Method and system for detecting abnormal power consumption of low-voltage power distribution network based on big data

Abstract

The invention relates to a method and a system for detecting abnormal power consumption of a low-voltage power distribution network based on big data, belongs to the technical field of power system monitoring, and aims to solve the problem that electricity stealing and other unmeasured abnormal power consumption behaviors in the low-voltage power distribution network are difficult to accurately and effectively monitor in the prior art. The method comprises the following steps: constructing a voltage drop-current inconsistency characteristic based on a split-phase voltage difference and a downstream current sum of adjacent nodes in a power distribution topology, calculating an instantaneous equivalent ratio of the voltage drop-current inconsistency characteristic, and comparing the instantaneous equivalent ratio with a historical reference to obtain a standardized fluctuation value; meanwhile, analyzing an empirical relationship between the 15-minute line loss rate and the power supply amount, and calculating a standardized residual error of the line loss rate; and finally, performing fusion and weighted scoring on the abnormal signals of the two dimensions through a sliding window comprehensive research and judgment mechanism, and when a comprehensive abnormal score exceeds a threshold value, judging that an abnormal power consumption behavior exists and reporting the abnormal power consumption behavior. According to the invention, the problem of high false alarm rate of a single index is effectively overcome, and high-precision detection and positioning of abnormal power consumption behaviors are realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power system monitoring and power consumption behavior analysis, and particularly relates to a low-voltage distribution network abnormal power consumption detection method and system based on big data, especially a comprehensive analysis method combining the "voltage drop-current inconsistency" feature and the 15-minute line loss rate, which is used for positioning electricity stealing and other un-metered abnormal power consumption behaviors in the low-voltage distribution network. BACKGROUND

[0002] With the global economic development, industrial expansion and technological progress, the demand for electricity continues to grow, especially under the dual driving of economic recovery in low-income countries and social and technological progress in developed countries, the load level of distribution network continues to rise. There are two categories of technical and non-technical losses in the operation of the distribution system, and the non-technical loss (NTL) is mainly caused by electricity stealing, metering device failure, and poor management, etc., which causes huge economic losses to the power company. Public statistics show that electricity stealing causes global economic losses of tens of billions of dollars each year, and nearly six hundred million dollars in the top 50 developing countries. Electricity stealing behavior also induces a series of safety hazards such as voltage drop, equipment overload, power supply reliability decline, and increased risk of power outages.

[0003] The promotion of smart grid and smart meters provides a basic support for improving the anti-electricity stealing capability. Smart meters can collect high-frequency voltage, current, power and other historical data, which are used to analyze user power consumption patterns and improve the accuracy of abnormal power consumption detection. However, actual electricity stealing behavior often has diversity and concealment, such as avoiding metering by combining various means such as phase shifting, undercurrent, undervoltage, bypass wiring, and pre-meter private installation, which makes it difficult to identify in time and accurately by relying on simple rules or threshold judgments of a single point and single dimension.

[0004] In the prior art, one class of methods focuses on identifying electricity stealing by analyzing user power consumption patterns based on single-meter load curves, but has limited positioning capability for specific physical locations; another class of methods focuses on analyzing the overall line loss rate of the transformer area, and triggers manual investigation when the line loss rate is long-term over-standard, but has insufficient positioning accuracy, delayed response, and is easily disturbed by factors such as natural three-phase imbalance, device noise, and changes in operating environment. Especially in the low-voltage distribution network scenario, there are many branches and complex user types, and the existing methods are difficult to accurately identify and locate electricity stealing users with a reasonable false positive rate.

[0005] Therefore, there is an urgent need for a low-voltage distribution network abnormal power consumption accurate detection and positioning method that can fully utilize the multi-dimensional power data collected by the master station, combine the characteristics of the distribution network topology, and balance the detection sensitivity and anti-interference capability. SUMMARY

[0006] The purpose of the present application is to overcome the defects of the prior art, provide a low-voltage power distribution network abnormal power consumption detection method and system based on big data, which can effectively reduce false positives and improve detection accuracy, by analyzing historical power data, combining time series power data with substation topology structure to dynamically detect abnormal power consumption behavior in the power distribution network.

[0007] The technical scheme of the present application is:

[0008] In a first aspect, the present application provides a detection method for low-voltage power distribution network abnormal power consumption, comprising the following steps: S1, obtaining the topology structure data of the target low-voltage power distribution network substation, and the time series electrical quantity data of the substation total table and the user electric meter, the electrical quantity data at least including voltage, phase current and power consumption;

[0009] S2, for any edge e=(i→j) in the topology, representing a line from upstream node i to downstream node j, for any edge e, there is a phase p∈{A,B,C}, the phase voltage drop of the adjacent node at time t is calculated according to the phase voltage of the adjacent node, and the downstream node phase current and are obtained, and the instantaneous equivalent ratio is calculated

[0010] S3, based on the historical data sequence of the instantaneous equivalent ratio, the dynamic reference benchmark, the robust standard deviation and the standardized fluctuation value of the equivalent R value are calculated;

[0011] S4, based on the total power supply of the substation, the 15-minute line loss rate is calculated, and the experience model of line loss rate and power supply is established by using historical data The standardized residual error of line loss rate is calculated to obtain the standardized fluctuation value of line loss rate;

[0012] S5, setting a sliding time window, for each time point t in the window W, when the standardized fluctuation value of the equivalent R value is greater than the threshold value and the standardized fluctuation value of the 15-minute line loss rate is greater than the threshold value, mark the point as an effective suspicious point;

[0013] S6, in the sliding time window, the continuity counting and abnormal intensity quantification of the continuously appearing effective suspicious points are carried out, and the comprehensive abnormal score in the window is calculated;

[0014] S7, comparing the comprehensive abnormal score with the preset decision threshold value, if the comprehensive abnormal score is greater than the decision threshold value, it is determined that there is abnormal power consumption behavior in the line in the window period, and an alarm information is generated.

[0015] Further, the calculation of the instantaneous equivalent ratio Comprises:

[0016] The phase voltage drop is calculated according to the phase voltage of the adjacent node Wherein Let be the phase voltages of upstream node i and downstream node j at time t, respectively;

[0017] Obtain the downstream node phase current and The downstream node phase current sum is the sum of the phase currents flowing from downstream node j to all its downstream loads;

[0018] Calculate the instantaneous equivalent ratio

[0019] Furthermore, S3 includes:

[0020] S31. Obtain the instantaneous equivalent ratio sequence within a preset historical time window. Where τ represents the moment within the historical time window, the median of the instantaneous equivalent ratio sequence is selected as the dynamic reference benchmark.

[0021] S32. Obtain the median absolute deviation (MAD) of the instantaneous equivalent ratio sequence and calculate the robust standard deviation. An estimate is made, where k is a constant;

[0022] S33. Calculate the standardized fluctuation value of the equivalent R value.

[0023] Furthermore, S4 includes:

[0024] S41. Calculate the 15-minute line loss rate. Among them, E sup (t) represents the total power supply of the transformer area, ∑E user (t) represents the sum of electricity consumption for all users;

[0025] S42. Build an empirical model using historical data through linear regression. Where a and b are regression coefficients;

[0026] S43. Calculate the residuals

[0027] S44, If E sup (t) is lower than the preset minimum load threshold E min Then set the standardized volatility value Z of the line loss rate. L (t) represents a normal value;

[0028] Otherwise, calculate the robust standard deviation σ of the residual series. L Take the minimum value of (1.4826 * MAD_r, 10th quantile - 90th quantile), where MAD_r is the median absolute deviation of the residual series; calculate the standardized volatility of the line loss rate.

[0029] Further, in S44, the robust standard deviation σ of the residual sequence is calculated. L include:

[0030] For the residual sequence r(τ), calculate the median absolute deviation MAD_r = median{|r(τ)-median{r(τ)}|};

[0031] Obtain the 90th and 10th quantiles of the residual series, and take the minimum of the median absolute deviation and the quantile difference as the robust standard deviation σ. L , min(1.4826*MAD_r, 10% quantile-90% quantile).

[0032] Furthermore, in S6, the continuous counting of consecutively occurring valid suspicious points includes:

[0033] Starting from the first valid suspicious point within the sliding time window, the continuity count C will be... t Initialize to 1; if the next time point is a valid suspicious point, then C t =C t +1; if the next time point is not a valid suspicious point, then C t Reset to 1.

[0034] Furthermore, in S6, the anomaly intensity quantification and the comprehensive anomaly score within the calculation window include;

[0035] The normalized fluctuation value z_score of the equivalent R value is mapped to obtain the anomaly intensity quantization value Q. t ;

[0036] Calculate the overall anomaly score S within the sliding time window. W =∑ t∈W (C t ×Q t ).

[0037] Furthermore, S1 also includes receiving the topology data and time-series electrical quantity data from the main station system via an API interface; preprocessing the time-series electrical quantity data, the preprocessing including missing value imputation, time alignment, and anomaly screening;

[0038] S7 also includes reporting alarm information to the main station system via API interface when abnormal power consumption behavior is determined to exist. The alarm information includes the time and location of the abnormality.

[0039] Secondly, an abnormal power consumption detection system for implementing the method includes:

[0040] The data acquisition module is used to receive topology data and time-series electrical quantity data of the low-voltage distribution network from the master station system;

[0041] The feature calculation module is used to calculate the standardized volatility of the equivalent R value and the standardized volatility of the line loss rate;

[0042] The comprehensive analysis module is used to execute dual-condition triggering, continuous counting, and weighted scoring within the sliding window to generate a comprehensive anomaly score;

[0043] The alarm output module is used to report alarm information, including the time and location of the anomaly, to the main station system when an anomaly is detected.

[0044] A computer device having a computer program stored thereon, which, when executed by a processor, implements the method.

[0045] The beneficial effects of this invention are:

[0046] This invention discloses a method and system for detecting abnormal power consumption in low-voltage distribution networks. Through topology analysis and electrical quantity data monitoring, it proposes an innovative solution to the problem of accurately locating and detecting abnormal power consumption behavior in real time in low-voltage distribution networks. The specific priorities are as follows:

[0047] 1. Driven by physical mechanism and with high positioning accuracy: This invention utilizes the physical constraint relationship between "voltage drop - current sum" in low-voltage distribution networks. By constructing the "voltage drop - current inconsistency" feature and monitoring the abnormal rise of equivalent impedance, it reveals the impact of unmetered bypass branches on the electrical quantities of the line from a physical mechanism perspective, thereby achieving precise positioning of abnormal power consumption locations.

[0048] 2. Multi-dimensional feature fusion significantly reduces false alarm rate: This invention combines the "voltage drop-current inconsistency" feature with the abnormal fluctuation of line loss rate on a 15-minute scale, and introduces line loss rate gating and low load protection mechanisms to effectively filter out false anomalies caused by natural three-phase imbalance, operating noise, etc., thereby improving the reliability and robustness of detection results.

[0049] 3. Robust statistical modeling to adapt to complex working conditions and abnormal data contamination: This invention extensively uses robust statistical methods such as median and median absolute deviation (MAD) in estimating the baseline of equivalent ratio and the fluctuation range of line loss residuals, and introduces upper limit constraints for quantiles, which can effectively resist the interference of isolated outliers or short-term disturbances in historical data and ensure the stability of the baseline model in long-term operation.

[0050] 4. Sliding window scoring system, summarizing and considering the intensity and duration of anomalies: This invention adopts a continuous scoring and fluctuation value z_score weight quantification method within a sliding time window, so that the duration and intensity of anomalies can be reflected in the window score, distinguishing between short-term spikes and continuous anomalies, and improving the sensitivity to real electricity theft.

[0051] 5. Easy to implement in engineering and easy to integrate with existing master station systems: This invention works in collaboration with the master station data acquisition system through a standardized API interface. It can be deployed based on existing smart meters and transformer area topology data without the need for additional expensive on-site hardware equipment, and has good engineering application value and promotion prospects.

[0052] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0053] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the invention.

[0054] Figure 1 A flowchart of the low-voltage power consumption detection method based on big data of the present invention is shown.

[0055] Figure 2 A data interaction flowchart between the master station system and the abnormal power consumption detection system in an embodiment of the present invention is shown. Detailed Implementation

[0056] Preferred embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0057] like Figure 1 and Figure 2 As shown, this invention provides a method and system for detecting abnormal power consumption in low-voltage distribution networks based on big data, the execution of which depends on coordination with the power master station system. The master station system pushes voltage, current, and power consumption data of the main meter and user meters of the transformer substations for the past hour (granularity of 15 minutes, a total of 4 points) to the server of this system on an hourly basis.

[0058] Taking phase A of a three-phase line in a certain transformer area as an example, the specific steps are as follows:

[0059] 1. Data preparation and preprocessing: The algorithm service receives data pushed by the main site. After obtaining the data, it needs to organize the data into time series data so that the algorithm can analyze it.

[0060] At the same time, based on the pre-stored transformer area topology, the specific location of the smart meter in the low-voltage distribution network is perceived, and the line edge e = (i → j) participating in the calculation is determined to represent a line segment from upstream node i = up to downstream node j = down (such as from the transformer outlet to the first branch box).

[0061] 2. Calculate the "voltage drop-current inconsistency" characteristic:

[0062] After acquiring the data, it is necessary to calculate the voltage of the upstream node at time t. Downstream node voltage and downstream current and

[0063] After calculating the voltages of adjacent nodes, the phase voltage drop of adjacent nodes can be calculated.

[0064] After the above calculations, the instantaneous equivalent ratio can be calculated.

[0065] Then, take the historical data window of the previous 7 days (7*96=672 points in total), and base it on the instantaneous equivalent ratio sequence within the window. Calculate two core baseline statistics: the baseline median. and robust standard deviation

[0066] Finally, we can calculate the standardized fluctuation value of the equivalent R value at the current moment. Where ε is a very small positive number to ensure that the denominator is not zero, the z_score quantitatively characterizes the degree of deviation of the current observation from the historical normal range;

[0067] Set a threshold θ1 for the calculated z_score value, if This is recorded as an abnormal equivalent ratio value, indicating that there may be abnormal electricity consumption downstream, resulting in unmetered electricity consumption.

[0068] 3. Calculate the 15-minute line loss rate anomaly:

[0069] Calculate the power supply E of phase A in transformer substation at time t. sup A(t) and the sum of electricity consumption of all users in phase A, ∑E user A(t) yields the line loss rate.

[0070] Fitting a model using historical non-outlier data

[0071] Calculate residuals

[0072] Calculate the robust standard deviation σ based on the residual history series. L (Min(1.4826*MAD, 10th percentile - 90th percentile));

[0073] Calculate the standardized volatility of the line loss rate If Esup A(t) <E min If a minimum power supply threshold is set, then this step is skipped.

[0074] 4. Comprehensive assessment:

[0075] Set a sliding window W of length L = 24 (i.e., 6 hours); within W, for each time point t, if simultaneously satisfying... And Z L If (t) > θ2, then mark it as a valid suspicious point;

[0076] Perform continuous counting of valid suspicious points within the window (C) t and weighted quantization Q t ;

[0077] Among them, the standardized fluctuation value z_score of the equivalent R value is weighted and quantized as Q. t This amplifies the impact of high-intensity anomalous signals. The mapping rule is z_score∈[2,3)→Q t =2, z_score∈[3,4)→Q t =3; z_score∈[4,5)→Q t =4; and so on;

[0078] Calculate the window's overall anomaly score S W =∑ t∈W (C t ×Q t );

[0079] If S W >S th (e.g. S) th If the value is 200, then it is determined that phase A of the line has highly suspicious abnormal electricity consumption behavior within the current 6-hour window.

[0080] 5. Result Reporting: The algorithm service will return the judgment results (including abnormal areas, lines, phases, and time windows) to the main station system through the API interface. The main station system will generate a work order and push it to the operation and maintenance personnel for on-site verification.

[0081] In this embodiment, the instantaneous equivalent ratio and standardized fluctuation value are calculated, and combined with line loss rate residual analysis, a dual-condition triggering mechanism is constructed. Continuous counting within a sliding time window and anomaly intensity quantification are introduced to generate a comprehensive anomaly score for judgment. This method effectively captures the correlation characteristics between local line anomalies and overall line loss anomalies, significantly improving the detection accuracy and real-time performance of abnormal power consumption behavior. It achieves efficient identification and alarm of abnormal power consumption in low-voltage distribution networks, reduces grid operation risks, and provides important technical support for the safe and stable operation of the power system.

[0082] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for detecting abnormal power consumption in low-voltage distribution networks based on big data, characterized in that, Includes the following steps: S1. Obtain the topology data of the target low-voltage distribution network area, as well as the time-series electrical quantity data of the main meter and user meters of the area. The electrical quantity data includes at least voltage, phase current and power consumption. S2. For any edge e = (i→j) in the topology, representing a line segment from upstream node i to downstream node j, and for any edge e, having phases p∈{A,B,C}, calculate the phase voltage drop of the adjacent nodes at time t based on the phase voltages of the adjacent nodes, and obtain the phase current of the downstream node, and calculate the instantaneous equivalent ratio. S3. Based on the historical data sequence of the instantaneous equivalent ratio, calculate its dynamic reference benchmark, robust standard deviation, and standardized fluctuation value of the equivalent R value; S4. Based on the total power supply of the transformer area, calculate the 15-minute line loss rate and establish an empirical model of line loss rate and power supply using historical data. Calculate the standardized residual of the line loss rate to obtain the standardized fluctuation value of the line loss rate; S5. Set a sliding time window. For each time point t within the window W, when the standardized fluctuation value of the equivalent R value is greater than the threshold and the standardized fluctuation value of the 15-minute line loss rate is greater than the threshold, mark the point as a valid suspicious point. S6. Within the sliding time window, continuously count and quantify the abnormality intensity of continuously appearing valid suspicious points, and calculate the comprehensive abnormality score within the window. S7. Compare the comprehensive anomaly score with the preset decision threshold. If the comprehensive anomaly score is greater than the decision threshold, it is determined that there is abnormal power consumption behavior in the line during the window period, and an alarm message is generated.

2. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 1, characterized in that, The calculation of the instantaneous equivalent ratio value include: Calculate the phase voltage drop based on the phase voltages of adjacent nodes. in Let be the phase voltages of upstream node i and downstream node j at time t, respectively; Obtain the downstream node phase current and The downstream node phase current sum is the sum of the phase currents flowing from downstream node j to all its downstream loads; Calculate the instantaneous equivalent ratio 3. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 1, characterized in that S3 include: S31. Obtain the instantaneous equivalent ratio sequence within a preset historical time window. Where τ represents the moment within the historical time window, the median of the instantaneous equivalent ratio sequence is selected as the dynamic reference benchmark. S32. Obtain the median absolute deviation (MAD) of the instantaneous equivalent ratio sequence and calculate the robust standard deviation. An estimate is made, where k is a constant; S33. Calculate the standardized fluctuation value of the equivalent R value.

4. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 1, characterized in that S4 include: S41. Calculate the 15-minute line loss rate. Among them, E sup (t) represents the total power supply of the transformer area, ∑E user (t) represents the sum of electricity consumption for all users; S42. Build an empirical model using historical data through linear regression. Where a and b are regression coefficients; S43. Calculate the residuals S44, If E sup (t) is lower than the preset minimum load threshold E min Then set the standardized volatility value Z of the line loss rate. L (t) represents a normal value; Otherwise, calculate the robust standard deviation σ of the residual series. L Take the minimum value of (1.4826 * MAD_r, 10th quantile - 90th quantile), where MAD_r is the median absolute deviation of the residual series; calculate the standardized volatility of the line loss rate.

5. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 4, characterized in that... In S44, the robust standard deviation σ of the residual sequence is calculated. L include: For the residual sequence r(τ), calculate the median absolute deviation MAD_r = median{|r(τ)-median{r(τ)}|}; Obtain the 90th and 10th quantiles of the residual series, and take the minimum of the median absolute deviation and the quantile difference as the robust standard deviation σ. L , min(1.4826*MAD_r, 10% quantile-90% quantile).

6. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 1, characterized in that... In S6, the continuous counting of consecutively occurring valid suspicious points includes: Starting from the first valid suspicious point within the sliding time window, the continuity count C will be... t Initialize to 1; if the next time point is a valid suspicious point, then C t =C t +1; if the next time point is not a valid suspicious point, then C t Reset to 1.

7. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 6, characterized in that... In S6, the abnormality intensity quantification and the comprehensive abnormality score within the calculation window include; The normalized fluctuation value z_score of the equivalent R value is mapped to obtain the anomaly intensity quantization value Q. t ; Calculate the overall anomaly score S within the sliding time window. W =∑ t∈W (C t ×Q t ).

8. The method for detecting abnormal power consumption in low-voltage distribution networks based on big data as described in claim 1, characterized in that, S1 also This includes receiving the topology data and time-series electrical quantity data from the main station system via an API interface; and preprocessing the time-series electrical quantity data, including missing value imputation, time alignment, and initial anomaly screening. S7 also includes reporting alarm information to the main station system via API interface when abnormal power consumption behavior is determined to exist. The alarm information includes the time and location of the abnormality.

9. An abnormal power consumption detection system for implementing the method according to any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to receive topology data and time-series electrical quantity data of the low-voltage distribution network from the master station system; The feature calculation module is used to calculate the standardized volatility of the equivalent R value and the standardized volatility of the line loss rate; The comprehensive analysis module is used to execute dual-condition triggering, continuous counting, and weighted scoring within the sliding window to generate a comprehensive anomaly score; The alarm output module is used to report alarm information, including the time and location of the anomaly, to the main station system when an anomaly is detected.

10. A computer device having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images