Transformer area theoretical equivalent resistance value measuring and calculating method based on multi-dimensional transformer area operation data

Abstract

The invention belongs to the field of power system line loss management, and particularly relates to a transformer area theoretical equivalent resistance value measuring and calculating method based on multi-dimensional transformer area operation data, which comprises the following steps: S1, automatically acquiring multi-dimensional data; s2, cleaning the multi-dimensional data; s3, feature extraction and screening; s4, constructing a line loss standard library; s5, training and optimizing a gradient lifting model; and S6, dynamic measurement and verification of the equivalent resistance. According to the transformer area theoretical equivalent resistance value measuring and calculating method based on the multi-dimensional transformer area operation data, electrical parameters, load characteristics and power grid structure data are automatically collected through the RPA technology, abnormal transformer areas are removed in combination with a GSA-SC clustering algorithm, 21 core characteristics are extracted, and high-precision equivalent resistance measuring and calculating are achieved through a Light GBM gradient lifting model. The method supports a dynamic load scene and a complex topological structure, the model generalization ability is optimized through k-fold cross validation and an early stop method, dynamic result verification is achieved in combination with a two-rate deviation rule, and an intelligent solution is provided for power enterprise line loss treatment.

Description

Technical Field

[0001] The present invention relates to the technical field of power system line loss management, and in particular to a method for calculating theoretical equivalent resistance of an electric power substation based on multi-dimensional electric power substation operation data. Background Art

[0002] All forms of power loss generated by distribution components during operation are considered distribution network losses. In the industry, distribution network losses are also referred to as line losses. Energy losses determined by operating and equipment parameters are referred to as theoretical or technical line losses, while losses caused by illegal electricity use, meter errors, or omissions during manual meter reading are referred to as management line losses. Theoretical and management line losses are collectively referred to as statistical line losses, which are numerically equal to the difference between the power supply and electricity sales as reported by the meter. In actual operation, each power supply company compiles monthly statistics on line losses for key power supply lines and the entire network, creating a statistical line loss report. This report not only serves as a basis for assessing system operation but also allows for the identification of abnormal power usage through comparison.

[0003] At present, in low-voltage distribution networks, accurate calculation of theoretical line loss in substations is the core basis for measuring grid energy efficiency and formulating loss reduction strategies. However, existing technologies have the following defects: single data dimension: traditional methods rely only on a few parameters such as current and voltage, ignoring key factors such as grid structure (number of nodes, cable ratio) and load characteristics (load rate, three-phase imbalance), resulting in large errors in equivalent resistance calculation; insufficient model adaptability: static formulas cannot adapt to complex substation topologies (such as star and fishbone types) and dynamic load changes, and have poor ability to fit nonlinear relationships; abnormal data interference: there is a lack of automated screening mechanism for abnormal substations such as power theft and equipment failure, and noisy data can easily cause the calculation results to deviate from the true value. In view of this, we propose a method for calculating the theoretical equivalent resistance of substations based on multi-dimensional substation operation data. Summary of the Invention

[0004] The main purpose of the present invention is to provide a method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data, which can solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned purpose, the present invention proposes a method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data, comprising the following steps:

[0006] S1. Automated multi-dimensional data collection: Automated collection of substation electrical parameters, load characteristics, grid structure, and operating status data through RPA technology;

[0007] S2. Multi-dimensional data cleaning: Abnormal data with actual line loss rate ≤ 1% are eliminated. The GSA-SC algorithm is used to calculate the proportion α of line loss sequence exceeding the standard range (0% to 8%). When α is greater than 50%, it is determined to be a long-term abnormal area and eliminated.

[0008] S3, Feature extraction and screening, based on the average current method theory to extract multiple core features, using the Spearman correlation coefficient to screen features with |SCC|>0.3;

[0009] S4. Line loss standard library construction: use the GSA-SC fusion algorithm to determine the optimal clustering number K, and use the binary K-means++ algorithm to construct the line loss standard library;

[0010] S5, gradient boosting model training and optimization, using the LightGBM framework to train the equivalent resistance estimation model, combining k=4 fold cross validation and early stopping to optimize model performance;

[0011] S6. Dynamic measurement and verification of equivalent resistance: input pre-processed data in real time, output theoretical equivalent resistance value, and dynamically verify the deviation range according to the power supply Q.

[0012] Preferably, the multi-dimensional data includes three-phase current, voltage, and active / reactive power, and the sampling frequency is ≤5 minutes.

[0013] Preferably, the determination formula for the long-term abnormal station area in step S2 is:

[0014]

[0015] Wherein, k is the number of data points in the line loss sequence that exceeds the range of 0% to 8%, and N is the total number of data points.

[0016] Preferably, the method for determining the optimal number of clusters K in step S4 is:

[0017] Calculate the gap value A(K) = E[lnWr(K)] - lnW(K)

[0018] The K value was selected when A(K) first increased significantly and the rate of change was <5%.

[0019] Preferably, in step S5, the hyperparameters of the LightGBM model are: num_leaves=31, max_depth=5, and learning_rate=0.05.

[0020] Preferably, in step S6, the dynamic rule of the deviation range is:

[0021] When Q>800kWh, the allowable deviation range is [-0.5%, 2%];

[0022] When 200kWh≤Q≤800kWh, the allowable deviation range is [-0.5%, 2.5%].

[0023] Preferably, the grid structure data includes the number of nodes, cable / overhead line length ratio, and maximum power supply radius.

[0024] Preferably, in step S3, the calculation formula for the data proportion with power factor > 0.9 is:

[0025]

[0026] Preferably, the triggering condition for the early stopping method in step S5 is that the MSE of the validation set does not decrease for 10 consecutive rounds, and the upper limit of the number of early stopping rounds is 15 rounds.

[0027] Preferably, the method is deployed in the power data middle platform, interacts with the metering system and load monitoring platform in real time through the API interface, and supports the automatic dispatch of abnormal warning work orders.

[0028] The present invention provides a method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data. It has the following beneficial effects:

[0029] (1) This method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data integrates 20+ dimensional data such as electrical parameters (such as three-phase current and voltage), load characteristics (average load rate and power factor), and grid structure (number of nodes and cable ratio), and combines the LightGBM gradient boosting algorithm to efficiently fit nonlinear relationships. The model test set has a mean absolute error (MAE) of ≤0.78Ω, a mean square error (MSE) of 1.35, and a determination coefficient R 2 =0.52, which is significantly improved compared with the traditional equivalent resistance calculation method.

[0030] (2) This method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data is based on the gradient boosting algorithm of LightGBM and supports dynamic load changes and various substation topologies (star, fishbone, and hybrid). Through feature importance analysis, key features such as the number of nodes (contribution 18%), cable length (contribution 15%), and three-phase imbalance rate (contribution 12%) play a dominant role in model prediction. The model was verified on a data set of 636 substations. The fitting error of the fishbone substation (MAE = 0.81Ω) and the star substation (MAE = 0.76Ω) differed by less than 6%, and the model was adaptive to complex scenarios.

[0031] (3) This method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data automatically identifies and eliminates long-term abnormal substations (abnormal line loss ratio α>50%) through the GSA-SC fusion algorithm. Combined with the early stopping method (training is terminated if the MSE of the validation set does not decrease for 10 consecutive rounds), it effectively reduces the interference of noisy data. Experiments show that in a test set containing 15% abnormal data, the error rate of this method is only 3.2%, which is 74% lower than the traditional association algorithm (error rate of 12.5%). The model accurately eliminates abnormal data through cluster analysis to avoid distortion of the measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0033] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0034] Figure 2 This is a schematic diagram of part of the process of the present invention Figure 1 ;

[0035] Figure 3 This is a schematic diagram of part of the process of the present invention Figure 2 ;

[0036] Figure 4 This is a schematic diagram of part of the process of the present invention Figure 3 ;

[0037] Figure 5 This is a schematic diagram of part of the process of the present invention Figure 4 ;

[0038] Figure 6 This is a schematic diagram of part of the process of the present invention Figure 5 ;

[0039] Figure 7 This is a schematic diagram of part of the process of the present invention Figure 6 .

[0040] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] See also Figure 1-Figure 7 The present invention proposes a method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data, comprising the following steps:

[0043] S1, automatic collection of multi-dimensional data;

[0044] In step S1, the RPA robot automatically collects the following multi-source data: electrical parameters: three-phase current, voltage, active power, and reactive power (sampling frequency ≤ 5 minutes); load characteristics: average load factor (ratio of the average load throughout the day to the rated capacity), maximum load factor (ratio of peak load to the rated capacity), load factor (percentage of low-load periods) < 20%, and power factor > 0.9; grid structure: number of nodes, number of outgoing cables, total line length, cable / overhead line length ratio, and maximum power supply radius; operating status: gateway metering parameters, transformer capacity, and reactive compensation equipment status. The data acquisition system interacts with the metering system and load monitoring platform in real time via an API interface. Distributed storage technology is used to structure the data by substation and time dimension and store it in the power data center to ensure data integrity and traceability. Missing data is filled using linear interpolation (e.g., weighted average of the current values ​​before and after each moment), and voltage data is normalized to per-unit value (using 230V as the reference value) to eliminate dimensionality effects.

[0045] S2, multidimensional data cleaning;

[0046] In step S2, abnormal data is removed: data with an actual line loss rate of ≤1% is filtered out. This data is usually caused by metering errors or electricity theft and does not reflect normal operating conditions. For example, a substation experienced a line loss rate of 0.8% for three consecutive days. On-site verification confirmed that the cause was loose meter wiring, and the data was removed to avoid noise interference. For identification of chronic abnormal substations, the GSA-SC (Gap Statistics-Contour Coefficient) fusion algorithm is used to calculate the abnormal proportion α of the line loss sequence that exceeds the standard range (0%-8%). When α is greater than 50%, it is determined to be a chronic abnormal substation (such as equipment failure or chronic electricity theft) and is removed. For example, 216 of the 300 line-loss data points in a certain substation exceeded the standard range (α = 72%), triggering a rejection mechanism to prevent abnormal data from contaminating the training set. The GSA-SC algorithm combines the advantages of gap statistics (GS) and silhouette coefficient (SC). GS determines the optimal number of clusters by comparing the natural logarithm of the cluster dispersion with the reference distribution, avoiding the subjectivity of the traditional elbow rule. SC measures the similarity between the sample and the cluster to which it belongs. The closer the value is to 1, the better the clustering effect. Combining the two can more accurately identify abnormal substations. Experiments show that this method improves the accuracy of abnormal data recognition by 20% compared with a single algorithm.

[0047] S3, feature extraction and screening;

[0048] In step S3, based on the average current method theory, multiple core features are extracted. According to load characteristics, the average load rate, maximum load rate, and low load ratio are classified; according to voltage characteristics, the maximum / minimum transformer output voltage is classified; according to grid structure, the number of nodes, cable length, and overhead line length are classified; according to imbalance, the three-phase maximum imbalance rate and high imbalance ratio are classified. The Spearman correlation coefficient (SCC) is then used to calculate the correlation between the features and the equivalent resistance, and features with |SCC|>0.3 are retained. For example, the number of nodes is positively correlated with the equivalent resistance (SCC=0.43) because more nodes mean more line branches and a greater total impedance; the cable length is negatively correlated with the equivalent resistance (SCC=-0.39) because the unit resistance of the cable is lower than that of the overhead line. The higher the length ratio, the lower the total resistance. After screening, 15-21 core features are finally included, and redundant parameters (such as ambient temperature with a correlation of less than 0.2 with the equivalent resistance) are eliminated.

[0049] S4. Construction of line loss standard library;

[0050] In step S4, the clustering dispersion W(K) and its natural logarithm under different K values ​​are first calculated to determine the optimal cluster number K. The binary K-means++ algorithm is used to divide the substation into normal clusters and abnormal clusters. The normal cluster with the largest sample size is selected, and the standard library interval is constructed with the minimum line loss value (ymin) and the maximum line loss value (ymax) to complete the construction of the line loss standard library. For example, after clustering a batch of substations, the normal cluster line loss interval is [2.1%, 7.3%]. Line loss values ​​outside this interval are judged to be abnormal. In addition, the standard library is updated regularly (such as once a month) to adapt to changes in the power grid structure (such as new users).

[0051] S5, gradient boosting model training and optimization;

[0052] In step S5, the model architecture builds an equivalent resistance estimation model based on the LightGBM framework. The input is the filtered 15-21 features, and the output is the equivalent resistance value. The model is optimized using k-fold cross-validation: the dataset is divided into k = 4 parts, and the training and validation cycles are repeated, and the average results are taken to improve generalization ability. The early stopping method terminates training when the mean square error (MSE) of the validation set does not decrease for 10 consecutive rounds. Hyperparameter tuning: the optimal parameter combination (num_leaves = 31, max_depth = 5, learning_rate = 0) is determined through grid search. 05), where num_leaves = 31 (controls tree complexity to avoid overfitting); max_depth = 5 (limits tree depth to improve generalization ability); learning_rate = 0.05 (balances training speed and accuracy, a smaller learning rate can reduce gradient oscillation). Tested on a dataset of 636 stations, the model test set has a mean absolute error (MAE) of ≤0.78Ω, a mean square error (MSE) of 1.35, and a coefficient of determination R2 of 0.52, which is 40% more accurate than the traditional average current method (MAE = 1.2Ω, R2 = 0.35).

[0053] S6. Dynamic measurement and verification of equivalent resistance.

[0054] In step S6, the model outputs the theoretical equivalent resistance value Req by inputting preprocessed substation data. This performs real-time measurement and dynamically adjusts the tolerance range based on the power consumption Q. For power consumption Q (kWh), when Q < 200, the tolerance range is [-0.5% to 3%] (low power consumption scenarios have slightly higher tolerance); when 200 ≤ Q ≤ 800, the tolerance range is [-0.5% to 2.5%]; and when Q > 800, the tolerance range is [-0.5% to 2%] (high power consumption scenarios require even higher accuracy). If Req exceeds the tolerance range, an alert is triggered and a review work order containing the substation number, the time of the anomaly, and the deviation value is automatically generated and assigned to the operations and maintenance personnel. For example, in a fishbone substation with Q = 1500 kWh, a 2.3% deviation in Req triggered an alert. On-site verification revealed incorrect wiring of the customer's meter, which, after correction, reduced the deviation to 0.5%.

[0055] In an embodiment of the present invention, the multi-dimensional data includes three-phase current, voltage, and active / reactive energy, and the sampling frequency is ≤ 5 minutes. In step S2, the determination formula for long-term abnormal substations is: Wherein, k is the number of data points in the line loss sequence that exceeds the range of 0% to 8%, and N is the total number of data points.

[0056] Furthermore, in step S4, the method for determining the optimal number of clusters K is: calculating the gap value A(K) = E[lnWr(K)] - lnW(K), and selecting the K value when A(K) increases significantly for the first time and the rate of change is <5%.

[0057] Furthermore, in step S5, the hyperparameters of the LightGBM model are: num_leaves = 31, max_depth = 5, learning_rate = 0.05, and in step S6, the dynamic rule of the deviation range is: when Q > 800 kWh, the allowable deviation range is [-0.5%, 2%]; when 200 kWh ≤ Q ≤ 800 kWh, the allowable deviation range is [-0.5%, 2.5%].

[0058] Furthermore, the grid structure data includes the number of nodes, the cable / overhead line length ratio, and the maximum power supply radius. In step S3, the calculation formula for the proportion of data with a power factor greater than 0.9 is: In step S5, the trigger condition for the early stopping method is that the validation set MSE does not decrease for 10 consecutive rounds, and the upper limit of the early stopping round is 15 rounds.

[0059] Furthermore, the method is deployed in the power data middle platform, interacting with the metering system and load monitoring platform in real time through the API interface, supporting the automatic dispatch of abnormal warning work orders.

[0060] Example 1: Calculation of equivalent resistance of a star-shaped substation

[0061] Step S1: Automated collection of multidimensional data

[0062] Data source: The following data is collected automatically through RPA robots:

[0063] Electrical parameters: A, B, C three-phase current (sampling interval 5 minutes), voltage (220V-235V), daily active power (720kWh), reactive power (150kvar);

[0064] Load characteristics: average load rate 35%, maximum load rate 85%, load rate < 20% data account for 12%, power factor > 0.9 data account for 78%;

[0065] Grid structure: 40 nodes, 8 outgoing cables, 650m cable length, 320m overhead line length, 55% cable ratio, and a maximum power supply radius of 500m.

[0066] Operating status: Transformer capacity 400kVA, reactive power compensation equipment operation rate 95%.

[0067] Data storage: Data is automatically stored in the power data center through RPA to form a structured data set (total data volume 300 items).

[0068] Step S2: Multidimensional data cleaning

[0069] Abnormal data removal:

[0070] Rule: Data with actual line loss rate ≤ 1% (e.g. line loss rate 0.8%) are directly filtered, and a total of 18 records are eliminated;

[0071] Identification of long-term abnormal substations: The abnormal proportion of line loss sequences exceeding the standard range (0%-8%) is calculated to be α=72% (k=216, N=300), and they are determined to be long-term abnormal substations and eliminated.

[0072] Preprocessing: Missing current data are filled using linear interpolation, and voltage data are normalized to per-unit value (reference value 230V).

[0073] Step S3: Feature extraction and screening

[0074] Feature extraction: 21 features are extracted based on the average current method, including:

[0075] Load characteristics: average load rate, maximum load rate, low load ratio;

[0076] Voltage characteristics: The maximum value of the transformer output voltage is 235V and the minimum value is 220V;

[0077] Grid structure: number of nodes, cable length, overhead line length;

[0078] Unbalance: The maximum three-phase imbalance rate is 15%, and the imbalance rate > 25% accounts for 5%.

[0079] Feature screening:

[0080] Correlation analysis: The Spearman correlation coefficient (SCC) was used for calculation. 16 features, including the number of nodes (SCC = 0.43) and cable length (SCC = 0.39), satisfied |SCC| > 0.3 and were included in the model input.

[0081] Step S4: Line loss standard library construction

[0082] Clustering:

[0083] Determination of the optimal number of clusters: The gap value A(K) = E[lnWr(K)] - lnW(K) is calculated by the GSA-SC fusion algorithm. When K = 3, A(K) = 2.15 (significantly increased and the change rate is <5%), and the optimal number of clusters K = 3 is determined;

[0084] Binary K-means++ clustering: The station area is divided into three clusters: normal cluster (sample proportion 82%), abnormal cluster A (12%), and abnormal cluster B (6%).

[0085] Standard library generation: Select the minimum line loss value ymin = 2.1% and the maximum line loss value ymax = 7.3% of the normal cluster to construct the standard library interval [2.1%, 7.3%].

[0086] Step S5: Gradient boosting model training and optimization

[0087] Model architecture: The equivalent resistance estimation model is built based on the LightGBM framework, with 16 features as input and equivalent resistance value as output.

[0088] Parameter configuration and optimization:

[0089] Hyperparameters: num_leaves = 31 (control tree complexity), max_depth = 5 (prevent overfitting), learning_rate = 0.05 (balance training speed and accuracy);

[0090] K-fold cross validation: the data was divided into k = 4 parts, and the average MSE after cyclic training was 1.35;

[0091] Early stopping method: When the validation set MSE does not decrease for 10 consecutive rounds (for example, MSE = 1.33 → 1.34 in the 15th round), terminate the training.

[0092] Model evaluation: Test set determination coefficient R 2 =0.52, mean absolute error MAE=0.78Ω.

[0093] Step S6: Dynamic calculation and verification of equivalent resistance

[0094] Real-time measurement: Input real-time data (including three-phase current, number of nodes, etc.), and the model output Req = 0.0268Ω;

[0095] Dynamic verification:

[0096] Power supply Q = 720kWh. According to the regulations, the allowable deviation range is [-0.5%, 2.5%];

[0097] Calculation of actual deviation: Compared with the field measured value of 0.0265Ω, the deviation = +0.0013Ω (0.48%), which is within the allowable range, and the judgment result is valid.

[0098] Example 2: Abnormal warning for a fishbone substation

[0099] Step S1: Automated collection of multidimensional data

[0100] Data source:

[0101] Electrical parameters: A, B, C three-phase current (sampling interval 5 minutes), voltage (220V-230V), daily active power 1500kWh, reactive power 320kvar;

[0102] Grid structure: 25 nodes, fishbone topology parameters (clear order of branch nodes), cable length 450m, overhead line length 1050m, and cable proportion 30%;

[0103] Operating status: Transformer capacity 630kVA, reactive power compensation equipment operation rate 88%.

[0104] Step S2: Multidimensional data cleaning

[0105] Abnormal data removal:

[0106] Rule: Filter data with actual line loss rate ≤ 1% (e.g. line loss rate 0.9%), and remove 5 records in total;

[0107] Identification of long-term abnormal substations: The calculated abnormal proportion of line loss sequence is α=28% (k=84, N=300), and the rejection condition is not triggered.

[0108] Step S3: Feature extraction and screening

[0109] Feature extraction: 18 features are extracted, including the maximum three-phase unbalance rate of 28%, node cumulative current (a parameter unique to the fishbone type), and a cable length of 450m.

[0110] Feature screening: Through SCC calculation, the correlation between user A's power consumption and line loss is SCC = 0.62 (strong correlation), and is included in the anomaly analysis.

[0111] Step S4: Line loss standard library construction

[0112] Clustering: The GSA-SC algorithm is used to determine K = 2 (gap value A(K) = 1.98), and the normal cluster line loss range is [2.5%, 7.8%].

[0113] Step S5: Gradient boosting model training and optimization

[0114] Model parameters: num_leaves = 31, max_depth = 5, learning_rate = 0.05, MSE of the test set after training = 1.28, R 2 =0.55.

[0115] Step S6: Dynamic calculation and verification of equivalent resistance

[0116] Real-time measurement: input pre-processed data, model output Req = 0.035Ω;

[0117] Dynamic verification:

[0118] Power supply Q = 1500kWh, with an allowable deviation range of [-0.5%, 2%];

[0119] Actual deviation: Compared with the theoretical reasonable value of 0.0342Ω, the deviation = +0.0008Ω (2.3%), which exceeds the allowable range;

[0120] Early warning and handling:

[0121] The system automatically triggers an early warning, generates a work order, and dispatches it to the operation and maintenance personnel;

[0122] On-site inspection revealed that user A's meter wiring was incorrect (electricity theft). After the repair, the line loss rate returned to 5.2% and the deviation dropped to 0.5%.

[0123] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data, characterized in that: The following steps are involved: S1, automatic collection of multi-dimensional data; S2, multidimensional data cleaning; S3, feature extraction and screening; S4. Construction of line loss standard library; S5, gradient boosting model training and optimization; S6. Dynamic measurement and verification of equivalent resistance.

2. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: The multi-dimensional data includes three-phase current, voltage, and active / reactive power, with a sampling frequency of ≤5 minutes.

3. The method for calculating theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: The determination formula for the long-term abnormal station area in step S2 is: Wherein, k is the number of data points in the line loss sequence that exceeds the range of 0% to 8%, and N is the total number of data points.

4. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: In step S4, the method for determining the optimal number of clusters K is: Calculate the gap value A(K) = E[lnWr(K)] - lnW(K) The K value is selected when A(K) increases significantly for the first time and the rate of change is less than 5%.

5. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: In step S5, the hyperparameters of the LightGBM model are: num_leaves=31, max_depth=5, and learning_rate=0.

05.

6. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: In step S6, the deviation range dynamic rule is: When Q>800kWh, the allowable deviation range is [-0.5%, 2%]; When 200kWh≤Q≤800kWh, the allowable deviation range is [-0.5%, 2.5%].

7. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: The grid structure data includes the number of nodes, cable / overhead line length ratio, and maximum power supply radius.

8. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: In step S3, the calculation formula for the data ratio with power factor > 0.9 is:

9. The method for calculating theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: In step S5, the trigger condition for the early stopping method is that the validation set MSE does not decrease for 10 consecutive rounds, and the upper limit of the early stopping round is 15 rounds.

10. The method for calculating the theoretical equivalent resistance of a substation based on multi-dimensional substation operation data according to claim 1, characterized in that: The method is deployed in the power data middle platform, interacts with the metering system and load monitoring platform in real time through the API interface, and supports the automatic dispatch of abnormal warning work orders.

Images