time.

[0065] Ultimately, these quantities are combined into a multidimensional feature vector, such as feature vector = [rate of change, fluctuation amplitude, continuous field of view, starting point, peak time, recovery time]. This dynamic feature contains more information than a simple average imbalance, and can distinguish between instantaneous impulsive imbalance and continuous steady-state imbalance, providing a key basis for diagnosing abnormal types (such as sudden start-up and shutdown of high-power single-phase equipment vs. permanent imbalance of three-phase loads).

[0066] The synchronicity of load fluctuations is characterized by two quantitative indicators, specifically: The total active power time series at the incoming side of the metering box and the active power time series of each sub-meter within the metering box are obtained, and the two time series are normalized. A sliding time window is used to calculate the Pearson correlation coefficient between the total power series and the power series of each sub-meter within each time window. Within each characteristic statistical window, the mean correlation coefficient between the power series of each sub-meter falling within it and the total power series is calculated as the first quantitative indicator to measure the synchronicity of load fluctuations within the characteristic statistical window. The variance of the correlation coefficient between the power series of each sub-meter falling within the characteristic statistical window and the total power series is calculated as the second quantitative indicator.

[0067] Normalization is used to eliminate the influence of dimensions and amplitude, allowing correlation analysis to focus on the similarity of fluctuation patterns. The preferred normalization method is Z-score normalization. The Pearson correlation coefficient is used to measure the degree of linear correlation between two variables. When calculating the Pearson correlation coefficient, the width of the sliding window should be sufficient to capture typical load fluctuation patterns; for example, it can be set to 15 minutes, corresponding to power data for 900 power frequency cycles.

[0068] For each submeter, the mean and variance of all correlation coefficients falling within the statistical window of this feature are calculated. The mean reflects the average degree of synchronization between the submeter's electricity consumption behavior and the main meter (the entire metering box) during that period; a mean close to 1 indicates that the submeter's load changes are highly synchronized with the total load, making it the main load of the metering box; a mean close to 0 indicates that the submeter's load changes are not significantly related to the total load. The variance reflects the stability of the synchronization relationship between the submeter and the main meter; a small variance indicates that the submeter's load and the total load always maintain a consistent relationship (either always synchronized or always out of sync); a large variance indicates that the submeter's synchronization mode has changed, for example, a device may be on at times (increased load, stronger synchronization) and off at other times (decreased load, weaker synchronization). These two indicators together constitute the load fluctuation synchronization characteristic, which is of great significance for detecting post-meter theft (new loads bypassing submeters, causing the total power to lose synchronization with the sum of the submeters) or submeter malfunctions (submeter metering distortion, causing its power curve to be abnormally shaped compared to the main meter curve).

[0069] In this embodiment, the energy conservation deviation vector is obtained as follows: For each characteristic statistical window, the cumulative active power value on the metering box's incoming side within that window is obtained as the total incoming energy. Specifically: let the length of the w-th characteristic statistical window be... The sequence of instantaneous total active power values ​​collected by the intelligent measurement unit on the incoming line side within this window is as follows: Then the total energy of the incoming line in the feature statistics window for: Where g is the sampling time index, and G is the number of sampling points within the feature statistics window. The sampling interval is... Let t be the time of the g-th sampling moment, and t be the time index in the feature statistics window.

[0070] The summation of the cumulative active power values ​​of all energy meters in the metering box within this characteristic statistical window yields the sum of the energy measured by each meter. Specifically: The design box contains e electricity meters, and their cumulative electricity consumption values ​​within the same window are respectively... (These values ​​can be obtained directly by reading the frozen electricity of each sub-meter or by integrating the power of the sub-meter), so the sum of the energy measured by the sub-meters is the sum of the cumulative active power of all the electricity meters.

[0071] Fixed line loss energy is a correction term in the energy conservation equation; line loss power It is proportional to the square of the current, that is... Where R is the equivalent resistance of the wiring inside the metering box (from the inlet to the outlet of each sub-meter). Since the wiring inside the box is usually short and fixed, its resistance value R can be considered a constant. This constant can be calculated from the wire diameter and length on the design drawings or measured by a microresistance meter under power failure conditions. Therefore, the fixed line loss energy within a characteristic statistical window is: in, This is the instantaneous value of the incoming line current.

[0072] The energy conservation deviation value is in, It is not an independent physical quantity, but rather a difference operator or deviation symbol used to represent the final calculated difference.

[0073] Energy conservation deviation vector: Arrange the deviation values ​​of W consecutive windows in time order to form a W-dimensional vector. The energy conservation deviation vector mentioned later in this embodiment refers to the vector composed of the energy conservation deviation value after deducting the fixed line loss energy; this vector contains all the information about the evolution of the deviation over time, and the positive or negative sign of the deviation value has a clear physical meaning: The incoming energy is more than the sum of the sub-meter readings, meaning there is excess energy. Possible causes include: under-metering (e.g., sub-meter malfunction, electricity theft), over-metering by the main meter, or the presence of unmetered loads (e.g., incorrect wiring).

[0074] The incoming energy is less than the sum of the sub-meter readings, meaning there is less energy. Possible reasons include: under-metering by the main meter, over-metering by the sub-meters, or reverse power transmission (such as photovoltaic power generation not being correctly measured).

[0075] This embodiment establishes a simulated low-voltage distribution transformer test platform in a laboratory environment. The platform includes a 10kVA distribution transformer, three feeder branches, and 12 smart metering boxes, each containing five energy meters with communication capabilities. The test platform is equipped with a smart measurement unit featuring synchronous phasor measurement functionality, with a sampling frequency set to 5120Hz (i.e., 256 samples per power frequency cycle), achieving microsecond-level time synchronization accuracy.

[0076] During data acquisition, using the zero-crossing point of the transformer's output voltage as a reference, the continuous time axis was divided into 15-minute characteristic statistical windows, each containing 45,000 power frequency cycles. The following test scenarios were designed for the three electrical characteristics: Harmonic characteristic test: A 6-pulse rectifier was connected to the incoming line side of metering box C. By controlling its start / stop and load size, an abnormal harmonic source was simulated. At the same time, normal linear loads were maintained at metering boxes A and B, and continuous operating data was collected for 150 minutes.

[0077] Three-phase imbalance test: A high-power single-phase resistive load (5kW) is switched on and off via a controllable switch on the metering box input side to simulate a three-phase imbalance event caused by a sudden load change. The switching time is set at the 8th minute, and the test continues until the 42nd minute when the load is removed, recording a complete dynamic evolution trajectory including the rise, peak, fall, and recovery process.

[0078] Load Synchronization Test: Submeters 1-3 and 5 in the metering box are maintained under normal fluctuating loads, with their power curves highly synchronized with the total load. Submeter 4 is in a normal state in windows 1-11 and 21-30 (the load pattern has inherent differences from the total load). In window 12-20, its voltage sampling circuit is manually disconnected to simulate a submeter metering fault. The total power and the power data of each submeter are collected synchronously, and the Pearson correlation coefficient is calculated.

[0079] The experiment collected data from 30 consecutive feature statistical windows, and the specific data are shown in the table below: Table 1: Comparison of Harmonic Content Rate Data Table 2: Schematic diagram of dynamic evolution trajectory data of three-phase imbalance Table 3: Schematic diagram of load fluctuation synchronicity analysis data Table 1 and Figure 2The graph shows the variation trend of the third harmonic content (HRU3) of three metering boxes on the same feeder branch over 150 minutes. As can be seen from the graph, the harmonic content of metering boxes A and B remained stable between 3.1% and 3.8%, with small fluctuations, which is within the normal nonlinear load background harmonic level. However, the harmonic content of metering box C continuously increased from the 15th minute, reaching a peak of 8.5% at the 60th minute, and then gradually decreased. Throughout the abnormal period (15-70 minutes), it significantly deviated from the branch average level (with a maximum deviation of 63%). This phenomenon reflects the presence of harmonic source equipment (such as a 6-pulse rectifier) ​​inside metering box C, causing a significant change in its harmonic energy distribution characteristics. Therefore, by comparing the harmonic content of each metering box on the same branch, metering boxes deviating from the branch average level can be effectively identified as candidates for harmonic anomalies.

[0080] Table 2 and Figure 3 The complete evolutionary trajectories of negative-order imbalance and zero-order imbalance over 58 minutes are shown. From Figure 3 The data clearly identifies four stages of the imbalance event: the first 8 minutes represent a steady-state period, with the imbalance remaining at a low level (negative order <1.5%, zero order <0.7%); minutes 8-18 represent the rising phase, where the imbalance rapidly increases; minutes 18-26 represent the peak phase, with the negative order imbalance reaching its highest point of 9.3% at 20 minutes, and the zero order imbalance simultaneously reaching 6.7%; minutes 26-38 represent the declining phase, where the imbalance gradually decreases; and after 38 minutes, the recovery phase begins, finally returning to pre-event levels at 42 minutes. This complete dynamic evolution trajectory is a concrete manifestation of the change trajectory and characteristic points. Figure 3 The starting point (8 min), peak point (20 min), and recovery point (42 min) constitute a feature vector describing the dynamic evolution of the imbalance. This vector can effectively distinguish between instantaneous impact imbalance (such as the start-up and shutdown of high-power single-phase equipment) and continuous steady-state imbalance (such as permanent imbalance of three-phase load), providing a key basis for the diagnosis of abnormal types.

[0081] Table 3 Figure 4 and Figure 5 Two quantitative indicators are presented to reflect the overall synchronization level and stability of synchronization relationships of load fluctuations within each time window. Figure 4 The correlation coefficients of sub-tables 1-3 and 5 within each time window are relatively close, indicating that the load fluctuations of these sub-tables maintain good synchronization with the main table; and... Figure 5 In the data, the line graphs of these sub-tables fluctuated steadily between 0.85 and 0.96. However, sub-table 4, during windows 12-20, showed a significantly lower correlation coefficient (0.06-0.15) than the mean correlation coefficient (0.73-0.76), creating a clear visual contrast and intuitively revealing an anomaly in this sub-table during this period. Figure 5 In this study, the line segment of sub-table 4 maintained a normal difference level of 0.32-0.42 in windows 1-11 and 21-30, while it dropped sharply to 0.06-0.15 in windows 12-20. Simultaneously, the variance line segment remained at a low level below 0.05 in windows 1-11 and 21-30, while it rose sharply to 0.10-0.13 in windows 12-20, peaking in windows 15-17. This change in the variance line segment quantitatively characterizes the degree of overall synchronicity instability caused by the anomalies in sub-table 4.

[0082] Step 3: Perform frequency domain analysis on the energy conservation deviation vector to obtain frequency domain energy characteristics, and match the frequency domain energy characteristics with a pre-built library of anomalous physical patterns to determine the anomaly type; In this embodiment, the energy conservation deviation vector is converted into a frequency domain energy spectrum distribution map, as follows: Wavelet packet decomposition is performed on the energy conservation deviation vector to decompose the time-domain signal into different frequency bands. The wavelet basis function and the number of decomposition levels are selected so that each frequency band after decomposition at least covers the frequency range corresponding to a specific frequency band.

[0083] Wavelet packet decomposition is a more refined time-frequency analysis method than the traditional Fourier transform. It can not only give the frequency components of a signal, just like the Fourier transform, but also the time when these frequency components appear. It is particularly effective for analyzing non-stationary signals such as energy conservation deviations.

[0084] The specific implementation steps are as follows: The energy conservation deviation vector obtained in step 3 is considered as a discrete-time signal; the choice of wavelet basis function affects the decomposition effect. This embodiment provides two commonly used wavelet basis choices: The Daubechies series (such as db4, db6) has good regularity and tight support, making it suitable for detecting abrupt changes and transient characteristics of signals, such as transient pulses caused by external strong magnetic field interference or loose terminals.

[0085] The Symlets series (such as Sym8) is an improvement on the Daubechies series, offering better symmetry and reducing phase distortion during signal decomposition and reconstruction. It is suitable for analyzing periodic stationary signals such as harmonics. The number of decomposition levels J determines the frequency resolution. If the sampling rate of the original signal (i.e., the frequency of the characteristic statistical window) is... (For example, one point every 15 minutes, then) After J-level decomposition, Hz can be obtained. Each frequency band has an equal width of [number] bands. When selecting J, it is necessary to ensure that the narrowest frequency band after decomposition can cover the narrowest specific frequency band of interest. For example, if we want to analyze signal components with a period between 1 hour and 1 day (corresponding to a frequency range of approximately...), Hz to (Hz), so we need to choose a suitable J so that a certain sub-band falls exactly within this range. For example, if we choose J=4 levels for decomposition, then we get 16 frequency bands.

[0086] Perform wavelet packet decomposition: Perform J-level wavelet packet decomposition on the discrete-time signal to obtain... A sequence of node coefficients, where each node coefficient represents the projection of the original signal into a specific frequency band and a specific time window.

[0087] Calculate the energy values ​​of wavelet coefficients in each frequency band after decomposition. Construct a frequency domain energy spectrum distribution map showing the energy variation with frequency, with frequency as the abscissa and energy value as the ordinate. Store the map as a matrix, where the rows of the matrix correspond to the frequency bands, the columns correspond to the feature statistics window, and the matrix elements are the energy values ​​of the frequency bands within the feature statistics window.

[0088] After wavelet packet decomposition, we obtain the coefficient sequence of the k-th frequency band under the j-th level decomposition. For each feature statistical window (corresponding to a sampling point of the original signal), we can align it with the decomposition coefficients in time. However, more commonly, we calculate the total energy of each frequency band over a longer time scale (e.g., a day).

[0089] For a fixed frequency band, its energy over the entire analysis period is defined as the sum of the squares of all wavelet packet coefficients in that frequency band.

[0090] To construct the frequency domain energy spectrum distribution map, we plot a bar chart or line graph with the frequency band (i.e., frequency range) as the x-axis and the calculated sum of squares as the y-axis. This graph visually displays the frequency component distribution of the energy conservation deviation signal.

[0091] To support more refined time-varying analysis, a sliding window approach can be used to construct an energy spectrum matrix. A time window of length L² feature statistics is set, with a step size of one window. Each time the window slides, the wavelet packet coefficient energy of each frequency band within the window is calculated, resulting in an energy spectrum row vector. Stacking all the row vectors of the sliding windows in chronological order forms a two-dimensional matrix. The rows of the matrix correspond to the time window, the columns correspond to the frequency band intervals, and the matrix elements are the energy values ​​of the corresponding frequency band within a given time window. This matrix contains both frequency information and information about the changes of these frequency components over time.

[0092] Based on the physical mechanism of abnormal electricity metering, the frequency range is divided into at least four preset specific frequency bands; the sum of the energy values ​​of all frequency intervals within each specific frequency band is calculated as the frequency domain energy characteristic of that specific frequency band.

[0093] The entire frequency range (from 0 to the Nyquist frequency) The frequency bands are divided into several specific frequency bands with physical meaning. This division is not arbitrary, but based on physical simulations and understanding of different time scales of anomalous events. In this embodiment, the following four specific frequency bands are divided: Frequency band A0 (ultra-low frequency band, for example) Hz: corresponds to changes with a period longer than one day, such as slow deviations caused by seasonal load changes, meter reading cycle errors, temperature drift, etc.

[0094] Frequency band B0 (low frequency band, for example) Hz: corresponds to the variation of the cycle from a few hours to a day, such as the metering deviation pattern caused by typical daily load curve fluctuations and time-of-use pricing responses.

[0095] Frequency band C0 (intermediate frequency band, for example) Hz): This corresponds to a period of variation between minutes and hours. This is the characteristic frequency band for many abnormal events. For example, loose input terminals, due to the heat-cooling cycle, cause contact resistance changes with a period typically ranging from minutes to hours; metering error drift caused by chip aging also falls into this frequency band.

[0096] Frequency band D0 (high frequency band, for example) Hz): This corresponds to rapid changes in frequency between a few seconds and a few minutes. For example, strong external magnetic field interference usually manifests as short, high-energy pulses with a wide frequency band and concentrated energy in the high-frequency range; load impacts such as the start-up and shutdown of large motors also leave traces in this frequency band.

[0097] For a given energy spectral matrix, for the a-th time window, we calculate its frequency domain energy characteristics in frequency band A. That is, all matrix elements belonging to column b (frequency band) of frequency band A0. The sum. Similarly, we can obtain... , , Therefore, for the a-th time window, we obtain a four-dimensional feature vector. This vector forms the basis for subsequent pattern matching.

[0098] The pre-built abnormal physical model library includes four abnormal types, each corresponding to a simulation of a metering box abnormal type; the energy conservation deviation vector under each abnormal type is collected and converted into a frequency domain energy spectrum distribution map to extract the frequency domain energy characteristics of each specific frequency band.

[0099] The abnormal physical pattern library is the core knowledge base for intelligent diagnosis using this method. Its construction is a supervised learning process that requires obtaining the standard fingerprint of the energy conservation deviation signal under known anomalies.

[0100] The specific simulation methods for the four anomaly types are as follows: The first type of anomaly: Accelerated aging test of computing chips. In the laboratory, the same model of metering chip is placed in a high-temperature environment (such as 70°C) to accelerate its aging process. At the same time, the measured power and the actual input standard source power are continuously recorded. The difference sequence between the two is the energy conservation deviation vector under this mode. This deviation vector reflects the evolution law of the metering error caused by chip aging.

[0101] The second type of anomaly: simulation of loose incoming terminals. In the experimental environment, the screws of the incoming terminals of the metering box are manually loosened to different torque levels (such as 30% and 50% of the rated torque). Then, a large current is passed through (simulating a high-load condition). A high-precision data acquisition card is used to record the instantaneous voltage drop and power loss caused by the change in contact resistance, and an energy conservation deviation vector is constructed. This vector will exhibit periodic fluctuation characteristics related to the square of the load current.

[0102] The third type of anomaly: simulation of external strong magnetic field interference. A strong magnet (such as a neodymium iron boron permanent magnet) is placed close to a running electricity meter, applying interference at different angles and distances to simulate real-world electricity theft. The amount of electricity lost by the interfered meter is recorded, constructing an energy conservation deviation vector, which is typically represented by a very short-duration negative pulse (representing electricity loss).

[0103] The fourth type of anomaly: simulation of three-phase unbalanced load. A strong three-phase unbalanced condition is created by repeatedly switching high-power single-phase loads on the outgoing side of the metering box using a controllable switch. The energy conservation deviation caused by increased line losses or decreased metering performance of certain energy meters under this special condition is recorded, and a deviation vector is constructed.

[0104] For each anomaly type, multiple repeated experiments are conducted to obtain multiple samples of energy conservation deviation vectors. Then, for each sample, the frequency domain energy spectrum analysis described earlier in this step is performed to obtain its frequency domain energy characteristics in four specific frequency bands. Finally, the average (or median) of all samples is taken as the standard feature vector for that anomaly type.

[0105] The frequency domain energy features extracted under each anomaly type are used as standard features to establish a correspondence between anomaly physical patterns and frequency band frequency domain energy features, forming an anomaly physical pattern library. The final anomaly physical pattern library can be in the form of a simple lookup table, with one column representing the anomaly type and the other column representing the corresponding standard feature vector. As field operational data accumulates, this library can be updated using new, confirmed anomaly data to make it more accurate and complete.

[0106] The specific matching process is as follows: The frequency domain energy characteristics of each metering box in each specific frequency band are obtained to form the feature vector to be measured; the standard feature vector corresponding to each anomaly type is retrieved from the abnormal physical pattern library; the standard feature vector is composed of the standard features of each specific frequency band; the similarity between the feature vector to be measured and each standard feature vector in the abnormal physical pattern library is calculated, and the anomaly type that is successfully matched is determined based on the similarity.

[0107] The matching process is essentially a pattern recognition problem. There are various methods for calculating similarity, and this embodiment provides two preferred solutions: Option 1: Based on cosine similarity For the feature vector to be tested and a standard feature vector in the library Their cosine similarity Defined as: in, , represents the index of the frequency band; the range of cosine similarity is [-1, 1]. The closer the value is to 1, the more consistent the directions of the two vectors are, that is, the more similar their feature distribution patterns are.

[0108] Option 2: Based on Euclidean distance Euclidean distance measures the absolute distance between vectors in a multidimensional space. It involves calculating the Euclidean distance between the eigenvector to be measured and the standard eigenvector. : The smaller the Euclidean distance, the closer the two vectors are.

[0109] Calculate the similarity (or distance) between the vector to be tested and all four pattern vectors to obtain a similarity list. If the maximum similarity exceeds a preset matching threshold (e.g., 0.85), or the minimum Euclidean distance is less than a preset distance threshold, the match is considered successful, and the corresponding pattern is identified as the diagnosed anomaly type.

[0110] If all similarities are below the threshold, or all distances are above the threshold, then the current anomaly is considered not to belong to any category in the pattern library. It may be a new anomaly type and needs to be marked as an unknown pattern for further analysis.

[0111] The matching threshold or distance threshold can be determined using cross-validation. A portion of known samples from the pattern library is reserved as a test set. The similarity between these test samples and all pattern vectors in the library is calculated. By adjusting the threshold, a balance is struck between precision and recall. This balance point becomes the final matching threshold. For example, we can set a threshold that maximizes the F1 score (the harmonic mean of precision and recall) on the test set.

[0112] After matching with the abnormal physical pattern library, a time-domain backtracking verification step is also included: When the frequency domain energy characteristics of the meter under test successfully match a certain anomaly type, the anomaly type and the corresponding characteristic frequency band are recorded.

[0113] Based on the frequency range of the characteristic frequency band, the corresponding time scale is determined in reverse; the original energy conservation deviation vector time-domain waveform is traced back to, and time-domain waveform segments of the successfully matched characteristic statistical window and its adjacent characteristic statistical windows are extracted; the morphological characteristics of the time-domain waveform segments are analyzed, including pulse amplitude, pulse width, rate of change, and periodicity, which are consistent with the typical time-domain performance of this anomaly type; if the morphological characteristics of the time-domain waveform segments are consistent with the typical time-domain performance of this anomaly type, the matching result is confirmed to be valid; if they are inconsistent, the confidence level of the matching result is lowered and marked as requiring on-site verification.

[0114] For example, if the matched anomaly is a loose terminal (its characteristic frequency band is mid-frequency band C), the period range corresponding to this frequency band is from a few minutes to a few hours. Then, we go back to the time domain and focus on observing the waveform on this time scale.

[0115] Analysis of temporal morphology: The typical time-domain manifestation of load current loosening is that the deviation value is proportional to the square of the load current. On the daily load curve, this manifests as a periodic pattern where the deviation increases significantly during the evening peak period and is almost zero during the off-peak period.

[0116] The typical time-domain manifestation of strong magnetic interference is that the deviation value shows a sharp negative pulse at a certain moment, the pulse duration is very short (a few seconds to a few minutes), and then it recovers quickly.

[0117] The typical time-domain manifestation of chip aging is that the deviation value exhibits a slow, almost linear long-term drift trend.

[0118] If the backtracked time-domain waveform is highly consistent with the linear performance described above, the matching result is confirmed to be valid with high confidence. If the frequency domain matches the loose terminal, but the backtracking time domain reveals that the deviation value consists of a series of irregular pulses unrelated to the load, the frequency domain matching may be a coincidence. In this case, the confidence level should be lowered, and a prompt should be output indicating suspected loose terminal, but the time-domain characteristics do not match, suggesting on-site verification.

[0119] Step 4: After matching and determining the anomaly type, determine the upstream and downstream nodes of the target metering box based on the real-time topological correlation matrix, eliminate the influence of upstream and downstream nodes on the energy conservation deviation, separate the anomaly contribution of the target metering box, and output the anomaly analysis results.

[0120] In this embodiment, the location of the currently matched metering box is located in the real-time topology association matrix. Based on the real-time topology association matrix, the upstream nodes of the metering box are determined, including the transformer outgoing line side and other metering boxes located on the feeder branch where the metering box is located and located before the metering box. The child nodes of the metering box, that is, the various energy meters connected inside the metering box, are determined as the downstream nodes of the metering box.

[0121] The following information can be obtained by dynamically constructing and continuously updating the real-time topological association matrix: Upstream nodes: All nodes along the path from the transformer to the target metering box, including the transformer and other metering boxes along the way. Anomalies at these nodes can propagate like contamination along the current flow, affecting observations of the target metering box.

[0122] Downstream nodes: the various energy meters connected inside the target metering box. The metering accuracy of these energy meters directly affects the calculation results of the energy conservation deviation vector.

[0123] Eliminating the impact of upstream nodes specifically includes: Following the path order from the root node to the metering box, the frequency domain energy features of each upstream node within the same feature statistics window are extracted sequentially. These frequency domain energy features are the frequency domain energy features of specific frequency bands corresponding to the same anomaly type as the target metering box. Simultaneously, the frequency domain energy features of the target metering box within the same feature statistics window are also extracted. The logic here is: if both the upstream nodes and the target metering box exhibit the same anomaly type, and their frequency domain energy features are correlated, then a portion of the anomaly signal observed by the target metering box may originate from the upstream nodes.

[0124] If an upstream node also has the same frequency domain energy characteristics, the ratio of the frequency domain energy characteristics of the upstream node to the frequency domain energy characteristics of the metering box is used as the contribution coefficient of the upstream node to the anomaly of the metering box; the product of the contribution coefficient and the frequency domain energy characteristics of the metering box is the contribution of the upstream node to the metering box.

[0125] In a more refined model, abnormal signals attenuate as they propagate along the line. Therefore, the contribution coefficient calculation can also consider the effect of electrical distance, introducing a distance attenuation factor. The electrical distance can be quantified based on the line impedance or the phase difference calculated in step 2. The distance attenuation factor can be preset through analysis of historical data; for example, the greater the distance, the smaller the attenuation factor.

[0126] The contributions of all upstream nodes are summed to obtain the total contribution of upstream nodes to the anomaly of the metering box; the total contribution is subtracted from the frequency domain energy characteristics of the metering box to obtain the net characteristic value after excluding the influence of upstream nodes.

[0127] Eliminating the impact of downstream nodes specifically includes: Obtain the electrical characteristics of each energy meter downstream of the metering box within the same feature statistics window, compare the correlation between each energy meter's electrical characteristics and the energy conservation deviation vector, and mark the energy meter as a suspected abnormal sub-meter when the correlation coefficient of at least one electrical characteristic exceeds the corresponding correlation threshold.

[0128] Anomalies at downstream nodes are typically reflected directly in their contribution to the energy conservation deviation. By calculating the correlation coefficient between the energy conservation deviation value and the electrical characteristics of each energy meter (such as harmonic content, three-phase imbalance, load fluctuation synchronicity, etc.), it can be determined which energy meter changes are closely related to the total deviation. The correlation threshold can be set using statistical methods. For example, a period of historical data from normal operation can be selected, the correlation coefficients between all sub-meters and the total deviation can be calculated, and the higher quantile of these coefficients can be used as the threshold to ensure that only energy meters with a significantly stronger correlation than normal are flagged.

[0129] The contribution of the suspected abnormal sub-meter to the energy conservation deviation vector was analyzed using multiple linear regression. The contribution values ​​of all suspected abnormal sub-meters were summed to obtain the total impact value of the downstream node on the metering box.

[0130] When multiple suspected abnormal meter sub-meters are identified, a multiple linear regression model can be established, using the energy conservation deviation as the dependent variable and the key electrical characteristics of these meters as independent variables. Regression analysis can then be used to estimate the contribution of each suspected abnormal meter sub-meter to the total deviation. For meters not identified as suspected, their contribution can be considered zero.

[0131] The portion of the metering box's own abnormal contribution is obtained by subtracting the total impact value from the net characteristic value after excluding the influence of upstream nodes.

[0132] After the above two steps of elimination, what we finally obtain is the true contribution of the target metering box's own anomaly. If this contribution is still significant, it indicates that the metering box itself does indeed have a problem; conversely, if it is close to zero after elimination, it indicates that the real source of the anomaly may be upstream or downstream.

[0133] The final anomaly analysis report should be clear and structured to guide on-site operations and maintenance. The report should include at least the following sections: Basic information: Substation area number, metering box ID, and time window in which the anomaly occurred.

[0134] Diagnostic conclusions: main anomaly type (e.g., loose incoming terminal), confidence level, and quantitative value of the contribution of the meter box itself to the anomaly (e.g., power loss caused by its own anomaly during this period).

[0135] Upstream impact analysis: List of affected upstream nodes, quantitative contribution value of each upstream node, and quantitative contribution value of the total upstream nodes.

[0136] Downstream impact analysis: List of suspected abnormal electricity meters, quantitative contribution value of each suspected abnormal electricity meter, and quantitative value of total downstream impact.

[0137] Maintenance recommendations: Based on the diagnostic results, provide specific troubleshooting order suggestions, such as prioritizing the inspection of the incoming terminals of the target metering box, while simultaneously verifying the on-site operation of the downstream electricity meter with ID 001.

[0138] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0139] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0140] 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; 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, depending on actual needs.

[0141] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for determining abnormal electrical energy in a metering box, characterized in that, The specific steps include: Step 1: Collect voltage and current data from the transformer outgoing side and the metering box incoming side of the target low-voltage distribution area. Based on the voltage data, determine the phase difference between the transformer outgoing side and the metering box incoming side, and construct a real-time topology correlation matrix between the transformer and the metering box by combining the phase difference and voltage change relationship. Step 2: When a phase difference or voltage change is detected, update the real-time topology correlation matrix. Based on the real-time topology correlation matrix, obtain the electrical characteristics of the metering box, and calculate the difference sequence between the incoming power of each metering box and the sum of the power measured by the power meter in the metering box according to the energy conservation relationship, so as to form an energy conservation deviation vector. Step 3: Perform frequency domain analysis on the energy conservation deviation vector to obtain frequency domain energy characteristics, and match the frequency domain energy characteristics with a pre-built library of anomalous physical patterns to determine the anomaly type; Step 4: After matching and determining the anomaly type, determine the upstream and downstream nodes of the target metering box based on the real-time topological correlation matrix, eliminate the influence of upstream and downstream nodes on the energy conservation deviation, separate the anomaly contribution of the target metering box, and output the anomaly analysis results.

2. The method for determining electrical energy anomalies in a metering box according to claim 1, characterized in that, Acquire voltage and current sampling data from the transformer outgoing side and the incoming side of each metering box, and synchronize and align the voltage and current sampling data in time. The aligned sampled data is divided into multiple feature statistical windows according to the preset time segmentation rules; Within each of the aforementioned feature statistics windows, voltage characteristic parameters for phase difference calculation are determined based on voltage sampling data. These voltage characteristic parameters include at least voltage amplitude parameters and phase angle parameters characterizing the phase.

3. The method for determining abnormal electrical energy in a metering box according to claim 2, characterized in that, The phase difference calculation and the generation of the real-time topological correlation matrix include the following steps: Within each feature statistics window, the voltage phase angle on the transformer outgoing side is used as the reference phase. The difference between the voltage phase angle on the incoming side of each metering box and the reference phase is calculated to obtain the phase difference of each metering box within the feature statistics window. The metering boxes are sorted according to the size of the phase difference, and the sorted metering boxes are divided into at least two phase difference partitions based on the preset phase difference partition threshold. Each phase difference partition corresponds to a different feeder branch on the transformer outgoing line side. The phase difference partition threshold is adaptively determined according to the statistical distribution of the phase difference. For metering boxes that fall into the same phase difference zone, their relative position order in the corresponding feeder branch is determined according to the magnitude of the voltage amplitude on the incoming side of the metering box. Based on the feeder branch affiliation relationship corresponding to the phase difference partition and the relative position order within the same partition, a real-time topology association matrix describing the connection relationship between the transformer outgoing side and each metering box is constructed. Changes in the time series of phase difference and / or voltage amplitude are detected, and when a change is detected that meets preset conditions, an update of the real-time topology correlation matrix is ​​triggered.

4. The method for determining electrical energy anomalies in a metering box according to claim 3, characterized in that, The real-time update of the topological association matrix includes the following steps: When a change detection triggers an update, the phase difference of the target metering bin that has changed within one or more feature statistics windows after the update is recalculated. The recalculated phase difference is compared with the phase difference partition threshold: if the recalculated phase difference crosses the partition boundary, it is determined that the feeder branch affiliation of the target metering box has changed, and the upper-level association information of the target metering box in the real-time topology association matrix is ​​updated. If the recalculated phase difference does not cross the partition boundary, the relative position order of the target metering box in its feeder branch in the real-time topology association matrix is ​​updated according to whether the voltage amplitude change of the target metering box meets the preset conditions, while keeping its feeder branch affiliation unchanged.

5. The method for determining electrical energy anomalies in a metering box according to claim 3, characterized in that, Within one or more power frequency cycles corresponding to the feature statistics window, the voltage sampling data and current sampling data are decomposed in the frequency domain to obtain the fundamental component and the amplitude parameters corresponding to at least one set of preset harmonic components. The active power corresponding to the fundamental component and harmonic components is calculated, and the ratio of the active power of the harmonic components to the active power of the fundamental component is processed to obtain the harmonic content rate sequence. Based on the feeder branch attribution relationship determined by the real-time topology association matrix, the harmonic content rate sequences of each metering box within the same feeder branch are compared horizontally, and metering boxes whose harmonic content rate deviates from the statistical level of the feeder branch by more than the preset condition are identified as candidates for harmonic anomalies. Obtain the three-phase voltage amplitude and three-phase current amplitude, and calculate the negative sequence unbalance and zero sequence unbalance based on the symmetrical components; Within the feature statistics window, statistical analysis is performed on negative-order imbalance and zero-order imbalance to obtain feature vectors characterizing their dynamic evolution. The feature vectors include at least the change magnitude feature and the change persistence feature. Obtain the total active power time series on the incoming side of the metering box and the active power time series of each sub-meter in the metering box, and perform normalization processing. The correlation between the total active power time series and the active power time series of each sub-meter is calculated using a sliding time window to obtain the correlation sequence; Within each feature statistical window, statistical analysis is performed on the correlation sequences falling into that window to obtain at least two indicators for quantifying the synchronicity of load fluctuations. The indicators include at least the correlation mean indicator and the correlation dispersion indicator. The electrical characteristics include harmonic anomaly candidates, feature vectors characterizing the dynamic evolution of negative-sequence and zero-sequence unbalance, and indices used to quantify load fluctuation synchronicity.

6. The method for determining abnormal electrical energy in a metering box according to claim 5, characterized in that, The formation of the energy conservation deviation vector specifically includes: For each characteristic statistical window, the cumulative value of active power on the incoming side of the metering box within that characteristic statistical window is obtained as the total incoming energy; and the cumulative values ​​of active power of each energy meter in the metering box within that characteristic statistical window are summed to obtain the sum of the energy of each meter. Based on the metering box connection relationship and / or relative position within the branch determined by the real-time topology correlation matrix, the line loss energy used for energy correction is determined, and the line loss energy is deducted from the total incoming line energy as a fixed line loss energy. The energy difference between the total incoming energy, the energy measured by the sub-meter, and the fixed line loss energy is processed to obtain the energy conservation deviation value within the characteristic statistical window; The energy conservation deviation values ​​corresponding to multiple feature statistical windows are arranged in chronological order to form an energy conservation deviation vector, where the positive or negative sign of the deviation value is used to characterize the direction of the deviation.

7. The method for determining electrical energy anomalies in a metering box according to claim 6, characterized in that, The extraction of the frequency domain energy features includes: The energy conservation deviation vector is decomposed into multiple frequency bands by performing time-frequency transformation on the energy conservation deviation vector. Calculate the energy value of each frequency band within each characteristic statistical window, construct a frequency domain energy spectrum distribution map of energy variation with frequency, and store it in matrix form, where the matrix elements are used to characterize the energy value of the corresponding frequency band within the corresponding characteristic statistical window; The preset frequency range is divided into multiple specific frequency bands, and the energy values ​​within each specific frequency band are aggregated to obtain the frequency domain energy characteristics corresponding to each specific frequency band.

8. The method for determining electrical energy anomalies in a metering box according to claim 7, characterized in that, The construction and matching of the abnormal physical pattern library includes: The corresponding sample energy conservation deviation vectors are obtained in advance for various metering box anomaly types and converted into frequency domain energy spectrum distribution maps to extract the standard frequency domain energy features corresponding to each anomaly type, thereby establishing the correspondence between anomaly types and standard frequency domain energy features and forming an anomaly physical pattern library. The frequency domain energy characteristics of the metering box under test are extracted to form the feature vector to be measured; the standard feature vectors corresponding to each anomaly type are retrieved from the abnormal physical pattern library; Calculate the similarity between the feature vector to be tested and each standard feature vector, and determine the anomaly type that matches the measuring box to be tested based on the similarity results.

9. The method for determining electrical energy anomalies in a metering box according to claim 2, characterized in that, The topological location of the target metering box is determined in the real-time topological correlation matrix, and the upstream node set of the target metering box is determined by tracing upstream based on the real-time topological correlation matrix. The upstream node set includes nodes located on the power supply path from the power supply side to the target metering box; at the same time, the downstream node set of the target metering box is determined. For the same feature statistical window, extract the frequency domain energy features of the target metering box, and then extract the frequency domain energy features of each upstream node in the upstream node set in turn. When the abnormal feature value of an upstream node and the abnormal feature value of the target meter box meet the preset similarity conditions, the ratio of the abnormal feature value of the upstream node and the abnormal feature value of the target meter box is processed to obtain the contribution coefficient of the upstream node to the abnormality of the target meter box, and the contribution amount of the upstream node is obtained according to the contribution coefficient. The total upstream contribution is obtained by summing the contributions of each upstream node, and then the total upstream contribution is deducted from the abnormal characteristic value of the target metering box to obtain the net characteristic value after removing the influence of upstream nodes.

10. The method for determining electrical energy anomalies in a metering box according to claim 9, characterized in that, The results of eliminating the influence of downstream nodes and outputting anomaly analysis include: Obtain the electrical characteristics of each downstream energy meter in the downstream node set within the same feature statistics window, and calculate the correlation between the electrical characteristics of each downstream energy meter and the energy conservation deviation vector of the target metering box. When the correlation meets the preset conditions, the corresponding downstream electricity meter is identified as a suspected abnormal downstream node, and the influence of the suspected abnormal downstream node on the energy conservation deviation vector of the target meter box is estimated. The total downstream impact value is obtained by summing the impact of each suspected abnormal downstream node, and then the total downstream impact value is subtracted from the net feature value to obtain the feature value corresponding to the abnormal contribution of the target metering box itself. Output the topological location information of the target metering box, the anomaly type determination result, the total upstream contribution, and the total downstream impact value.

Images