Photovoltaic low-voltage transformer area line loss abnormity sensing method considering intermittent electricity stealing
By constructing feature vectors for low-voltage photovoltaic distribution areas and using adaptive learning algorithms to identify electricity theft users, the problem of abnormal line loss detection after distributed photovoltaic access has been solved, achieving accurate identification of electricity theft and improving grid security.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING INST OF TECH
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-01
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The large-scale integration of distributed photovoltaic power generation has increased the difficulty of detecting abnormal line losses. Traditional methods cannot dynamically adapt to fluctuations in line loss rates, and the concealment of electricity theft methods makes it difficult to distinguish between abnormal and normal fluctuations, resulting in a lack of multi-level diagnostic capabilities.
By acquiring key operating indicators of low-voltage photovoltaic distribution areas, calculating cross-correlation functions and maximum information coefficients, constructing feature vectors, using the isolated forest algorithm to detect abnormal distribution areas, and combining adaptive ensemble learning and a two-state hidden Markov model to identify electricity theft users, the line loss rate and photovoltaic penetration rate are corrected to identify the causes of anomalies.
It enables accurate identification of power consumption anomalies, photovoltaic anomalies, and electricity theft, improving the accuracy of anomaly cause diagnosis and the economy and security of the power grid, and solving the problem of unbalanced electricity theft samples.
Smart Images
Figure CN121965523A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power system monitoring technology, specifically relating to a method for detecting abnormal line losses in photovoltaic low-voltage distribution areas that takes into account intermittent electricity theft. Background Technology
[0002] With the rapid development of the national economy, my country's power grid construction, especially in areas affecting residential life, has undergone significant changes. The large-scale integration of distributed photovoltaic (PV) power has greatly impacted the power flow characteristics of these areas. Due to the randomness and uncertainty of its output, these areas may experience increased line losses and reduced reliability. Furthermore, the mismatch and incoordination between PV and load further limit the absorption capacity of distributed PV power.
[0003] The large-scale integration of distributed photovoltaic (PV) power has exacerbated the difficulty of detecting line loss anomalies: bidirectional power flow leads to significant fluctuations in line loss rates, which traditional threshold methods cannot dynamically adapt to; the coupling of multiple factors (such as PV penetration rate, three-phase imbalance, and power factor) requires the establishment of highly interpretable correlation models; and the concealment of electricity theft methods (such as tampering with PV metering data) makes it difficult to distinguish between anomalies and normal fluctuations. Existing methods are mostly designed for traditional unidirectional power supply scenarios, rely on grid parameters or ignore the influence of PV, and lack multi-level diagnostic capabilities for technical anomalies (such as equipment overload) and non-technical anomalies (such as electricity theft). Summary of the Invention
[0004] This invention addresses the shortcomings of existing technologies by providing a method for detecting abnormal line losses in low-voltage photovoltaic distribution areas that takes into account intermittent electricity theft. This method is topology-independent and adaptable to photovoltaic grid connection, and combines anomaly detection, cause diagnosis, and data correction functions, thereby improving the economy and security of the power grid.
[0005] This invention provides the following technical solution:
[0006] A method for detecting abnormal line losses in low-voltage photovoltaic distribution areas, considering intermittent electricity theft, includes: Step S1: Obtain the key operating indicators for each low-voltage photovoltaic distribution area, including photovoltaic penetration rate and several secondary indicators; Step S2: For each transformer substation, calculate the cross-correlation function and maximum information coefficient between each key operating indicator and the line loss rate, and construct the feature vector for each transformer substation; Step S3: After summarizing the feature vectors of all transformer areas, input them into the Isolation Forest algorithm to detect outliers and filter out abnormal transformer areas; Step S4: By comparing the deviation of key operating indicators of abnormal transformer areas from those of normal transformer areas, identify the causes of abnormality in each abnormal transformer area, including: power consumption-side abnormalities, photovoltaic abnormalities, and minor indicator abnormalities. Step S5: Output the results for abnormal distribution areas where the cause of the abnormality is an abnormality on the power consumption side or a minor indicator abnormality; at the same time, identify electricity theft for each abnormal photovoltaic distribution area. If there are no electricity theft users, output the result as photovoltaic technology abnormality as the cause of the abnormality. If there are electricity theft users, correct the line loss rate and photovoltaic penetration rate of the abnormal photovoltaic distribution area, and determine whether the abnormal photovoltaic distribution area has recovered to a normal distribution area. If it has recovered, output the electricity theft users, the electricity theft period, and the cause of the abnormality as photovoltaic user electricity theft. If it has not recovered, return to step S4 to re-identify the cause of the abnormality, and combine the electricity theft identification result with the re-identified cause of the abnormality before outputting it.
[0007] Optionally, some of the aforementioned secondary indicators are one or more of the following: three-phase imbalance rate, load factor, and power factor.
[0008] Optionally, step S2 specifically involves: calculating the key operating indicators in the current transformer area according to the following formula. Cross-correlation function and maximum information content of time series and line loss rate time series; ; ; in, and Key performance indicators and line loss rate The cross-correlation function and the maximum information number, Key performance indicators Time series, , The length of the time series. Line loss rate Time series, , Key performance indicators and line loss rate The joint probability density, and Key performance indicators and line loss rate marginal probability density, and These represent the number of intervals into which each dimension is divided. ; Based on all key operational indicators of the transformer area, construct the transformer area feature vector: ;in, Taiwan District eigenvectors, , , and They are respectively the Taiwan area Line loss rate and photovoltaic penetration rate Three-phase imbalance rate Load factor and power factor The maximum number of information items.
[0009] Optionally, step S3 specifically includes: The feature vectors of all transformer areas are summarized to form a total matrix. , , The number of stations, Taiwan District eigenvectors; Outlier detection is performed using the isolated forest algorithm to obtain anomaly scores for each transformer area. Transformers with anomaly scores exceeding a set threshold are marked as abnormal transformer areas, while the rest are normal transformer areas.
[0010] Optionally, step S4 specifically includes: For each abnormal transformer area, the deviation of its key operating indicators from the average key operating indicators of normal transformer areas is calculated and scored. Key operating indicators with scores exceeding the set value are classified into the outlier set. If the outlier set includes all key operating indicators, the abnormality of the current abnormal transformer area is due to an abnormality on the power consumption side. If the outlier set does not include all key operating indicators, it is determined whether the score of photovoltaic penetration rate is the maximum value. If it is the maximum value, the abnormality is due to a photovoltaic abnormality; otherwise, the abnormality is due to a minor indicator abnormality.
[0011] Optionally, in step S5, electricity theft identification is performed on each abnormal photovoltaic distribution area. Specifically, an adaptive ensemble learning algorithm is used to identify suspicious users in the abnormal photovoltaic distribution area, and a dual-state hidden Markov model is used to independently analyze each suspicious user to identify the electricity theft user and its corresponding electricity theft period.
[0012] Optionally, the step of identifying suspicious users in abnormal photovoltaic distribution areas using an adaptive ensemble learning algorithm, and independently analyzing each suspicious user using a dual-state hidden Markov model to identify electricity theft users and their corresponding theft periods, specifically involves: By using the Pearson correlation coefficient to screen meteorological indicators that are strongly correlated with photovoltaic power generation, and combining them with the long-term unit capacity power generation of users in the photovoltaic abnormal area, the time-series characteristics of each user in the photovoltaic abnormal area are obtained. The hyperparameters of the RUSBoost model are iteratively optimized using a Bayesian optimization algorithm. For each user in the abnormal photovoltaic distribution area, their time-series characteristics are input into the optimized RUSBoost model to obtain the user's probability of electricity theft at each time point; based on the electricity theft probability sequence, suspicious users are initially screened. If there are no suspicious users, then there are no users stealing electricity, and the cause of the abnormality in the photovoltaic abnormal area is photovoltaic technology abnormality as the output result; if there are suspicious users, then the probability of electricity theft by suspicious users is used to construct an observation sequence, and a two-state hidden Markov model is used to analyze each suspicious user, decode to obtain the optimal hidden state sequence, so as to identify electricity theft users and several electricity theft periods of electricity theft users.
[0013] Optionally, the step of constructing an observation sequence using the probability of electricity theft by suspicious users, and analyzing each suspicious user using a two-state hidden Markov model to decode and obtain the optimal hidden state sequence, specifically: Constructing observation sequences ,in, express The observed value at time, , For the current user The probability of electricity theft at any given moment; Establish a two-state hidden Markov model and define the set of hidden states. State transition probability matrix and initial state probability , , This indicates normal power usage. Indicates a state of electricity theft; Construct the likelihood function and define the observed emission probability. , , Representing state The following observations The probability of; , Indicates the current hidden state; The Viterbi dynamic programming algorithm is used to find the optimal hidden state sequence and identify several intermittent periods of electricity theft.
[0014] Optionally, step S5 further includes: optimizing the decoded optimal hidden state sequence to obtain an optimized electricity theft period, specifically: For each user's optimal hidden state sequence, a sliding window is used to obtain the local anomaly density within each window; Construct the dynamic local anomaly density threshold for the current user: ;in, This serves as the baseline value for the local anomaly density of the current user. The standard deviation of the historical local anomaly density for the current user. This represents the current user's dynamic local anomaly density threshold. This is the sensitivity coefficient; The system sequentially checks whether the local anomaly density within each window is greater than the current user's dynamic local anomaly density threshold. If so, the current window is considered a period of electricity theft; otherwise, the current window is considered a period of normal electricity use.
[0015] Optionally, in step S5, the line loss rate of the abnormal photovoltaic distribution area is corrected using the following formula: ; in, To update the line loss rate, This represents the original line loss value for the current abnormal photovoltaic distribution area. This refers to the number of solar power theft users within the distribution area. For users who steal electricity in the current abnormal photovoltaic distribution area The installed capacity, and For users who steal electricity in the current abnormal photovoltaic distribution area The measured value of power generation per unit capacity and the average value of power generation of normal photovoltaic users. The power supply of a general-purpose electricity meter during the same period. For photovoltaic users in the current abnormal photovoltaic distribution area Electricity generated per unit capacity This represents the total number of photovoltaic users in the current photovoltaic anomaly area.
[0016] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention can identify abnormal distribution areas by combining MIC and IF algorithms, and identify power consumption anomalies, photovoltaic anomalies and other minor indicator anomalies by analyzing the causes of abnormal distribution areas, thereby distinguishing between power consumption anomalies and minor indicator technical anomalies; This invention further identifies abnormal photovoltaic distribution areas by identifying electricity theft, thereby distinguishing between photovoltaic technical anomalies and photovoltaic user electricity theft anomalies; This invention re-analyzes the causes of anomalies after correcting for line loss rate and photovoltaic penetration rate, which can ensure the accuracy of photovoltaic user electricity theft anomaly identification, and further identify whether there are other anomalies in abnormal photovoltaic distribution areas besides photovoltaic user electricity theft anomalies; This invention has a line loss anomaly perception framework that does not depend on topology and is adapted to photovoltaic access, which improves the accuracy and comprehensiveness of anomaly cause diagnosis, thereby improving the economy and security of the power grid.
[0017] (2) This invention combines adaptive ensemble learning algorithm, dual-state hidden Markov model and electricity theft period optimization, which can effectively solve problems such as imbalance of electricity theft samples and identification of photovoltaic electricity theft users, and accurately lock intermittent electricity theft behavior. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating the present invention's method for detecting abnormal line losses in low-voltage photovoltaic distribution areas, taking into account intermittent electricity theft.
[0019] Figure 2 This is a flowchart of the photovoltaic electricity theft identification method of the present invention.
[0020] Figure 3 This is a visualization of the feature vectors of all transformer areas in Embodiment 2 of the present invention.
[0021] Figure 4 This is a graph showing the abnormal scoring results of key operating indicators when analyzing the causes of abnormalities in all transformer areas in Embodiment 2 of the present invention.
[0022] Figure 5 This is the photovoltaic electricity theft detection result in Embodiment 2 of the present invention.
[0023] Figure 6 This is a visualization of the characteristic vector of the transformer area after correcting the line loss rate and photovoltaic penetration rate in Embodiment 2 of the present invention.
[0024] Figure 7 This is a graph showing the abnormal scoring results of key operating indicators after correcting the line loss rate and photovoltaic penetration rate in Embodiment 2 of the present invention. Detailed Implementation
[0025] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be used to limit the scope of protection of the present invention. It should be noted that the term "comprising" and any variations thereof in the specification, claims and the above-mentioned drawings of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or devices.
[0026] Example 1 like Figure 1 As shown, a method for detecting abnormal line losses in low-voltage photovoltaic distribution areas considering intermittent electricity theft includes: Step S1: Obtain the key operating indicators for each photovoltaic low-voltage distribution area.
[0027] The division of distribution transformer areas can refer to existing technologies. Specifically, distribution transformer areas can be classified according to electricity consumption type, power supply radius, number of users, and photovoltaic (PV) grid connection level. Since electricity is used for various purposes, consumer electricity consumption and demand may vary at different times of the day. Therefore, based on the main type of electricity consumption, distribution transformer areas are divided into four categories: residential, industrial and commercial, agricultural production, and others. The power supply radius captures the line length between the substation and the users in the distribution transformer area, characterizing the line loss rate. Generally, urban areas have higher consumer concentration and smaller supply radii than suburban and rural areas. Combining the number of users with the power supply radius can effectively describe the user density of the distribution transformer area, facilitating the clustering of similar distribution transformer areas. Appropriate distributed PV grid connection can shorten the power supply distance for some users, effectively reducing line losses; in this case, the line loss rate in transformer fault areas may not be high. However, excessive PV power generation can also lead to electricity being sent back to the substation, causing higher line losses. The grid connection level of distributed PV is calculated by combining the PV installed capacity of the transformer area and general meter information, using the following formula: In the formula, The level of distributed photovoltaic grid connection in the transformer area; and These are the transformer substations within the time period to be analyzed. The installed capacity of each photovoltaic user and the power supply of the general electricity meter; For the time span of the data, This refers to the number of photovoltaic users within the designated area.
[0028] In this embodiment, the key operating indicators include photovoltaic penetration rate and several secondary indicators, which are one or more of the following: three-phase imbalance rate, load factor, and power factor. Of course, other secondary indicators can also be added based on existing technology.
[0029] (1) Photovoltaic penetration rate ; Photovoltaic penetration rate refers to the percentage of electricity generated by distributed photovoltaic (PV) systems over a certain period, reflecting the extent to which distributed PV systems support the electricity supply of a power distribution area. The calculation formula is as follows: ; In the formula, For the first The amount of electricity supplied by a photovoltaic user per unit of time; The power supply for general-purpose electricity meters during the same period.
[0030] (2) Three-phase imbalance rate ; Excessive three-phase imbalance can lead to reduced equipment operating efficiency, increased heat generation in equipment and lines, and serious damage to the operation of distribution transformers. According to the State Grid Corporation's enterprise standards, the formula for calculating this indicator is as follows:
[0031] In the formula, and These represent the maximum and minimum RMS values of the current in the three stages, respectively.
[0032] (3) Load factor ; This indicator reflects the current load-bearing status of transformers and photovoltaic systems. Under normal circumstances, the inherent losses in lines and equipment result in a high line loss rate when the transformer area is unloaded, while overload will lead to a decrease in equipment operating efficiency, which may increase the line loss rate.
[0033] (4) Power factor ; The power factor is the ratio of active power to apparent power, reflecting the energy utilization efficiency of electrical equipment in a distribution network. Excessive inductive loads or improper equipment configuration can lead to increased reactive power, a decreased power factor, and higher line loss rates in the distribution area.
[0034] Step S2: Calculate the cross-correlation function and maximum information coefficient between each key operating indicator and the line loss rate, and construct the feature vector for each transformer area.
[0035] For a normal transformer substation, the line loss rate can be expressed as: If an electricity user steals electricity within a transformer area, the line loss rate can be expressed by the following equation: In the formula, This refers to electricity stolen by users, determined by their electricity consumption behavior and unaffected by operational indicators. Therefore, in this case, the independent variable of the function increases, and the correlation between various operational indicators and the line loss rate sequence should be lower than the normal range.
[0036] By analyzing the correlation between performance indicators and line loss rates, the characteristic vector of a transformer substation is constructed. MIC (Microcontroller Interpreter Analyzer) can effectively evaluate the linear and nonlinear correlations between two time series, exhibiting high robustness and low computational complexity; therefore, it is used as a basis for implementing correlation analysis of time series.
[0037] Step S21: Assume the time series of key operating indicators and transformer area line loss rate are as follows: and In the formula, Where is the length of the time series. The MI (mutual information) and MIC of the two series can be obtained as follows: ; ; in, and Key performance indicators and line loss rate The cross-correlation function and the maximum information number, Key performance indicators Time series, The length of the time series. Line loss rate Time series, , Key performance indicators and line loss rate The joint probability density, and Key performance indicators and line loss rate marginal probability density, and These represent the number of intervals into which each dimension is divided. , for , , , one of the.
[0038] Step S22: Construct the transformer area feature vector based on all key operational indicators of the transformer area: ;in, Taiwan District eigenvectors, , , and They are respectively the Taiwan area Line loss rate and photovoltaic penetration rate Three-phase imbalance rate Load factor and power factor The maximum number of information items.
[0039] In this application, since time series data of key operating indicators and transformer area line loss rates are required to have equal granularity and length, it is recommended that the data acquisition period for the time series be 15 minutes / 30 minutes / 1 hour, ensuring that the series lengths are equal. Furthermore, Lagrange interpolation is used to supplement missing values in the series.
[0040] Through the above steps, time delay interference is eliminated, feature correlation is accurately quantified, and a feature vector reflecting the actual operating status of the transformer area is constructed. MIC can effectively evaluate the linear and nonlinear correlations between two time series, exhibiting high robustness and low computational complexity; therefore, it is used as a basis for time series correlation analysis.
[0041] Step S3: After summarizing the feature vectors of all transformer areas, input them into the isolated forest algorithm to detect outliers and filter out abnormal transformer areas.
[0042] Step S3 is as follows: Step S31: Summarize the feature vectors of all transformer areas to form a total matrix. , , The number of stations, Taiwan District eigenvectors.
[0043] Step S32: Perform outlier detection using the isolated forest algorithm to obtain anomaly scores for each transformer area. Transformers with anomaly scores exceeding a set threshold are marked as abnormal transformer areas, while the rest are normal transformer areas. Normal transformer areas are not processed.
[0044] The Isolation Forest (IF) algorithm exhibits excellent outlier detection performance. Its principle involves randomly selecting features and split points, causing samples to fall into leaf nodes of a binary tree, forming an isolated tree. Conversely, by constructing multiple isolated trees and observing the path length from the root node to a leaf node, outlier sample identification can be achieved. Typically, compared to normal samples, outliers are more likely to be split into leaf nodes earlier, and their paths are relatively shorter.
[0045] The method for calculating the anomaly score of a transformer area is as follows: ; ; In the formula, It is a sample (The normal score for each station area); yes Average path length in an isolated tree; It is Euler's constant. This is the normalization constant.
[0046] Step S4: By comparing the deviation of key operating indicators of abnormal transformer areas from those of normal transformer areas, identify the causes of abnormality in each abnormal transformer area, including: power consumption abnormality, photovoltaic abnormality, and minor indicator abnormality.
[0047] Specifically, for each abnormal transformer area, the deviation of its key operating indicators from the average key operating indicators of normal transformer areas is calculated and scored. Key operating indicators with scores (abnormal indicator scores) exceeding a set value are categorized into an outlier set. If the outlier set includes all key operating indicators, the cause of the abnormality for the current transformer area is an electricity consumption-side abnormality, and the transformer area with an electricity consumption-side abnormality is output. If the outlier set does not include all key operating indicators, it is determined whether the score of photovoltaic penetration rate is the maximum value. If it is the maximum value, the cause of the abnormality is a photovoltaic abnormality; otherwise, the cause of the abnormality is a secondary indicator abnormality. When a secondary indicator is abnormal, the secondary indicator with the highest abnormal score is the further cause of the abnormality. That is, if the outlier set includes load factor and power factor, and the abnormal score of power factor is higher, the cause of the abnormality is a secondary indicator abnormality, more specifically, a power technology abnormality.
[0048] In this embodiment, the construction of the outlier set is specifically as follows: For each abnormal transformer area, firstly, the mean values of all normal transformer areas for each key operational indicator are calculated. Then, the deviation of each key operational indicator of the abnormal transformer area from the mean of the normal transformer areas is calculated. This deviation can typically be measured using relative deviation or standardized distance. Next, a score is assigned based on the magnitude of the deviation, and the scoring rules can be set according to business needs (e.g., mapped to intervals). If the score of a certain indicator exceeds a preset threshold, it will be included in the outlier set.
[0049] Taking photovoltaic penetration rate as an example, calculate photovoltaic penetration rate. The mean is calculated using the following formula: ;in, The number of normal transformer stations. Normal station area Photovoltaic penetration rate.
[0050] Step S5: Output the results for abnormal distribution areas where the cause of the abnormality is an abnormality on the power consumption side or a minor indicator abnormality; at the same time, identify electricity theft for each abnormal photovoltaic distribution area. If there are no electricity theft users, output the result as photovoltaic technology abnormality as the cause of the abnormality. If there are electricity theft users, correct the line loss rate and photovoltaic penetration rate of the abnormal photovoltaic distribution area, and determine whether the abnormal photovoltaic distribution area has recovered to a normal distribution area. If it has recovered, output the electricity theft users, the electricity theft period, and the cause of the abnormality as photovoltaic user electricity theft. If it has not recovered, return to step S4 to re-identify the cause of the abnormality, and combine the electricity theft identification result with the re-identified cause of the abnormality before outputting it.
[0051] Step S5 is as follows: S51: Output the results for abnormal transformer areas whose abnormality is due to power consumption abnormality or minor indicator abnormality. The specific output format can be set as needed, including but not limited to: the ID of the abnormal transformer area and the abnormality reason of the abnormal transformer area.
[0052] S52: Identify electricity theft in each abnormal photovoltaic distribution area.
[0053] In this embodiment, an adaptive ensemble learning algorithm is used to identify suspicious users in abnormal photovoltaic distribution areas, and a dual-state hidden Markov model is used to independently analyze each suspicious user to identify electricity theft users and their corresponding electricity theft periods.
[0054] like Figure 2 As shown, step S52 specifically includes: S521: Using the Pearson correlation coefficient to screen meteorological indicators that are strongly correlated with photovoltaic power generation, and combining them with the long-term unit capacity power generation of users in the photovoltaic abnormal area, the time-series characteristics of each user in the photovoltaic abnormal area are obtained.
[0055] Photovoltaic anomaly areas are those where the anomaly is caused by photovoltaic (PV) anomalies. Considering the strong coupling between PV output and meteorological conditions, the Pearson correlation coefficient (PCC) is used to screen meteorological indicators (such as temperature and irradiance) that are strongly correlated with PV power generation. These indicators, along with the PV user's power generation per unit capacity, constitute the input characteristics. The PCC can be obtained from the following equation: ; In the formula, and It is the first in the time series Number.
[0056] S522: The hyperparameters of the RUSBoost model are iteratively optimized using the Bayesian optimization algorithm.
[0057] The structure of the RUSBoost model can refer to existing technologies, and the use of Bayesian optimization (BO) algorithm to optimize hyperparameters for iterative optimization can also refer to existing technologies.
[0058] S523: For each user in the photovoltaic abnormal distribution area, input their time-series characteristics into the optimized RUSBoost model to obtain the user's probability of electricity theft at each time point; based on the electricity theft probability sequence, preliminarily screen suspicious users.
[0059] For each user, meteorological time-series data is combined with the user's power generation per unit capacity to obtain the user's time-series characteristics. These characteristics are then input into an optimized RUSBoost model to obtain the user's probability of electricity theft at each time point. The number of users whose probability of electricity theft exceeds a set threshold in each electricity theft probability sequence is obtained. If the number exceeds the set value, the user is considered a suspicious user; the rest are considered normal users.
[0060] If there are no suspicious users, then there are no users stealing electricity. The cause of the abnormality in the photovoltaic distribution area will be attributed to photovoltaic technology abnormality as the output result.
[0061] If suspicious users exist, an observation sequence is constructed using the probability of electricity theft by the suspicious users, and a two-state hidden Markov model is used to analyze each suspicious user and decode the optimal hidden state sequence to identify electricity theft users and several electricity theft periods.
[0062] In this embodiment, decoding to obtain the optimal hidden state sequence specifically includes the following steps: Step a: Construct the observation sequence ,in, express The observed value at time, , For the current user The probability of electricity theft at any given moment.
[0063] Step b: Establish a two-state hidden Markov model and define the set of hidden states. State transition probability matrix and initial state probability , , This indicates normal power usage. This indicates a state of electricity theft.
[0064] State transition probability matrix This describes the user's inertia when switching between normal and electricity theft states. Let... From state Transferred to The probability of: Initial state probability , ,express The probability of being in each state at any given time.
[0065] Step c: Construct the likelihood function and define the observed emission probability. , , Representing state The following observations The probability of; , This indicates the hidden state at the current moment.
[0066] Specifically, ; .
[0067] Step d: Use the Viterbi dynamic programming algorithm to find the optimal hidden state sequence and identify several intermittent periods of electricity theft.
[0068] Optimal Hidden State Sequence Through the above steps, the output state sequence It can automatically filter out short-term misjudgments caused by cloud cover or communication noise (such as individual normal points in a continuous electricity theft process) and accurately locate intermittent electricity theft behavior with temporal continuity.
[0069] In some other embodiments, step S5 further includes: step e: optimizing the optimal hidden state sequence obtained by decoding to obtain the optimized electricity theft period.
[0070] Step S53 specifically involves: for each user's optimal hidden state sequence, using a sliding window to obtain the local anomaly density within each window. The method for calculating the local anomaly density refers to existing techniques. Construct the dynamic local anomaly density threshold for the current user: ;in, This serves as the baseline value for the local anomaly density of the current user. The standard deviation of the historical local anomaly density for the current user. This represents the current user's dynamic local anomaly density threshold. This is the sensitivity coefficient. It sequentially checks whether the local anomaly density within each window exceeds the current user's dynamic local anomaly density threshold. If so, the current window is considered a period of electricity theft; otherwise, the current window is considered a period of normal electricity use.
[0071] S53: After correcting the line loss rate and photovoltaic penetration rate of the abnormal photovoltaic distribution area, determine whether the abnormal photovoltaic distribution area has recovered to a normal distribution area. If it has recovered, output the electricity theft user, the electricity theft period, and the abnormal reason as electricity theft by the photovoltaic user as the result. If it has not recovered, return to step S4 to re-identify the abnormal reason, and output the electricity theft identification result and the abnormal reason identified in the second identification.
[0072] The reasons for abnormalities in secondary identification include: abnormalities on the power consumption side, abnormalities in secondary indicators, or abnormalities in photovoltaic technology.
[0073] The specific formula for correcting the line loss rate of abnormal photovoltaic distribution areas is as follows: ; in, To update the line loss rate, This represents the original line loss value for the current abnormal photovoltaic distribution area. This refers to the number of solar power theft users within the distribution area. For users who steal electricity in the current abnormal photovoltaic distribution area The installed capacity, and For users who steal electricity in the current abnormal photovoltaic distribution area The measured value of power generation per unit capacity and the average value of power generation of normal photovoltaic users. The power supply of a general-purpose electricity meter during the same period. For photovoltaic users in the current abnormal photovoltaic distribution area Electricity generated per unit capacity This represents the total number of photovoltaic users in the current photovoltaic anomaly area.
[0074] Example 2 Here is a specific example of using the method of this invention for line loss anomaly detection: Based on the operational data of a photovoltaic power station in western my country and local meteorological data for the same period, 20 simulated low-voltage photovoltaic distribution areas were constructed. The main electricity consumption type, power supply radius, number of users, and distributed photovoltaic access level of the simulated transformer areas were residential, 0-100 m, 1-100 m, and 0.5-1 m, respectively. The time series of each variable spanned one week, and the data collection interval was 15 minutes. Then, some distribution areas were selected for line loss anomaly transformation, resulting in areas with electricity theft on the consumer side (Region 5), photovoltaic user electricity theft (Region 10), areas with both anomalies (photovoltaic user electricity theft and photovoltaic technology anomalies) (Region 15), and areas with power factor anomalies (Region 20). The results of feature vector extraction for these 20 areas are shown in Table 1 below, and the results were plotted as a four-dimensional visualization, as shown in... Figure 3 As shown.
[0075] Table 1 shows the feature vector extraction results for the 20 given regions.
[0076] from Figure 4 It can be seen that outlier elements exist in the feature vectors of regions 5, 10, 15, and 20. Since the outlier feature vector elements in region 5 include all key indicators, it is determined that there is an anomaly on the electricity consumption side of this distribution area. In the vectors of regions 10 and 15, photovoltaic penetration rate is the highest outlier element, therefore, it is determined that there is anomalies on the photovoltaic customer side in these two regions. Furthermore, it is necessary to detect and re-analyze photovoltaic electricity theft. Additionally, power factor is the element with the highest outlier in the vector of region 20; therefore, the line loss anomaly in this distribution area is diagnosed as a minor indicator anomaly, specifically: power factor anomaly.
[0077] Based on the RUSBoost algorithm, photovoltaic electricity theft detection was performed in regions 10 and 15. Selected meteorological variables included temperature, humidity, and radiation intensity. The sliding window length was set to 6 hours, corresponding to 24 metering points, and the sliding window moved 24 points at a time. The detection results are as follows: Figure 5 As shown, Figure 5 (a) shows the results of the photovoltaic power theft detection in area 10. Figure 5 (b) shows the results of the photovoltaic power theft detection in area 15. The results showed that one photovoltaic user in each of the two regions did indeed commit electricity theft. The amount of electricity stolen by these abnormal users was assessed, and the time series of line loss rates was corrected before the distribution areas were re-analyzed. The results are shown in Table 2. Figure 5 , Figure 6 and Figure 7 .
[0078] Table 2. Corrected eigenvectors for regions 10 and 15
[0079] According to Table 2, Figures 5-7 It can be seen that the feature vector of region 10 is no longer abnormal, while the photovoltaic penetration rate and load factor elements in the vector of region 15 are still outliers, and these are all low-level outliers. Therefore, it is determined that there are still photovoltaic technology anomalies in region 15, that is, there are both photovoltaic technology anomalies and photovoltaic user electricity theft anomalies in region 15.
[0080] This invention is adaptable to distributed photovoltaic scenarios and can achieve highly interpretable anomaly detection and multi-cause diagnosis; at the same time, it can effectively solve problems such as imbalanced electricity theft samples, identification of photovoltaic electricity theft users, and accurate identification of intermittent electricity theft behavior.
[0081] For more detailed information on the above methods, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.
[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Those skilled in the art will clearly understand that the technologies in the embodiments of this invention can be implemented using software and necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments of this invention.
[0083] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.
Claims
1. A method for detecting abnormal line losses in low-voltage photovoltaic distribution areas considering intermittent electricity theft, characterized in that, include: Step S1: Obtain the key operating indicators for each low-voltage photovoltaic distribution area, including photovoltaic penetration rate and several secondary indicators; Step S2: For each transformer substation, calculate the cross-correlation function and maximum information coefficient between each key operating indicator and the line loss rate, and construct the feature vector for each transformer substation; Step S3: After summarizing the feature vectors of all transformer areas, input them into the isolated forest algorithm to detect outliers and filter out abnormal transformer areas; Step S4: By comparing the deviation of key operating indicators of abnormal transformer areas from those of normal transformer areas, identify the causes of abnormality in each abnormal transformer area, including: power consumption-side abnormalities, photovoltaic abnormalities, and minor indicator abnormalities. Step S5: Output the results for abnormal distribution areas where the cause of the abnormality is an abnormality on the power consumption side or a minor indicator abnormality; at the same time, identify electricity theft for each abnormal photovoltaic distribution area. If there are no electricity theft users, output the result as photovoltaic technology abnormality as the cause of the abnormality. If there are electricity theft users, correct the line loss rate and photovoltaic penetration rate of the abnormal photovoltaic distribution area, and determine whether the abnormal photovoltaic distribution area has recovered to a normal distribution area. If it has recovered, output the electricity theft users, the electricity theft period, and the cause of the abnormality as photovoltaic user electricity theft. If it has not recovered, return to step S4 to re-identify the cause of the abnormality, and combine the electricity theft identification result with the re-identified cause of the abnormality before outputting it.
2. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, The aforementioned secondary indicators include one or more of the following: three-phase imbalance rate, load factor, and power factor.
3. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, Step S2 specifically involves: calculating the key operating indicators in the current transformer area according to the following formula. Cross-correlation function and maximum information content of time series and line loss rate time series; ； ； in, and Key performance indicators and line loss rate The cross-correlation function and the maximum information number, Key performance indicators Time series, , The length of the time series. Line loss rate Time series, , Key performance indicators and line loss rate The joint probability density, and Key performance indicators and line loss rate marginal probability density, and These represent the number of intervals into which each dimension is divided. ; Based on all key operational indicators of the transformer area, construct the transformer area feature vector: ;in, Taiwan District eigenvectors, , , and They are respectively the Taiwan area Line loss rate and photovoltaic penetration rate Three-phase imbalance rate Load factor and power factor The maximum number of information items.
4. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, Step S3 specifically includes: The feature vectors of all transformer areas are summarized to form a total matrix. , , The number of stations, Taiwan District eigenvectors; Outlier detection is performed using the isolated forest algorithm to obtain anomaly scores for each transformer area. Transformers with anomaly scores exceeding a set threshold are marked as abnormal transformer areas, while the rest are normal transformer areas.
5. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, Step S4 specifically includes: For each abnormal transformer area, the deviation of its key operating indicators from the average key operating indicators of normal transformer areas is calculated and scored. Key operating indicators with scores exceeding the set value are classified into the outlier set. If the outlier set includes all key operating indicators, the abnormality of the current abnormal transformer area is due to an abnormality on the power consumption side. If the outlier set does not include all key operating indicators, it is determined whether the score of photovoltaic penetration rate is the maximum value. If it is the maximum value, the abnormality is due to a photovoltaic abnormality; otherwise, the abnormality is due to a minor indicator abnormality.
6. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, In step S5, electricity theft is identified for each abnormal photovoltaic distribution area. Specifically, an adaptive ensemble learning algorithm is used to identify suspicious users in the abnormal photovoltaic distribution area, and a dual-state hidden Markov model is used to independently analyze each suspicious user to identify the electricity theft user and the corresponding electricity theft period.
7. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 6, characterized in that, The process involves identifying suspicious users in abnormal photovoltaic distribution areas using an adaptive ensemble learning algorithm, and then independently analyzing each suspicious user using a dual-state hidden Markov model to identify electricity theft users and their corresponding theft periods. Specifically: By using the Pearson correlation coefficient to screen meteorological indicators that are strongly correlated with photovoltaic power generation, and combining them with the long-term unit capacity power generation of users in the photovoltaic abnormal area, the time-series characteristics of each user in the photovoltaic abnormal area are obtained. The hyperparameters of the RUSBoost model are iteratively optimized using a Bayesian optimization algorithm. For each user in the abnormal photovoltaic distribution area, their time-series characteristics are input into the optimized RUSBoost model to obtain the user's probability of electricity theft at each time point; based on the electricity theft probability sequence, suspicious users are initially screened. If there are no suspicious users, then there are no users stealing electricity, and the cause of the abnormality in the photovoltaic abnormal area is photovoltaic technology abnormality as the output result; if there are suspicious users, then the probability of electricity theft by suspicious users is used to construct an observation sequence, and a two-state hidden Markov model is used to analyze each suspicious user, decode to obtain the optimal hidden state sequence, so as to identify electricity theft users and several electricity theft periods of electricity theft users.
8. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 7, characterized in that, The method involves constructing an observation sequence based on the probability of electricity theft by suspicious users, and then using a two-state hidden Markov model to analyze each suspicious user and decode the optimal hidden state sequence. Specifically: Constructing observation sequences ,in, express The observed value at time, , For the current user The probability of electricity theft at any given moment; Establish a two-state hidden Markov model and define the set of hidden states. State transition probability matrix and initial state probability , , This indicates normal power usage. Indicates a state of electricity theft; Construct the likelihood function and define the observed emission probability. , , Representing state The following observations The probability of; , Indicates the current hidden state; The Viterbi dynamic programming algorithm is used to find the optimal hidden state sequence and identify several intermittent periods of electricity theft.
9. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 7, characterized in that, Step S5 further includes: optimizing the optimal hidden state sequence obtained from decoding to obtain the optimized electricity theft period, specifically: For each user's optimal hidden state sequence, a sliding window is used to obtain the local anomaly density within each window; Construct the dynamic local anomaly density threshold for the current user: ;in, This serves as the baseline value for the local anomaly density of the current user. The standard deviation of the historical local anomaly density for the current user. This represents the current user's dynamic local anomaly density threshold. This is the sensitivity coefficient; The system sequentially checks whether the local anomaly density within each window is greater than the current user's dynamic local anomaly density threshold. If so, the current window is considered a period of electricity theft; otherwise, the current window is considered a period of normal electricity use.
10. The method for detecting abnormal line losses in photovoltaic low-voltage distribution areas considering intermittent electricity theft according to claim 1, characterized in that, In step S5, the line loss rate of the abnormal photovoltaic distribution area is corrected using the following formula: ； in, To update the line loss rate, This represents the original line loss value for the current abnormal photovoltaic distribution area. This refers to the number of solar power theft users within the distribution area. For users who steal electricity in the current abnormal photovoltaic distribution area The installed capacity, and For users who steal electricity in the current abnormal photovoltaic distribution area The measured power generation per unit capacity and the average power generation of normal photovoltaic users. The power supply of a general-purpose electricity meter during the same period. For photovoltaic users in the current abnormal photovoltaic distribution area Electricity generated per unit capacity This represents the total number of photovoltaic users in the current photovoltaic anomaly area.