Table-user relation identification method based on table-user relation identification model

Abstract

The invention relates to a platform user relation identification method based on a platform user relation identification model, and the process of the identification method comprises the following steps: collecting the electricity consumption of a user, processing the error data in the collection process, carrying out the standardization processing of all processed data, and obtaining the electricity consumption of the user; the power utilization information of different users can be compared and analyzed under the same scale; based on the standardized data, extracting electricity consumption information feature values from a time domain and a frequency domain, and constructing feature vectors reflecting electricity consumption habits of users; a random forest algorithm is adopted to establish a station user relation model, training and evaluation are carried out, a plurality of decision trees are constructed and are respectively predicted, and a strong classifier is combined to calculate a final classification result based on the prediction result of each decision tree. According to the method, a final user area classification result is obtained by extracting the user electricity consumption information characteristic value, establishing the user relation identification model, constructing a plurality of decision trees and carrying out independent prediction, and integrating the decision tree prediction values.

Description

Technical Field

[0001] The present invention relates to the technical field of station-user relationship identification, and in particular to a station-user relationship identification method based on a station-user relationship identification model. Background Art

[0002] Identifying the relationship between low-voltage stations and users is a key link in distribution network management. Traditional methods rely on manual inspections, ledger data, and station user identification devices. These methods are not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in inaccurate or incomplete data. With the rapid development of smart grids, the amount of distribution network data has increased dramatically. How to use these massive data to accurately identify and maintain the relationship between stations and users in low-voltage distribution networks has become a key challenge to improve the operating efficiency of distribution networks, reduce operation and maintenance costs, and enhance power supply reliability.

[0003] The invention patent with publication number CN110389267A discloses a method for identifying the relationship between low-voltage substations and households based on data collected by smart energy meters, including: collecting and obtaining the original voltage data of a preset number of gateway smart energy meters and user smart energy meters in the low-voltage substation to be identified for a preset number of days and cleaning them to obtain the valid data of each smart energy meter; according to the valid data obtained for each smart energy meter, obtaining the daily average voltage and standard deviation of the smart energy meter; performing correlation analysis on the average value series and standard deviation series of the user smart energy meter and the gateway smart energy meter respectively; if the analysis results show that the correlation coefficients of the average value and standard deviation of a certain user smart energy meter and a certain gateway smart energy meter are both the maximum, it is determined to be a match. The household matching accuracy of this invention is not high, and the accuracy of household identification needs to be further improved.

[0004] Therefore, providing a method with high recognition accuracy is an urgent problem to be solved. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for identifying household relationships based on a household relationship identification model.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a method for identifying a customer relationship based on a customer relationship identification model is provided, wherein the identification method comprises the following steps:

[0008] S1. Collecting the user's electricity consumption, where the electricity consumption is the user's daily electricity consumption for a continuous period of time;

[0009] S2. Process the erroneous data in the collection process and standardize all the processed data to ensure that the electricity consumption information of different users can be compared and analyzed on the same scale;

[0010] S3, based on the standardized data, extract the characteristic values ​​of electricity consumption information from the time domain and frequency domain, and construct a characteristic vector reflecting the user's electricity consumption habits;

[0011] S4. Use the random forest algorithm to establish a household relationship model and perform training and evaluation. By constructing multiple decision trees and making predictions separately, a strong classifier is combined to calculate the final classification result based on the prediction results of each decision tree.

[0012] As a preferred technical solution, the types of error data to be processed include data anomalies, data missing and invalid data;

[0013] Among them, combined with the user's commissioning date information, users who are not in operation or have not generated electricity are eliminated; based on the user's design capacity, a reasonable threshold range is set for electricity consumption, and electricity consumption data that exceeds the normal operating range is marked, corrected or eliminated; based on the electricity consumption characteristics corresponding to the user type, the electricity consumption data is smoothed.

[0014] As a preferred technical solution, the missing data is processed by an interpolation method or a filling method, and the interpolation method includes a linear interpolation method and a polynomial interpolation method.

[0015] As a preferred technical solution, the characteristic values ​​of electricity consumption information include average electricity consumption, median electricity consumption, 25% quantile of electricity consumption, 75% quantile of electricity consumption, interquartile range of electricity consumption, maximum electricity consumption, minimum electricity consumption, range of electricity consumption, standard deviation of electricity consumption, variance of electricity consumption, coefficient of variation of electricity consumption, skewness of electricity consumption, kurtosis of electricity consumption and peak factor of electricity consumption.

[0016] As a preferred technical solution, the original data set of electricity consumption is divided into a training set and a validation set.

[0017] As a preferred technical solution, the parameter optimization of the platform-user relationship model adopts a cross-validation method.

[0018] As a preferred technical solution, the training process of the household relationship model includes the following steps:

[0019] S41, using the K-fold cross-validation partitioning method, divide the data set into K equal parts, use K-1 parts for training each time, and use the remaining parts for validation;

[0020] S42. Finally, the average result of each verification is taken as the performance evaluation index of the model.

[0021] During the cross-validation process, the key parameters of the model are gradually adjusted, including the number of trees and the maximum tree depth parameters; through continuous adjustment and verification, the optimal parameter combination is selected.

[0022] As a preferred technical solution, the trained household relationship model is evaluated using a validation set, and the evaluation indicators include accuracy, precision and recall.

[0023] As a preferred technical solution, the decision tree is trained independently as a base model, and multiple base models are combined into a strong classifier.

[0024] As a preferred technical solution, the classification result is obtained by voting or weighted averaging.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. The present invention extracts characteristic values ​​of electricity consumption information, establishes a station-user relationship recognition model, constructs multiple decision trees and performs independent predictions, and the prediction results do not affect each other. The final classification result is obtained based on the prediction results.

[0027] 2. The present invention uses the decision tree as a base model to be independently trained, and multiple base models are combined into a strong classifier. The classification result is obtained by voting or weighted averaging to ensure the accuracy of the result.

[0028] 3. The electricity consumption information characteristic values ​​of the present invention can reflect the user's electricity consumption habits in many aspects. The establishment of a station-user relationship identification model is more in line with the actual situation and the classification result is more accurate.

[0029] 4. The present invention combines the electricity consumption characteristics corresponding to the user type to smooth the electricity consumption data, eliminate random fluctuations, improve data stability, and improve the accuracy of user identification.

[0030] 5. During the cross-validation process of the present invention, the key parameters of the model are gradually adjusted, including the number of trees and the maximum tree depth parameters; through continuous adjustment and verification, the optimal parameter combination is selected so that the model can maintain a high prediction accuracy on new data, thereby improving the accuracy of household identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of user identification of the present invention;

[0032] Figure 2 Schematic diagram of the random forest algorithm of the present invention;

[0033] Figure 3 This is a confusion matrix diagram of an example of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work should fall within the scope of protection of the present invention.

[0035] Identifying the relationship between low-voltage stations and users is a key link in distribution network management. Traditional methods rely on manual inspections, ledger data, and station user identification devices. These methods are not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in inaccurate or incomplete data. With the rapid development of smart grids, the amount of distribution network data has increased dramatically. How to use these massive data to accurately identify and maintain the relationship between stations and users in low-voltage distribution networks has become a key challenge to improve the operating efficiency of distribution networks, reduce operation and maintenance costs, and enhance power supply reliability.

[0036] Therefore, the present invention provides a method for identifying the relationship between a station and a household based on a station-to-household relationship identification model. The present invention extracts the characteristic value of electricity consumption information, establishes the station-to-household relationship identification model, constructs multiple decision trees and performs independent predictions, and the prediction results do not affect each other, and the final classification result is obtained according to the prediction result. The present invention trains the decision tree independently as a base model, combines multiple base models into a strong classifier, and obtains the classification result by voting or weighted averaging to ensure the accuracy of the result. The characteristic value of electricity consumption information of the present invention reflects the user's electricity consumption habits in many aspects, and the establishment of a station-to-household relationship identification model is more in line with the actual situation, and the classification result is more accurate. The present invention combines the electricity consumption characteristics corresponding to the user type, smoothes the electricity consumption data, eliminates random fluctuations, improves data stability, and improves the accuracy of station-to-household identification. In the cross-validation process of the present invention, the key parameters of the model are gradually adjusted, including adjusting the number of trees and the maximum tree depth parameters; through continuous adjustment and verification, the optimal parameter combination is selected, so that the model can also maintain a high prediction accuracy on new data, and improve the accuracy of station-to-household identification.

[0037] Example 1

[0038] like Figure 1 As shown, a method for identifying a household relationship based on a household relationship identification model, the identification method comprises the following steps:

[0039] S1. Collecting the user's electricity consumption, where the electricity consumption is the user's daily electricity consumption for a continuous period of time;

[0040] S2. Process the erroneous data in the collection process and standardize all the processed data to ensure that the electricity consumption information of different users can be compared and analyzed on the same scale;

[0041] S3, based on the standardized data, extract the characteristic values ​​of electricity consumption information from the time domain and frequency domain, and construct a characteristic vector reflecting the user's electricity consumption habits;

[0042] S4. Use the random forest algorithm to establish a household relationship model and perform training and evaluation. By constructing multiple decision trees and making predictions separately, a strong classifier is combined to calculate the final classification result based on the prediction results of each decision tree.

[0043] The types of error data that need to be processed include data anomalies, data missing and invalid data; among them, in combination with the user's commissioning date information, users who are not in operation or have not generated electricity are eliminated; in combination with the user's design capacity, a reasonable threshold range is set for electricity consumption, and electricity consumption data that exceeds the normal operating range is marked, corrected or eliminated; in combination with the electricity consumption characteristics corresponding to the user type, the electricity consumption data is smoothed.

[0044] The missing data is processed by interpolation or filling method, and the interpolation method includes linear interpolation and polynomial interpolation.

[0045] In this embodiment, the user's electricity consumption data is collected in real time through the smart meter and the data acquisition and monitoring system, mainly including the user's voltage, current, active power, reactive power and power factor information. Compared with the high-frequency voltage and current data with strong time series, the power consumption data is time-accumulated data, and the time synchronization requirements are relatively low. Therefore, the power consumption data analysis has low data quality requirements and has a wider applicability.

[0046] The dynamic changes in the low-voltage distribution network topology, such as commissioning, withdrawal, power supply adjustment, etc., pose many challenges to data collection, and the accuracy and integrity of the data must be ensured during the data collection process. When collecting electricity consumption data, there may be a variety of reasons that lead to data anomalies or data loss. These reasons may include but are not limited to: the user has applied for access, but has not been put into operation, the user table number only exists in the index table, and there is no electricity consumption data; the user has been put into operation, but the electricity consumption data is missing or zero; the user has been put into operation, and there is only electricity consumption data after the commissioning date in the first month of operation; random data loss caused by errors in the measurement, transmission, and entry process.

[0047] When processing data, we combine the operation log to judge and consider the actual situation to avoid generating values ​​that do not conform to physical laws. First, data cleaning is used to identify and process data anomalies and invalid data. Combined with the user's commissioning date information, the reasons for the missing user data are determined, and users who are not commissioned or have not generated electricity are eliminated; combined with the user's design capacity, a reasonable threshold range is set for electricity consumption, and the electricity consumption data that exceeds the normal operating range is marked, corrected or eliminated; combined with the electricity consumption characteristics corresponding to the user type, the electricity consumption data is smoothed to eliminate random fluctuations and improve data stability. Secondly, for the problem of missing data, a combination of interpolation and filling methods is used to handle it. For missing data in a small range, the linear interpolation method is used, that is, the value of the missing data is estimated by the values ​​of the two valid data points before and after. For cases where the data is missing for a long time or the data fluctuates greatly, the polynomial interpolation method is used to fit a polynomial curve using multiple adjacent data points, and the missing values ​​are estimated based on the curve, so as to more accurately restore the trend of the data. In addition, since electricity consumption is highly cyclical, data from similar historical time periods are used to fill in missing values, especially when the missing data occurs at the head or tail of the list, which makes it difficult for interpolation methods to handle accurately.

[0048] After data processing is completed, all information will be standardized to ensure that the electricity consumption information of different users can be compared and analyzed on the same scale. The standardization process includes normalizing various data and converting data of different dimensions into a dimensionless unified standard. This step helps with subsequent feature extraction and model training, and improves recognition accuracy.

[0049] The characteristic values ​​of electricity consumption information include average electricity consumption, median electricity consumption, 25% quantile of electricity consumption, 75% quantile of electricity consumption, interquartile range of electricity consumption, maximum electricity consumption, minimum electricity consumption, range of electricity consumption, standard deviation of electricity consumption, variance of electricity consumption, coefficient of variation of electricity consumption, skewness of electricity consumption, kurtosis of electricity consumption and peak factor of electricity consumption.

[0050] In this embodiment, combined with the information of power consumption change trend, power fluctuation characteristics, time series correlation, etc., the power consumption information feature values ​​are extracted from the time domain and frequency domain to construct a set of feature vectors to reflect the user's daily power consumption habits, reveal the changes in the power grid operation status, and improve the recognition ability of the model in complex scenarios. i ) as an example, the time domain features include the following;

[0051] Average power consumption reflects the overall power consumption level of users. The formula is as follows:

[0052]

[0053] The median power consumption represents the median value of the user's power consumption (the data is arranged in ascending order before calculation). The formula is as follows:

[0054]

[0055] The 25% quantile of electricity consumption means that 25% of users in the data consume less than this value. The formula is as follows:

[0056] Q1=x (0.25×n)

[0057] The 75% quantile of electricity consumption means that 75% of users in the data consume less than this value. The formula is as follows:

[0058] Q3=x (0.75×n)

[0059] The interquartile range of electricity consumption (indicating the range of the middle 50% of the data) is used to measure the concentration and dispersion of the data. The formula is as follows:

[0060] IQR=Q3-Q1

[0061] The maximum value of power consumption is used to analyze the user's peak power consumption. The formula is as follows:

[0062] Max(x)=max(x1,x2,…,x n )

[0063] The minimum value of power consumption is used to analyze the user's minimum power consumption. The formula is as follows:

[0064] Min(x)=min(x1,x2,…,x n )

[0065] The range of electricity consumption indicates the fluctuation range of the user's electricity consumption. The formula is as follows:

[0066] R=Max(x)-Min(x)

[0067] The standard deviation and variance of electricity consumption reflect the fluctuation degree of user electricity consumption. The formula is as follows:

[0068]

[0069] The coefficient of variation of electricity consumption (standard deviation divided by the mean) indicates the relative degree of fluctuation, and the formula is as follows:

[0070]

[0071] The skewness of electricity consumption describes the symmetry of the distribution of user electricity consumption. The formula is as follows:

[0072]

[0073] The kurtosis of electricity consumption describes the sharpness of the distribution, and the formula is as follows:

[0074]

[0075] The peak factor of power consumption (the ratio of maximum value to RMS value) is used to measure the peak value of the data. The formula is as follows:

[0076]

[0077] The original electricity consumption dataset is divided into a training set and a validation set.

[0078] The parameters of the household relationship model are optimized by cross-validation method.

[0079] The training process of the household relationship model includes the following steps:

[0080] S41, using the K-fold cross-validation partitioning method, divide the data set into K equal parts, use K-1 parts for training each time, and use the remaining parts for validation;

[0081] S42. Finally, the average result of each verification is taken as the performance evaluation index of the model.

[0082] During the cross-validation process, the key parameters of the model are gradually adjusted, including the number of trees and the maximum tree depth parameters; through continuous adjustment and verification, the optimal parameter combination is selected.

[0083] The trained household relationship model is evaluated using the validation set, and the evaluation indicators include accuracy, precision and recall.

[0084] The decision trees are trained independently as base models, and multiple base models are combined into a strong classifier.

[0085] The classification result is obtained by voting or weighted averaging.

[0086] In this embodiment, based on the constructed user power consumption information feature value, a random forest algorithm with strong classification ability and good generalization performance is used to train and identify the station-user relationship. By constructing multiple decision trees and determining the final classification result through a majority voting mechanism, the overfitting problem that is prone to occur in a single model is avoided. Figure 2 shown.

[0087] To ensure the accuracy of the model, the cross-validation method is used to optimize the model parameters during the training process. Cross-validation divides the data set into multiple subsets, trains the model multiple times on different training-validation combinations, and evaluates its performance to avoid the problem of overfitting the model to the training data and select the best model parameters. First, the original data set of electricity consumption is divided into a training set and a validation set. The division method used is K-fold cross-validation, which divides the data set into K equal parts, uses K-1 parts for training each time, and the remaining parts are used for validation. This is repeated K times, and the average result of each validation is finally taken as the performance evaluation indicator of the model. Secondly, in the process of cross-validation, the key parameters of the model are gradually adjusted, including the number of adjustable trees and the maximum tree depth. Through continuous adjustment and verification, the optimal parameter combination is selected so that the model can maintain a high prediction accuracy on new data.

[0088] Ensemble learning is used to combine multiple base models into a strong classifier, so that the final model has stronger noise resistance and higher accuracy. Multiple random forest models are trained independently as base models. These base models make predictions separately, and finally obtain the final classification results by voting or weighted averaging. By integrating these models, the errors that may occur in a single model are reduced, different modes of data are captured to fully reflect the characteristics of the data, and the model's ability to resist data noise and outliers is enhanced.

[0089] After the model training is completed, the model is evaluated by using an independent test set to verify its performance in practical applications. The test set is an independent subset separated from the original data set. Unlike the training set, it does not participate in the model training process. The main function of the test set is to simulate the performance of the model in practical applications. It can effectively detect the generalization ability of the model, that is, the accuracy of the model's prediction of unknown data. In order to comprehensively evaluate the performance of the model, multiple evaluation indicators are selected, including accuracy, precision, recall, etc. Through the comprehensive analysis of the above evaluation indicators, the effect of the model in practical applications is judged.

[0090] Example 2

[0091] Taking the daily electricity consumption data of 225 users in two substations within one month as an example, the data is processed. The data of some users are shown in Table 1. The electricity consumption data of user No. 579 within one month is complete, the electricity of user No. 213 was delivered on the 25th of this month, and the data of user No. 214 was missing for two days this month. For user No. 213, there was no actual electricity consumption before the date of power supply, and the processing method was to set the data of the corresponding date to zero. For user No. 214, linear interpolation and historical filling methods were used to calculate the missing values, respectively. The interpolation method calculated the results of 2.12kWh and 2.14kWh, and the historical filling method calculated the results of 2.08kWh and 2.04kWh.

[0092] Table 1 Daily electricity consumption data of some users

[0093]

[0094]

[0095] Note: The unit of electricity consumption is kWh. NaN means non-number and no corresponding data.

[0096] Based on the collected electricity consumption data, its characteristic values ​​are constructed. Taking user No. 579 as an example, its characteristic parameter results are: average electricity consumption (5.25kWh), median electricity consumption (5.18kWh), 25% quantile of electricity consumption (4.87kWh), 75% quantile of electricity consumption (5.28kWh), interquartile range of electricity consumption (0.41kWh), maximum electricity consumption (5.31kWh), minimum electricity consumption (4.74kWh), range of electricity consumption (0.57kWh), standard deviation and variance of electricity consumption (0.364kWh / 0.1323kWh), coefficient of variation of electricity consumption (0.0693), skewness of electricity consumption (-0.35), kurtosis of electricity consumption (1.47), peak factor of electricity consumption (1.01).

[0097] Based on the characteristic values ​​of user electricity consumption data, the random forest algorithm is used to learn and judge the user station area. The model confusion matrix is ​​as follows: Figure 3 The accuracy and precision of the model, i.e., the recall rate, are 0.90909, 1, and 0.85714, respectively, indicating that the model has excellent performance in dividing the substations using only the user power consumption data.

[0098] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying a household relationship based on a household relationship identification model, characterized in that: The process of the identification method comprises the following steps: S1. Collecting the user's electricity consumption, where the electricity consumption is the user's daily electricity consumption for a continuous period of time; S2. Process the erroneous data in the collection process and standardize all the processed data to ensure that the electricity consumption information of different users can be compared and analyzed on the same scale; S3, based on the standardized data, extract the characteristic values ​​of electricity consumption information from the time domain and frequency domain, and construct a characteristic vector reflecting the user's electricity consumption habits; S4. Use the random forest algorithm to establish a household relationship model and perform training and evaluation. By constructing multiple decision trees and making predictions separately, a strong classifier is combined to calculate the final classification result based on the prediction results of each decision tree.

2. The method for identifying a household relationship based on a household relationship identification model according to claim 1, characterized in that: The types of error data to be processed include data anomalies, data missing and invalid data; Among them, combined with the user's commissioning date information, users who are not in operation or have not generated electricity are eliminated; based on the user's design capacity, a reasonable threshold range is set for electricity consumption, and electricity consumption data that exceeds the normal operating range is marked, corrected or eliminated; based on the electricity consumption characteristics corresponding to the user type, the electricity consumption data is smoothed.

3. The method for identifying a household relationship based on a household relationship identification model according to claim 2, characterized in that: The missing data is processed by interpolation or filling method, and the interpolation method includes linear interpolation and polynomial interpolation.

4. The method for identifying a customer relationship based on a customer relationship identification model according to claim 1, characterized in that: The characteristic values ​​of electricity consumption information include average electricity consumption, median electricity consumption, 25% quantile of electricity consumption, 75% quantile of electricity consumption, interquartile range of electricity consumption, maximum electricity consumption, minimum electricity consumption, range of electricity consumption, standard deviation of electricity consumption, variance of electricity consumption, coefficient of variation of electricity consumption, skewness of electricity consumption, kurtosis of electricity consumption and peak factor of electricity consumption.

5. The method for identifying a customer relationship based on a customer relationship identification model according to claim 1, characterized in that: The original electricity consumption dataset is divided into a training set and a validation set.

6. The method for identifying a customer relationship based on a customer relationship identification model according to claim 5, characterized in that: The parameters of the household relationship model are optimized by cross-validation method.

7. The method for identifying a household relationship based on a household relationship identification model according to claim 6, characterized in that: The training process of the household relationship model includes the following steps: S41, using the K-fold cross-validation partitioning method, divide the data set into K equal parts, use K-1 parts for training each time, and use the remaining parts for validation; S42. Finally, the average result of each verification is taken as the performance evaluation index of the model. During the cross-validation process, the key parameters of the model are gradually adjusted, including the number of trees and the maximum tree depth parameters; through continuous adjustment and verification, the optimal parameter combination is selected.

8. The method for identifying a household relationship based on a household relationship identification model according to claim 5, characterized in that: The trained household relationship model is evaluated using the validation set, and the evaluation indicators include accuracy, precision and recall.

9. The method for identifying a household relationship based on a household relationship identification model according to claim 1, characterized in that: The decision trees are trained independently as base models, and multiple base models are combined into a strong classifier.

10. The method for identifying a customer relationship based on a customer relationship identification model according to claim 1, characterized in that: The classification result is obtained by voting or weighted averaging.

Images