Low-voltage transformer area user phase identification method based on multiple measurement data

Abstract

The invention discloses a low-voltage transformer area user phase identification method based on multiple measurement data, which comprehensively utilizes two measurement data of voltage and power to carry out integrated similarity measurement, and solves the problems of a single voltage data method under the condition of three-phase voltage balance and a single power data method under the condition of low power consumption of users. According to the method, firstly, the redundancy of original data is reduced through voltage data screening and power feature extraction, then, agglomerated hierarchical clustering is carried out on users by using a screening voltage sequence to generate in-phase user clusters, and finally, secondary clustering is carried out on the users by using integrated similarity combining voltage and power similarity to generate in-phase user clusters. The phase attribution discrimination of the transformer area users is realized, and the method has relatively high phase identification accuracy and stability.

Description

Technical Field

[0001] The invention belongs to the field of low-voltage distribution station area topology identification, and in particular relates to a low-voltage station area user phase identification method based on multiple measurement data. Background Art

[0002] In recent years, with the development of the economy and the continuous advancement of urbanization, the number of users in low-voltage areas has grown rapidly. However, due to problems such as difficulty in investigation and untimely information updates, the archival information of user phases is often unknown or recorded incorrectly, which affects the upgrading and transformation of lines and the implementation of three-phase imbalance management. In order to solve the above problems, it is necessary to simultaneously carry out the identification of phase information of users in the area in actual engineering applications. At present, the methods for identifying the relationship between household transformers in low-voltage distribution stations are mainly divided into three categories: manual investigation, carrier signal detection, and measurement data analysis. The manual on-site investigation method has a large workload and low investigation efficiency. The signal detection method has weak anti-interference ability and high implementation and maintenance costs. The method based on area measurement data has the advantages of strong anti-interference ability and low implementation and maintenance costs, and has gradually become the mainstream method for solving the problem of phase identification in the area at this stage.

[0003] According to the different data used, the methods based on measurement data can generally be divided into methods based on voltage data and methods based on power data. The voltage data-based method is difficult to determine the user phase when the three-phase voltage in the substation is relatively balanced. The accuracy of the method using single power data is affected by the total number of users in the substation and the number of users that can be collected by the substation meter, or it requires sufficient power prominent change characteristics as support, that is, it requires a larger amount of data, resulting in an increase in the time complexity of the algorithm, and the method cannot distinguish between vacant users and low-power users well.

[0004] The recognition method that only uses a single data has a large deviation in recognition accuracy in some scenarios. In order to improve the accuracy of user phase recognition, the measurement data of the substation can be used in combination to make up for the shortcomings. The advantage of using both voltage and power data is that it can solve the problem that the single data method is difficult to guarantee the recognition accuracy in three-phase voltage balance scenarios and user low power consumption scenarios, improve the applicability in different scenarios, and have better performance. Summary of the invention

[0005] Purpose: The purpose of the present invention is to provide a low-voltage distribution area user phase identification method based on multiple measurement data in order to solve the phase identification problem in the topology identification of the low-voltage distribution area.

[0006] Technical solution: The low-voltage area user phase identification method based on multiple measurement data proposed in the present invention includes the following steps:

[0007] Step 1: Dataset preprocessing, including missing value filling, redundant information removal, and data standardization.

[0008] Step 2: Screening of voltage and power data: by screening the difference sub-segments of the voltage sequence and extracting the significant features of the power data, the original voltage and power data of the substation are processed.

[0009] Step 3: Filter the voltage sequence for clustering, cluster the users based on voltage similarity, and obtain the initial user clusters.

[0010] Step 4: Based on the secondary clustering of integrated similarity, the extracted voltage and power sequences are combined to measure the integrated similarity between the user and the substation master table, and the user clusters are clustered again using the clustering algorithm to achieve the final determination of user phase identification.

[0011] The step 1 comprises the following steps:

[0012] (11) Process the original data, delete illogical outliers, and use the data before the missing values ​​to fill in the missing position data.

[0013] (12) The original voltage data is standardized, and the formula is as follows:

[0014]

[0015] In the formula, It represents the measurement value of the nth measurement point at time t, T represents the total number of measurement times, and N represents the total number of measurement points, i.e. the total number of users. represents the mean voltage of all metering points at time t, Represents the voltage standard deviation of all metering points at time t. It represents the Z-Score standard value of the voltage at the nth metering point at time t.

[0016] Step 2 The specific method for voltage and power data screening is:

[0017] (21) The idea of ​​voltage difference sequence screening is to highlight the differences in user voltage sequences at different physical connection locations, so that the voltage changes of adjacent users in the same phase have a higher correlation with other users. According to the changes in the total meter power curve, the voltage sequence during the high power period of the phase line is screened.

[0018] Define the power threshold as θ p , using the minimum length T dur To exclude short data segments, the selected voltage sequence should satisfy the total three-phase power of the meter within the data segment time is greater than the power threshold. The process for selecting data segments can be formulated as:

[0019]

[0020]

[0021] In the formula, represents the voltage sequence after user i filters, represents the kth data segment, m k represents the index of the data point in the kth data segment, It represents the mean value of the total meter power data within the data period, and ΔT represents the sampling interval.

[0022] (22) The idea of ​​extracting the power significant change feature is to remove redundant small changes and extract the significant power changes relative to other users to better reflect the data similarity between the user and its phase total table. The steps are as follows:

[0023] 1) Calculate the user power change according to the formula, and combine the obtained changes to form a user power change sequence And the total meter power change sequence VP d .

[0024]

[0025]

[0026] In the formula, represents the power change of user i at time t relative to time tk, It represents the power change of the total meter phase d at time t relative to time tk, and the power change sequence is expanded by selecting different k values.

[0027] 2) According to the significant change extraction condition shown in formula 6, the significant change characteristics of user power are extracted, and the significant change characteristic sequence of user i is recorded as The index value corresponding to the significant change is The corresponding sequence of power changes in the total table extracted according to the index is recorded as VP′ d .

[0028]

[0029] In the formula, represents the power change of other users at time t, and α is the average change coefficient. Using this condition, the significant change features of each user are extracted, and the time corresponding to the significant change features is recorded.

[0030] The step 3 of screening the voltage sequence for a cluster comprises the following steps:

[0031] (31) According to the formula, the user similarity matrix D is calculated using the screening voltage sequence obtained in step 2. The formula is as follows:

[0032]

[0033] In the formula, pcc(U′ li，m , U′ lj，m ) filters the Pearson correlation coefficient of voltage series for users li and lj.

[0034] (32) Initially treat each user as a class;

[0035] (33) Merge the two closest classes into one class and calculate the change ΔS of the silhouette coefficient before and after the merger;

[0036] (34) If the silhouette coefficient change ΔS is less than 0, the process ends and a preliminary clustering result C is obtained; otherwise, the process returns to step (33).

[0037] The specific method of the secondary clustering based on integrated similarity in step 4 is:

[0038] The integrated similarity is used as the distance metric for secondary clustering. For the user cluster C in the clustering result C in step 3, k The user voltage sequence in the cluster is calculated, and the similarity distance between the user voltage sequence and the voltage sequence of each phase in the total table is calculated. The average D v As the voltage similarity distance between the user cluster and the total table. k The similarity distance with phase d is expressed as:

[0039]

[0040] For cluster C k The power data of the internal users are randomly selected and summed as the expansion sequence. Then, the similarity between the sequence and the corresponding sequence in the total table is calculated based on the power significant change feature sequence, and the average value D is taken. p As the power similarity distance between the user cluster and the total table.

[0041] The final joint similarity distance D is the sum of the two:

[0042] D=D v +βD p (9)

[0043] The parameter β is used to make the dimensions of the two changes consistent.

[0044] Based on this integrated similarity distance, the user clusters in the user cluster set C are clustered again. The specific steps are as follows:

[0045] 1) Initialize the cluster center as the three-phase voltage sequence and power change sequence of the total meter;

[0046] 2) Calculate the user cluster C to be identified based on the integrated similarity distancek The distance to each cluster center;

[0047] 3) Classify each user cluster into the cluster corresponding to the closest distance, and sort the minimum distance from each user cluster to the cluster center. Mark the user cluster corresponding to the minimum value as a high-credibility user cluster;

[0048] 4) Calculate the mean sequence of the voltage sequence of the users in the high-credibility user cluster and the corresponding phase total meter, and subtract the power change sequence in the user cluster from the power change sequence of the corresponding phase total meter;

[0049] 5) The cluster center is updated to the voltage mean sequence and the power change sequence after subtraction, and the power significant change feature extraction condition is updated and iterated according to the formula;

[0050]

[0051] In the formula, NH represents the total number of users included in the high-trust user cluster, Represents the power change of users in the high-trust user cluster at time t.

[0052] 6) If the user cluster to be identified is 0, end, otherwise return to step 2).

[0053] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The comprehensive use of two measurement data, voltage and power, reflects the advantages of multiple data fusion methods, solves the shortcomings of single data methods, and makes the present invention have higher accuracy and stronger applicability in user phase identification. 2. The original voltage data is screened to reduce data redundancy and reflect important data features. The primary clustering method based on screened voltage in the invention effectively guarantees the clustering effectiveness and stability of adjacent users in the same phase. 3. The power data is subjected to significant change feature extraction, and a voltage and power integrated similarity measurement method is proposed. The significant feature extraction iterative mechanism is introduced in the secondary clustering model, which effectively improves the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 : Overall flow chart of the present invention;

[0055] Figure 2 : Screening voltage sequence clustering flow chart;

[0056] Figure 3 :Flowchart of secondary clustering based on integrated similarity DETAILED DESCRIPTION

[0057] The technical solution is described in detail below in conjunction with the accompanying drawings:

[0058] The actual measurement data of smart meters in two substations in a certain city are taken as an example to illustrate and verify. The number of users in the two substations is 55, and they are single-phase users. The real phase of the users is obtained through on-site collection. The sampling interval of the substation measurement data is 15 minutes. The data of 20 consecutive days in the substation is selected as the total data, and the data of 3 consecutive days is selected as the model input data. A total of 18 experiments are conducted, and the average result is taken as the comparison data of the experiment. The recognition accuracy is defined as the ratio of the number of correctly identified users to the total number of users involved in the recognition. The basic information of users in the substation is shown in Table 1.

[0059] Table 1 Phase distribution information of users in the area

[0060]

[0061]

[0062] The overall process of the phase identification operation of the station area user using the present invention is as follows: Figure 1 The specific steps are as follows:

[0063] 1. Data preprocessing, including anomaly and missing value processing, data standardization, the specific steps are:

[0064] (1) Process the original data, delete illogical outliers, and use the data before the missing values ​​to fill in the missing data.

[0065] (2) Perform Z-Score normalization on the original voltage data, the formula is as follows:

[0066]

[0067] In the formula, It represents the measurement value of the nth measurement point at time t, T represents the total number of measurement times, and N represents the total number of measurement points, i.e. the total number of users. represents the mean voltage of all metering points at time t, Represents the voltage standard deviation of all metering points at time t. It represents the Z-Score standard value of the voltage at the nth metering point at time t.

[0068] 2. Voltage and power data screening, including voltage data screening and power significant change extraction.

[0069] (1) Define the power threshold as θ p , using the minimum length T dur To exclude short data segments, the selected voltage sequence should satisfy the total three-phase power of the meter within the data segment time is greater than the power threshold. The process for selecting data segments can be formulated as:

[0070]

[0071]

[0072] In the formula, represents the voltage sequence after user i filters, represents the kth data segment, m k represents the index of the data point in the kth data segment, It represents the mean value of the total meter power data within the data period, and ΔT represents the sampling interval.

[0073] (2) For the original power data, calculate the change sequence of the three-phase power data of the user and the total meter, and extract the significant change features of the user quantity sequence to form a significant change sequence. The steps are as follows:

[0074] 1) Calculate the user power change according to the formula, and combine the obtained changes to form a user power change sequence And the total meter power change sequence VP d .

[0075]

[0076]

[0077] In the formula, represents the power change of user i at time t relative to time tk, It represents the power change of the total meter phase d at time t relative to time tk, and the power change sequence is expanded by selecting different k values.

[0078] 2) According to the significant change extraction conditions, extract the significant change characteristics of user power, and record the significant change characteristic sequence of user i as The index value corresponding to the significant change is The corresponding sequence of power changes in the total table extracted according to the index is recorded as VP′ d The extraction conditions of significant changes are:

[0079]

[0080] In the formula, represents the power change of other users at time t, and α is the average change coefficient. Using this condition, the significant change features of each user are extracted, and the time corresponding to the significant change features is recorded.

[0081] 3. Use the screening voltage sequence to cluster users.

[0082] The obtained screening voltage sequence is used to cluster the users. The clustering process is as follows: Figure 2The specific steps of clustering are as follows:

[0083] (1) According to the formula, the user similarity matrix D is calculated using the screening voltage sequence;

[0084]

[0085] In the formula, pcc(U′ li，m , U′ lj，m ) filters the Pearson correlation coefficient of voltage series for users li and lj.

[0086] (2) Initially treat each user as a class;

[0087] (3) Merge the two closest classes into one class and calculate the change in silhouette coefficient ΔS before and after the merger;

[0088] (4) If the silhouette coefficient change ΔS is less than 0, the process ends and a preliminary clustering result C is obtained; otherwise, the process returns to step (3).

[0089] As shown in Table 2, the primary clustering method in step 3 of the present invention is compared with other methods. In the experiment, since the k-means algorithm sometimes converges to the local optimum, the clustering result is not stable, and the DBSCAN algorithm is greatly affected by the initial parameters and tends to cluster heterogeneous users with close distances into one category. It can be concluded from the experimental results that compared with the k-means algorithm and the DBSCAN algorithm, the primary clustering algorithm used in the present invention has better effectiveness and stability.

[0090] Table 2 Clustering effect of different clustering methods

[0091]

[0092] 4. Perform secondary clustering of user clusters based on integrated similarity

[0093] The integrated similarity is used as the distance metric for secondary clustering. For the user cluster C in the clustering result C k The user voltage sequence in the cluster is calculated, and the similarity distance between the user voltage sequence and the voltage sequence of each phase in the total table is calculated. The average D v As the voltage similarity distance between the user cluster and the total table. k The similarity distance with phase d is expressed as:

[0094]

[0095] For cluster C k The power data of the internal users are randomly selected and summed as the expansion sequence. Then, the similarity between the sequence and the corresponding sequence in the total table is calculated based on the power significant change feature sequence, and the average value D is taken. pAs the power similarity distance between the user cluster and the total table.

[0096] The final joint similarity distance D is the sum of the two:

[0097] D=D v +βD p

[0098] The parameter β is used to make the dimensions of the two changes consistent.

[0099] Based on this integrated similarity distance, the user clusters in the user cluster set C are clustered again. The secondary clustering process is as follows: Figure 3 As shown, the specific steps are:

[0100] 1) Initialize the cluster center as the three-phase voltage sequence and power change sequence of the total meter;

[0101] 2) Calculate the user cluster C to be identified based on the integrated similarity distance k The distance to each cluster center;

[0102] 3) Classify each user cluster into the cluster corresponding to the closest distance, and sort the minimum distance from each user cluster to the cluster center. Mark the user cluster corresponding to the minimum value as a high-credibility user cluster;

[0103] 4) Calculate the mean sequence of the voltage sequence of the users in the high-credibility user cluster and the corresponding phase total meter, and subtract the power change sequence in the user cluster from the power change sequence of the corresponding phase total meter;

[0104] 5) The cluster center is updated to the voltage mean sequence and the power change sequence after subtraction, and the power significant change feature extraction condition is updated and iterated according to the formula;

[0105]

[0106] In the formula, NH represents the total number of users included in the high-trust user cluster, Represents the power change of users in the high-trust user cluster at time t.

[0107] 6) If the user cluster to be identified is 0, end, otherwise return to step 2).

[0108] As shown in Table 3, in the comparison of the accuracy of phase recognition results, the accuracy of the method combining voltage and power data is generally higher than that of the method using single data, and the recognition accuracy of the present invention is the highest. In the two substations used in the experiment, there are more low-power users in Substation No. 7 than in Substation No. 8. The method using power data in the above experiment has better recognition effect in Substation No. 8 than in Substation No. 7. By comparing methods 1, 2, and 4, it can be seen that after using the integrated similarity metric, the accuracy of methods 1 and 2 in Substation No. 7 is higher than that of method 4, and method 1 is better than method 2. Therefore, when the power consumption characteristics of users in the substation area are not obvious, the method of the present invention can improve the recognition accuracy of low-power users, and after introducing the feature extraction condition iteration, the recognition accuracy has been further improved, which verifies the effectiveness of the integrated similarity and feature extraction condition iteration mechanism in the present invention.

[0109] Table 3 Phase recognition effects of different recognition methods

[0110]

Claims

1. A method for identifying phase of low voltage area users based on multiple measurement data, characterized in that: The following steps are involved: Step 1: Dataset preprocessing, including missing value filling, redundant information removal, and data standardization; Step 2: Screening of voltage and power data: by screening the voltage sequence difference sub-segments and extracting the significant features of power data, the original voltage and power data of the substation area are processed; Step 3: Filter the voltage sequence for clustering, cluster the users based on voltage similarity, and obtain the initial user clusters; Step 4: Based on the secondary clustering of integrated similarity, the extracted voltage and power sequences are combined to measure the integrated similarity between the user and the substation master table, and the user clusters are secondary clustered using the clustering algorithm to achieve the final determination of user phase identification.

2. According to claim 1, a method for identifying phases of low-voltage area users based on multiple measurement data is characterized in that: The specific steps of screening the voltage sequence difference sub-segments and extracting the significant features of the power data described in step 2 are as follows: Data screening includes two parts: voltage difference series screening and power significant change feature extraction. (21) According to the changes in the total meter power curve, the voltage sequence of the phase line during the high power period is selected. Define the power threshold as θ p , using the minimum length T dur To exclude short data segments, the selected voltage sequence should satisfy the total three-phase power of the meter within the data segment time is greater than the power threshold. Therefore, the process for selecting data segments can be formulated as: In the formula, represents the voltage sequence after user i filters, represents the kth data segment, m k represents the index of the data point in the kth data segment, It represents the mean value of the total meter power data within the data period, and ΔT represents the sampling interval. (22) Extraction of significant power change features: Extracting significant power changes relative to other users to better reflect the data similarity between the user and its phase table. The steps of extracting significant change features are as follows: 1) Calculate the user power change according to the formula, and combine the obtained changes to form a user power change sequence And the total meter power change sequence VP d . In the formula, represents the power change of user i at time t relative to time tk, It represents the power change of the total meter phase d at time t relative to time tk, and the power change sequence is expanded by selecting different k values. 2) According to the significant change extraction conditions, extract the significant change characteristics of user power, and record the significant change characteristic sequence of user i as The index value corresponding to the significant change is The corresponding sequence of power changes in the total table extracted according to the index is recorded as VP′ d The formula for extracting significant changes is as follows: In the formula, represents the power change of other users at time t, and α is the average change coefficient. Using this condition, the significant change features of each user are extracted, and the time corresponding to the significant change features is recorded.

3. The method for identifying the phase of a low voltage area user based on multiple measurement data according to claim 1, characterized in that: The specific steps of the primary clustering of the screening voltage sequence described in step 3 are: (31) According to the formula, the user similarity matrix D is calculated using the screening voltage sequence obtained in step 2; In the formula, pcc(U′ li，m , U′ lj，m ) is user I i and I j Pearson correlation coefficient of the screening voltage series. (32) Initially treat each user as a class; (33) Merge the two closest classes into one class and calculate the change ΔS of the silhouette coefficient before and after the merger; (34) If the silhouette coefficient change ΔS is less than 0, the process ends and a preliminary clustering result C is obtained; otherwise, the process returns to step (33).

4. The method for identifying the phase of a low voltage area user based on multiple measurement data according to claim 1, characterized in that: The specific method of secondary clustering based on integrated similarity described in step 4 is: The integrated similarity is used as the distance metric for secondary clustering. For the user cluster C in the clustering result C in step 3, k The user voltage sequence in the cluster is calculated, and the similarity distance between the user voltage sequence and the voltage sequence of each phase in the total table is calculated. The average D v As the voltage similarity distance between the user cluster and the total table. k The similarity distance with phase d is expressed as: For cluster C k The power data of the internal users are randomly selected and summed as the expansion sequence. Then, the similarity between the sequence and the corresponding sequence in the total table is calculated based on the power significant change feature sequence, and the average value D is taken. p As the power similarity distance between the user cluster and the total table. The final joint similarity distance D is the sum of the two: D=D v +βD p (8) The parameter β is used to make the dimensions of the two changes consistent. Based on this integrated similarity distance, the user clusters in the user cluster set C are clustered again. The specific steps are as follows: 1) Initialize the cluster center as the three-phase voltage sequence and power change sequence of the total meter; 2) Calculate the user cluster C to be identified based on the integrated similarity distance k The distance to each cluster center; 3) Classify each user cluster into the cluster corresponding to the closest distance, and sort the minimum distance from each user cluster to the cluster center. Mark the user cluster corresponding to the minimum value as a high-credibility user cluster; 4) Calculate the mean sequence of the voltage sequence of the users in the high-credibility user cluster and the corresponding phase total meter, and subtract the power change sequence in the user cluster from the power change sequence of the corresponding phase total meter; 5) The cluster center is updated to the voltage mean sequence and the power change sequence after subtraction, and the power significant change feature extraction condition is updated and iterated according to the formula; In the formula, NH represents the total number of users included in the high-trust user cluster, Represents the power change of users in the high-trust user cluster at time t. 6) If the user cluster to be identified is 0, end, otherwise return to step 2).

Images