A method for identifying household-transformer relationships based on the fusion of electricity and voltage information

Abstract

A method for identifying the relationship between a transformer and its user based on the fusion of electricity and voltage information, which is based on hot-restart stochastic gradient descent and a type of support vector machine information fusion, includes the following steps: (1) constructing a parameterized model of electricity consumption in the transformer area; (2) using stochastic gradient iteration and learning rate self-adjustment method to find the global optimal solution for preliminary identification; (3) using a sliding time window to make multiple judgments to obtain preliminary transformer identification results; (4) based on the preliminary identification results, using voltage data of users with normal transformer-user relationships to form training samples; (5) using a type of support vector machine to learn the voltage characteristics of users with normal transformer-user relationships in the transformer area, constructing a transformer-user relationship identification model, and realizing the correct identification of transformer-user relationships in low-voltage transformer areas. This invention can effectively identify users with incorrect transformer-user relationship records, with a significant improvement in recall and precision, saving investigation costs, and realizing a comprehensive judgment method based on the fusion of different feature information of the transformer area. Compared with the single feature information judgment method, the identification results are more stable and reliable.

Description

Technical Field

[0001] This invention relates to the field of power distribution, specifically to a method for identifying the relationship between a household and a transformer, which is based on the fusion of power and voltage information. Background Technology

[0002] With the rapid development of the social economy and the improvement of people's living standards, electricity consumption has increased, making the management of line losses in low-voltage distribution areas increasingly complex. To accelerate the construction of smart grids, fully utilize distribution-side data resources, and improve power supply service quality and customer satisfaction, traditional extensive management methods are no longer adequate for the needs of the times. Lean management of distribution areas has become a new management trend. However, for a long time, the addition and replacement of distribution transformers during line renovations and municipal engineering projects have led to highly complex cross-line issues between distribution areas and errors in the system files regarding customer-transformer relationships, severely hindering the lean management of distribution areas.

[0003] To address the issue of incorrect transformer-household relationship records in the aforementioned distribution areas, power companies primarily rely on manual checks. Traditional methods include on-site power outage verification and carrier communication verification. On-site power outage verification involves briefly shutting down each transformer, then using a handheld device to read the power outage records from the meters, and finally identifying the affiliation of each meter based on the outage duration. However, this method requires power outage verification, disrupting normal electricity usage, user experience, and power supply reliability. Carrier communication verification is currently widely used. This method uses a carrier communication terminal located on the transformer side and a handheld receiver on the user's meter side to conduct carrier communication. By analyzing the characteristics of the communication messages, it determines the distribution area to which the user belongs. However, this method is limited to distribution areas where the user's meter supports the carrier communication protocol, has poor robustness, is susceptible to interference, and has an accuracy rate of only 70%-80%. It cannot be used in distribution areas with significant load fluctuations. Therefore, researching a reliable and automated method for identifying transformer-household relationships is crucial. Summary of the Invention

[0004] This invention addresses the problems existing in the prior art by providing a method for identifying the relationship between transformer substations and households based on the fusion of power and voltage information. An objective function is constructed based on power data, and the objective function is solved using a hot-restart stochastic gradient descent algorithm to obtain preliminary identification results. Based on voltage information, a support vector machine is constructed to achieve accurate identification of the relationship between transformer substations and households in low-voltage transformer substations.

[0005] The technical solution of this invention is: a method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information, which is based on hot-restart stochastic gradient descent and a type of support vector machine information fusion, including the following steps:

[0006] Includes the following steps:

[0007] (1) Construct a parameterized model of electricity consumption in the distribution area based on the conservation of electricity consumption in the total meter of the distribution area and the electricity consumption of users;

[0008] (2) The process of solving the global optimal solution of the parameterized model of electricity consumption in the transformer area by using stochastic gradient iteration and learning rate self-adjustment method, and the preliminary identification of the affiliation between users and transformer areas.

[0009] (3) Using a sliding time window-based multiple judgment method, the historical data of the transformer area is divided into multiple time windows with different start dates. Based on steps (1) and (2), multiple solutions are performed, and the results of multiple solutions are combined to obtain the household transformer affiliation relationship, i.e., the preliminary household transformer identification result.

[0010] (4) Based on the preliminary identification results, training samples were constructed from the voltage data of users with normal household-transformer relationships;

[0011] (5) Using a support vector machine method, the voltage characteristics of normal users in the transformer substation are learned, and a transformer substation relationship identification model is constructed. The transformer substation relationship identification is performed on all users in the current files of the transformer substation, and users with incorrect transformer substation relationship files are identified, so as to achieve correct identification of transformer substation relationship.

[0012] This invention is based on the fusion of power and voltage information, and uses the characteristic information of power and voltage to identify the transformer in the user area, fully exploring the value of the data collected on the user side. Moreover, it can take into account the characteristics of each transformer area for a large number of low-voltage transformer areas, and has high applicability and effectiveness.

[0013] The present invention has positive effects: Compared with the prior art, the method of the present invention can use the actual operation data of existing electricity information systems, marketing systems, etc. as a basis, and use the Stochastic Gradient Descent with Warm Restarts (SGDR) and One-Class Support Vector Machine (OC-SVM) to establish a household-transformer relationship identification model by learning from the historical data of users in the transformer area. This model can then calculate the users with incorrect household-transformer relationships in the current files of a certain transformer area, thus clarifying the direction of on-site verification at the system level.

[0014] The beneficial effects of the present invention also include:

[0015] (1) In this invention, a multiple judgment method based on sliding time window is adopted to divide the historical data into multiple time windows. Each time, different start dates are selected for multiple solutions to calculate the users with incorrect household-transformation relationship in different periods. Then, the union of the incorrect users is taken to obtain the final household-transformation relationship identification result, thus avoiding the problem of verification omissions caused by differences in user electricity consumption characteristics.

[0016] (2) Based on the method in this invention, the list of electricity users with incorrect household-transformer relationships can be accurately identified, and the verification direction can be given. Then, based on the actual engineering maintenance personnel's door-to-door inspection, the specific incorrect users can be identified, which effectively reduces the workload of on-site verification, improves the efficiency of on-site verification, and saves verification costs. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process for identifying household change relationships based on hot-restart stochastic gradient descent and a type of support vector machine information fusion in this invention.

[0018] Figure 2 This is a learning rate self-adjustment graph based on a cosine annealing hot restart mechanism in this invention.

[0019] Figure 3 This is a schematic diagram of the multiple judgment method of the sliding time window in a household change relationship identification method based on hot restart stochastic gradient descent and a type of support vector machine information fusion in this invention. Detailed Implementation

[0020] This invention proposes a method for identifying the relationship between a user and a transformer station based on the fusion of electricity and voltage information. The method integrates electricity and voltage information for identification, employing a sliding time window-based multiple-judgment approach. A fixed number of time windows are selected from historical data. Based on electricity information, the Stochastic Gradient Descent (SGDR) method is used to perform a first-stage identification of the user-transformer relationship in the historical data. Abnormal users are removed based on the identification results to reduce the number of abnormal users in the sample. Voltage data of users with normal user-transformer relationships obtained from the first identification are used to construct training samples. A support vector machine is trained to learn the voltage characteristics of normal users. A second-stage identification of user-transformer relationships is then performed on users within the transformer station archives to identify users with incorrect user-transformer relationships within the archives.

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0022] Figure 1 This is a schematic diagram of the household change relationship identification method based on hot-restart stochastic gradient descent and a type of support vector machine information fusion in this invention. Figure 1 As shown, the specific implementation of the present invention is described below.

[0023] Electricity consumption data is a cumulative quantity, reflecting the logical summation relationship between households and transformers. Based on the law of conservation of energy, a relationship model is established between the electricity consumption of the gateway meter and the electricity consumption of each user, as follows.

[0024]

[0025] In the formula, n represents the number of users to be input in the station area, and y tThis represents the electricity consumption at the gate meter in hour t (t = 1, 2, ..., T, where T represents the total number of hours). This represents the electricity consumption of the i-th user in hour t (i = 1, 2, ..., n), a i This indicates whether the user profile belongs to the specified area. i When a is 1, it indicates that the i-th user file belongs to this station area; when a... i When the value is 0, it means that the user file does not belong to this distribution area, and μ represents the line loss of this distribution area.

[0026] Within the same distribution area, bus loss and total power consumption at the gateway are positively correlated. The total power consumption is the sum of the power consumption of each user. Therefore, the bus loss can be allocated to each user during calculation, as shown in equation (2):

[0027]

[0028] In the formula, the bus line loss consists of the line losses of n users, and the line loss of each user is related to their power consumption, b i The line loss coefficient represents the proportion of line loss in the total electricity consumption for that user.

[0029] In reality, residential electricity consumption does not fluctuate significantly, and the line loss coefficient b i It is approximately a fixed value, while a i and b i These are all coefficients for electricity consumption, which can be represented by c. i =a i +b i Simplifying the electricity consumption coefficient, the electricity conservation model (1) can be simplified to equation (3):

[0030]

[0031] From equation (3), the following set of logical equations relating household changes can be established:

[0032] Y = XC (4)

[0033] Y = [y1, y2, y3, ..., y n ]' (5)

[0034] C = [c1, c2, c3, ..., c n (6)

[0035]

[0036] In the formula, Y represents the electricity consumption data of the main meter for the distribution area over T days, C represents the electricity consumption coefficient to be solved, and X represents the electricity consumption of n users in the distribution area over T days. Each row represents the electricity consumption of one user over T days, and each column represents the electricity consumption of each user on day t. The set of logical equations relating users and electricity consumption is the parameterized model of electricity consumption in the distribution area.

[0037] The solution process for the household-variable relationship equations based on the principle of energy conservation is as follows:

[0038] 1) Input the power data and initialize the solution parameters;

[0039] Input the daily electricity consumption data of the transformer master meter of the area to be analyzed and n users for a total of T days. The electricity consumption data of the master meter constitutes a T*1 transformer master meter electricity matrix Y, and the electricity consumption data of the users constitutes a T*n user electricity matrix X.

[0040] Initialize the user power consumption coefficient C as an n*1 vector of all 1s. In the initial state, it is assumed that all users to be analyzed are users with correct user-transformer profiles. Initialize the user-transformer relationship equation set Y=XC.

[0041] 2) The system of equations is solved using stochastic gradient iteration and self-adjusting learning rate methods. This invention utilizes hot-restart stochastic gradient descent to solve Y = XC, with the following solution parameters: the number of iterations per cycle is set to K = 1 * 10^ 5 The number of hot restart iterations is set to R = 3;

[0042] The cost function E of the equation system in the s-th iteration, i.e., the sum of squared errors of the equation system Y = XC, is calculated using the following formula:

[0043]

[0044] In the formula, X t C represents the electricity consumption matrix consisting of n users on day t. s This represents the energy coefficient matrix corresponding to the s-th iteration, where s∈[1,K].

[0045] Find the cost function E with respect to the user's electricity consumption coefficient C. s The gradient δ is derived using the following formula:

[0046]

[0047] X t 'Is X' t For C s The derivative of .

[0048] The coefficients for power consumption are updated according to the current step size and gradient direction as shown in equation (10):

[0049]

[0050] Learning rate adjustment combining hot restart and cosine annealing, such as Figure 2 Japanese style (11):

[0051]

[0052] In the formula, s represents the number of iterations, and η max This represents the upper limit of the learning rate adjustment process, η. min This represents the lower bound of the learning rate adjustment process, T. cur T represents the number of iterations of the learning rate during the interval from the start to the end of each restart. s This indicates the restart cycle.

[0053] Furthermore, to avoid instability in the solution process causing some users' power coefficients to converge too quickly, thus affecting the overall solution of C, C needs to be reduced after each iteration. The reduced value is used as the initial value of C for the next iteration. The reduction formula is:

[0054]

[0055] c s Indicate C s For each element in the set, set the upper threshold to 0.7 and the lower threshold to 0, and compare c. s The system determines the electricity consumption of each user. Users whose consumption is below the lower threshold are considered suspicious due to incorrect user records, and their records do not belong to the relevant transformer area. Users whose consumption is above the upper threshold are considered users of the transformer area. Users in between indicate that the system is trapped in a local optimum, requiring parameter adjustments and iterative solving. The corrected electricity consumption coefficient is used as the initial value for the next run.

[0056] A multiple-judgment method based on a sliding time window is adopted, such as... Figure 3 Historical data is divided into multiple time windows, and different start dates are used each time to solve the parameterized model of electricity consumption in the transformer area multiple times. The union of all solution results is then used to obtain the household transformer affiliation relationship, i.e., the preliminary household transformer identification result.

[0057] Based on the preliminary household transformer identification results obtained using the power model, users with incorrect household transformer relationships in the transformer substation area are deleted, and new users with nearly correct household transformer relationships are obtained. Training samples are constructed using the voltage data of these users with normal household transformer relationships, and a support vector machine OC-SVM is trained to learn the voltage characteristics of users with normal household transformer relationships in the transformer substation area.

[0058] This invention determines the transformer-household relationship based on voltage information. The data used is voltage sampling data from users at 24:00 per day (collected 1 hourly). Utilizing the differences in voltage curves between users in different transformer areas, a single-class support vector machine (OC-SVM) method is used to classify users within a transformer area into those with normal transformer-household relationships and those with incorrect relationships, thus achieving transformer-household relationship identification. When identifying users with incorrect transformer-household relationship records, the number of users with incorrect records should be very small compared to normal users, which constitutes an imbalanced sample data. Therefore, identifying transformer-household relationships falls under the category of imbalanced sample data classification or anomaly user detection. To prevent overfitting, a single-class classification method is used. The single-class support vector machine is an unsupervised anomaly identification method designed to solve the single-class classification problem based on traditional support vector machines. OC-SVM automatically finds a hyperplane that contains the vast majority of the original data; data outside the hyperplane boundary is judged as anomaly data. Therefore, the single-class support vector machine can identify abnormal transformer-household relationships.

[0059] OC-SVM maps the input transformer area voltage data to the corresponding feature space through a kernel function, and uses the data in the feature space to construct a hyperplane and separate it from the origin to the maximum extent. The algorithm uses a decision function f(x) to evaluate which side of the hyperplane in the feature space the test sample data x falls on to determine whether it is a positive or negative sample. A positive sample indicates that the user belongs to the transformer area, and a negative sample indicates that the user does not belong to the transformer area. In order to separate the dataset from the origin, the OC-SVM algorithm is essentially solving a quadratic programming problem of equation (13):

[0060]

[0061] In the formula: ω is the hyperplane normal vector; v∈(0,1] is the preset negative sample ratio, l is the number of samples, vl is a trade-off parameter controlling the upper limit of the outlier ratio and the lower limit of the number of support vectors; ρ is the hyperplane intercept; φ is the nonlinear mapping function; ωφ(x i -ρ = 0 denotes the hyperplane, ||·|| denotes the Euclidean norm; ξ i These are slack variables.

[0062] The decision function can be expressed as equation (14):

[0063] f(x)=sgn((ω·φ(x))-ρ) (14)

[0064] f(x) will be positive for most of the data in the training set, while the support vector type regularization term ||ω|| remains small. The actual trade-off between these two objectives is controlled by v. Using the multiplier α i ,β i ≥0, we introduce the Lagrange multipliers to obtain equation (15):

[0065]

[0066] Taking the partial derivatives with respect to ω, ξ, and ρ respectively and setting them to 0, we obtain equation (16):

[0067]

[0068] In the above formula, {x i : i∈[l], α i >0} are support vectors. After kernel expansion, the decision function becomes equation (17):

[0069]

[0070] Substituting equation (16) and the kernel function k into equation (15), equation (13) can be transformed into its dual problem, as shown in equation (18):

[0071]

[0072] In the decision function, ρ can be represented by any corresponding Lagrange multiplier α. i Non-zero sample x j Find:

[0073]

[0074] The decision function is then:

[0075]

[0076] For the decision function f(x), when sample x is identified by the classifier as a user in the transformer area, the value is +1, and when it is identified as an abnormal user, the value is -1, thus realizing the identification of transformer area-household-transformer relationship based on voltage data. Equations (13)-(20) are the basic derivation formulas of a class of support vector machines, which are the basic theories of a class of support vector machine algorithms.

[0077] Considering that some support vector machine (SVM) methods are sensitive to outliers, their training sets should be as free of outlier contamination as possible, or contain a small proportion of outlier data. Furthermore, when using the hot-restart stochastic gradient descent (SGDR) method based on electricity consumption for transformer-household relationship identification, it requires comprehensive judgment across multiple time periods using a sliding time window; however, determining the optimal time window is difficult. As the number of time windows increases, recall and precision become incompatible, and computational time costs increase. Therefore, a transformer-household relationship identification method fusing electricity consumption and voltage information is proposed. A fixed number of time windows are selected empirically. Based on electricity consumption information, the SGDR method is used for the first identification of transformer-household relationships. Based on the identification results, outlier users are removed to reduce the number of outlier users in the sample. The voltage data of normal transformer-household relationships obtained from the first identification are used to construct the training sample. Then, OC-SVM is used for the second identification of transformer-household relationships to obtain the final incorrect transformer-household relationships for the transformer area. Test results show that this method achieves 100% recall and over 96% accuracy, improving the effectiveness of identification. It not only resolves the contradiction between recall and precision when using the SGDR method based on electricity consumption to jointly determine multiple time periods, as well as the increase in time cost, but also reduces the proportion of abnormal data in the OC-SVM training samples and makes full use of data resources.

[0078] The beneficial effects of this invention are that it can effectively identify users with incorrect household registration records, and the recall and precision rates are greatly improved. It can save the cost of manual investigation, realize a comprehensive judgment method based on the fusion of different feature information of the transformer area, and the recognition results are more stable and reliable compared with the single feature information judgment method.

[0079] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information, characterized in that: Based on hot-restart stochastic gradient descent and information fusion of a class of support vector machines, the following steps are included: (1) Construct a parameterized model of electricity consumption in the transformer area based on the conservation of total meter readings and user electricity consumption; (2) The process of solving the global optimal solution of the parameterized model of electricity consumption in the transformer area by using stochastic gradient iteration and learning rate self-adjustment method, and the preliminary identification of the affiliation between users and transformer areas; Solving the parameterized model of electricity consumption in the transformer area based on the principle of energy conservation: 1.1) Input the power data and initialize the solution parameters; Input the daily electricity consumption data of the transformer master meter of the area to be analyzed and n users for a total of T days. The electricity consumption data of the master meter constitutes a T*1 transformer master meter electricity matrix Y, and the electricity consumption data of the users constitutes a T*n user electricity matrix X. The user electricity consumption coefficient C is initialized as an n*1 vector of all 1s, that is, the initial state assumes that all users to be analyzed are users with correct user and transformer files, and the initialized transformer area electricity consumption parameterization model Y=XC is obtained. 1.2) Solve the parameterized model of power consumption in the transformer area using hot-restart stochastic gradient descent, setting the solution parameters as follows: the number of iterations per cycle is set to K = 1 * 10. ^5 The number of hot restart iterations is set to R=3; 1.3) Calculate the cost function E of the equation system in the s-th iteration, i.e., the sum of squared errors of Y=XC. The formula for calculation is: (8) In the formula, y t X represents the electricity consumption on day t of the meter. t Let C represent the electricity consumption matrix consisting of n users on day t. s Let represent the energy coefficient matrix corresponding to the s-th iteration, where s∈[1,K]; Find the cost function E with respect to the user's electricity consumption coefficient C. s The gradient δ is derived using the following formula: δ= (9) The coefficients of power consumption are updated according to the current iteration step size and gradient direction as shown in equation (10): (10) The learning rate for hot restart combined with cosine annealing; After each iteration, C is reduced, and the reduced value is used as the initial value for the next iteration of C. The reduction formula is: (12) c s Indicate C s For the elements in the table, users whose values ​​are below the lower threshold are identified as suspicious users with file errors, while those whose values ​​are above the upper threshold are identified as users of the same area. Users in between indicate that the system is trapped in a local optimum and the parameters need to be adjusted and iterated again. The corrected power consumption coefficient is used as the initial value for the next run. (3) Using a sliding time window-based multiple judgment method, the historical data of the transformer area is divided into multiple time windows with different start dates. Based on steps (1) and (2), multiple solutions are performed, and the results of multiple solutions are combined to obtain the household transformer affiliation relationship, i.e., the preliminary household transformer identification result. (4) Based on the preliminary identification results, the voltage data of users with normal household-transformer relationships are used to form training samples; (5) Using a support vector machine method, the voltage characteristics of normal users in the transformer substation are learned, and a transformer substation relationship identification model is constructed. The transformer substation relationship identification is performed on all users in the current files of the transformer substation, and users with incorrect transformer substation relationship files are identified, so as to achieve correct identification of transformer substation relationship.

2. The method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information as described in claim 1, characterized in that: When constructing a parameterized model of electricity consumption in a distribution area, the operating data of the distribution area includes: the daily electricity consumption and voltage readings of all users in the area, as well as the daily electricity consumption and voltage readings of the power supply transformers in the area. The objective of the parameterized model is to minimize the square of the total electricity consumption of the distribution area minus the sum of the electricity consumption of each user. The calculation formula is as follows: In the formula, y t X represents the electricity consumption on day t, where t = 1, 2, ..., T, and T represents the total number of days. t Let C represent the electricity consumption matrix consisting of n users on day t. s This represents the energy coefficient matrix corresponding to the s-th iteration.

3. The method for identifying the relationship between a household and its transformer based on the fusion of electricity and voltage information as described in claim 2 is characterized in that, in the parameterized model, the electricity consumption data is a cumulative quantity, reflecting the logical summation relationship between household and transformers. Based on the law of conservation of energy, a relationship model between the electricity consumption of the gate meter and the electricity consumption of each user is established, as follows: (1) In the formula, n represents the number of users to be input in the station area, and y t This represents the electricity consumption on day t of the gate meter, x i t This represents the electricity consumption of the i-th user on day t, where i = 1, 2, ..., n, and a i This indicates whether the user profile belongs to the specified area. i When a is 1, it indicates that the i-th user file belongs to this station area; when a... i When the value is 0, it means that the user file does not belong to this distribution area, and μ represents the line loss of this distribution area; Within the same distribution area, bus losses are distributed among users during calculation, as shown in equation (2): (2) In the formula, the bus line loss consists of the line losses of n users, and the line loss of each user is related to their power consumption, b i The line loss coefficient represents the proportion of line loss in the user's electricity consumption. Use c i =a i +b i The simplified electricity consumption coefficient and the electricity conservation model (1) are simplified to equation (3): (3) Based on equation (3), a set of logical equations relating households and transformers is established, which is the parameterized model of electricity consumption in the transformer area, as follows: Y=XC (4) Y = [y1,y2,y3,…,y T (5) C=[c1,c2,c3,…,c n ] (6) (7) In the formula, Y represents the electricity consumption data of the main meter of the distribution area for T days, C represents the electricity consumption coefficient to be solved, and X represents the electricity consumption of n users in the distribution area for T days.

4. The method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information as described in claim 1, characterized in that... The learning rate adjustment for restarting combined with cosine annealing is as shown in equation (11): (11) In the formula, s represents the number of iterations, and η max This represents the upper limit of the learning rate adjustment process, η. min This represents the lower bound of the learning rate adjustment process, T. cur T represents the number of iterations of the learning rate during the interval from the start to the end of each restart. s This indicates the restart cycle.

5. The method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information as described in claim 1, characterized in that: When using the sliding time window-based multiple judgment method, a fixed number of time windows are selected. For the data in each time window, a parameterized model of the power consumption of the transformer area is constructed according to steps (1) and (2). The stochastic gradient descent (SGDR) method is used to identify the relationship between the transformer and the user within the time window. Abnormal users are removed based on the results of the first identification to reduce the number of abnormal users in the training samples. The voltage data of normal users with normal transformer-user relationships are used as training samples to train a support vector machine to learn the voltage characteristics of normal users with normal transformer-user relationships in the transformer area and construct a transformer-user relationship identification model.

6. The method for identifying the relationship between a household and a transformer based on the fusion of electricity and voltage information according to claim 1, characterized in that: After the household-transformation relationship identification model identifies the current household-transformation relationships in the transformer substation area, manual verification is performed.

Images