A low voltage substation phase sequence identification method and system based on voltage curve segmentation feature clustering

Abstract

The present invention discloses a low-voltage substation phase sequence identification method and system based on voltage curve segmentation feature clustering, including: obtaining voltage data of substations and low-voltage users for multiple consecutive days; preprocessing voltage data; segmenting data to form segmented data sets; extracting the mean value, coefficient of variation, skewness, and kurtosis for each segment to form a feature sequence; using the central difference method to form a central difference feature sequence set for each segment; establishing a similarity measurement function; establishing a k-means phase sequence identification clustering model; using the model to classify low-voltage users, and labeling users with three-phase labels A, B, and C according to the cluster center to which the users ultimately belong; processing low-voltage users through the model for multiple consecutive days, and taking the phase labels that appear multiple times as the final prediction. The present invention generates feature sequences based on voltage curve segmentation, combines similarity measurement functions, and uses the improved k-means clustering method to significantly improve the accuracy of the low-voltage substation phase sequence identification method.

Description

Technical Field

[0001] The present invention belongs to the field of low-voltage distribution networks, and in particular relates to a low-voltage substation phase sequence identification method and system based on voltage curve segmentation feature clustering. Background Art

[0002] With the continuous expansion of the distribution network and the increasing complexity of its structure, the operation and maintenance of low-voltage distribution networks are facing increasingly severe challenges. The increase in new loads, the frequent relocation of lines, and the adjustment of three-phase unbalanced loads have made the phase topology connection information of low-voltage substations dynamic and changeable. However, the current update speed of archive record information is difficult to keep pace with it, which to a certain extent restricts the pace of intelligent and digital development of power grids.

[0003] In this complex network architecture, the accuracy of phase sequence plays a vital role. The correct phase sequence is not only the basis for the stable operation of the power grid, but also the guarantee for the safe use of electricity by users. If the phase sequence is deviated, the power grid operators will not be able to accurately understand the actual operating conditions of each phase of the low-voltage power distribution system, hindering the subsequent three-phase imbalance management, thus affecting the economic and stable operation of the low-voltage power distribution system. Therefore, ensuring the accuracy of the phase sequence in the distribution network is the key to ensuring the safe, reliable and efficient operation of power supply. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a low-voltage substation phase sequence identification method and system based on voltage curve segmentation feature clustering, which can significantly improve the accuracy of low-voltage substation phase sequence identification.

[0005] Technical solution: A low voltage phase sequence identification method based on voltage curve segmentation feature clustering of the present invention comprises:

[0006] Data acquisition: From the electricity consumption information collection system, obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, as well as the continuous single-phase voltage data of low-voltage single-phase users for multiple days to form a voltage curve data set;

[0007] Data preprocessing: preprocess the voltage curve data set to eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and outliers;

[0008] Segment feature sequence generation: segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segment data set; then extract the average value of each segment to form an average feature sequence; extract the coefficient of variation of each segment to form a coefficient of variation feature sequence; extract the skewness of each segment to form a skewness feature sequence; extract the kurtosis of each segment to form a kurtosis feature sequence; use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes;

[0009] Establishment of similarity measurement function: Establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change;

[0010] Model establishment and prediction: First, through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, the k-means phase sequence recognition clustering model is established by reasonably setting the initial cluster center; then, through the segmented feature sequence of low-voltage users, the k-means phase sequence recognition clustering model is used to divide low-voltage users into three categories, and the three categories of low-voltage users are marked with phase labels of A, B, and C based on the similarity measurement between the segmented feature sequence of the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation; finally, the low-voltage users are predicted by the k-means phase sequence recognition clustering model for many consecutive days, and the phase labels that appear multiple times are taken as the final prediction.

[0011] Furthermore, the data preprocessing process is as follows:

[0012] Delete irrelevant and duplicate data in the voltage curve dataset, and use the function in Python to identify the special identifiers NaN, NULL, and NA in the dataset to identify missing values;

[0013] Calculate the median absolute deviation of the voltage curve data set and use the median absolute deviation method to identify outliers. The calculation formula is as follows:

[0014] MAD=median(|v0-median(V)|,|v1-median(V)|,...,|v 95 -median(V)|)

[0015] Among them, MAD represents the median absolute deviation of the voltage curve data set; median represents the median; v0, v1...v 95 represents each numerical point in the voltage curve; V represents the voltage curve data set; therefore,

[0016] |v i -median(V)|>3×MAD

[0017] Any numerical point satisfies the above v i All of them were identified as outliers;

[0018] The Newton interpolation method is used to fill the identified missing values ​​and outliers. The formula of the Newton interpolation method is as follows:

[0019]

[0020] Where P(x) represents the polynomial function obtained by Newton interpolation method, which is used to estimate the interpolation result at a given position x of the measured data point; v0 represents the initial voltage value point; x j Indicates the position of the jth voltage value point; Δ k v0 represents the k-order difference quotient of v0, and the difference quotient is defined as follows:

[0021] Δ k v i =Δ k-1 v i+1 -Δ k-1 v i

[0022] Δ 1 v i =v i+1 -v i

[0023] Among them, Δ k-1 v i+1 Indicates v i+1 k-1 order difference quotient; v i represents the voltage value at the i-th position; v i+1 Indicates the voltage value at the i+1th position.

[0024] Furthermore, the average value is extracted for each segment to form an average value feature sequence, including:

[0025] Calculate the average value of each segment sequence in the segmented data set to form an average feature sequence in, represents [v0,...,v 11 ], and so on.

[0026] Furthermore, the coefficient of variation is extracted for each segment to form a characteristic sequence of the coefficient of variation, including:

[0027] Calculate the coefficient of variation value for each segment sequence in the segmented data set to form a characteristic sequence of the coefficient of variation in, represents [v0,...,v 11 The specific calculation process of the coefficient of variation is as follows:

[0028] First, calculate the average value of each segment of data Then, calculate the standard deviation S of each segment of data i , the formula is as follows:

[0029]

[0030] Among them, v j represents the voltage value of the jth point, and n represents the data length of the segment.

[0031] Finally, according to the standard deviation S i With the average Calculate the coefficient of variation for each segment of data The calculation formula is as follows:

[0032]

[0033] Furthermore, the skewness is extracted for each segment to form a skewness feature sequence, including:

[0034] Calculate the skewness value for each segment sequence in the segmented data set to form a skewness feature sequence in, represents [v0,...,v 11 The specific calculation process of skewness is as follows:

[0035] First, calculate the average value of each segment of data Then, find the median of each segment of data (V i );

[0036] Finally, according to the average value of each segment of data The median of each data segment (V i ) and the standard deviation S of each segment of data i , calculate the skewness characteristics of each segment of data The calculation formula is as follows:

[0037]

[0038] Furthermore, the kurtosis is extracted for each segment to form a kurtosis feature sequence, including:

[0039] Calculate the kurtosis value for each segmented sequence in the segmented data set to form a kurtosis characteristic sequence in, represents [v0,...,v 11 The specific calculation process of kurtosis is as follows:

[0040] First, find the fourth power of the standard deviation of each segment of data. The calculation formula is as follows:

[0041]

[0042] Then, based on the average value of each segment of data Find the fourth-order center distance of each segment of data The calculation formula is as follows:

[0043]

[0044] Finally, according to the fourth-order center distance of each segment of data and the fourth power of the standard deviation of each segment of data Calculate the kurtosis characteristic of each segment of data The calculation formula is as follows:

[0045]

[0046] Furthermore, the central difference method is used for each segment to form a central difference feature sequence set that reflects the local trend changes, including:

[0047] Based on the segmented dataset [v0,...,v 11 ], the formula for obtaining the central difference is as follows:

[0048]

[0049] Among them, f(x+h) and f(xh) are the values ​​at a distance h to the right and left of point x; according to the above formula, when h is equal to 1, the central differential feature sequence of the segmented data set is obtained in,

[0050]

[0051] Finally, the above operations are performed on all segmented data sets to obtain the central differential feature sequence set reflecting the local trend changes. The details are as follows:

[0052] [[v′1,v′2,...,v′ 10 ],[v′ 13 ,v′ 14 ,...,v′ 22 ],....,[v′ 85 ,v′ 86 ,...,v′ 94 ]].

[0053] Furthermore, a similarity measurement function is established, including:

[0054] Get the average feature sequence of user a Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence

[0055] Get the average feature sequence of user b Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence

[0056] The formula for the similarity measurement function between user a and user b is as follows:

[0057]

[0058] in, express express express express The Euclidean distance of the characteristic series of mean, coefficient of variation, skewness and kurtosis was calculated through these four variables. express The DTW distance between two sequences is calculated based on the set of all curved paths between two time series. s ,…,w r} to find the best bending path so that the total bending cost is minimized. The specific calculation is:

[0059]

[0060] Among them, w s is the coordinate of the sth element in the curved path; r is the total number of elements in the path; d(w s ) is w s Corresponding coordinates v′ a,i and v′ b,j The Euclidean distance between for and Specifically, the DTW distance between each segment is calculated by a dynamic programming method.

[0061] Furthermore, the process of establishing the k-means phase sequence recognition clustering model is as follows:

[0062] The three-phase voltage segment feature sequence of the substation is selected as the initial cluster center;

[0063] The distance between the three-phase voltage segment feature sequence of the substation and the voltage segment feature sequence of the user is calculated using the similarity measurement function. The smaller the distance, the higher the similarity, that is, the low-voltage user is classified into this cluster center.

[0064] After processing the data of all low-voltage users, for the low-voltage users in the same cluster, take the average value of their segmented feature sequences as the new cluster center, and recalculate the distance between the segmented feature sequence of the cluster center and the voltage segmented feature sequence of the low-voltage user. Similarly, the smaller the distance, the higher the similarity, that is, the user is classified into this cluster center;

[0065] Repeat the previous step until the phase category of all low-voltage users no longer changes. Finally, calculate the similarity measure between the cluster center and the segmented feature sequence of the three-phase voltage in the substation area, pair the cluster center with the three-phase voltage in the substation area one by one, and obtain the k-means phase sequence recognition clustering model.

[0066] Based on the same inventive concept, a low voltage phase sequence identification system based on voltage curve segment feature clustering of the present invention comprises:

[0067] The data acquisition module is used to obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, and the continuous single-phase voltage data of low-voltage single-phase users for multiple days from the power consumption information collection system to form a voltage curve data set;

[0068] A data preprocessing module is used to preprocess the voltage curve data set, eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and abnormal values;

[0069] The segmented feature sequence generation module is used to segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segmented data set; to extract the average value of each segment to form an average value feature sequence; to extract the coefficient of variation of each segment to form a coefficient of variation feature sequence; to extract the skewness of each segment to form a skewness feature sequence; to extract the kurtosis of each segment to form a kurtosis feature sequence; to use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes;

[0070] A similarity measurement function establishment module is used to establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change;

[0071] The model building and prediction module is used to establish a k-means phase sequence recognition clustering model through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, by reasonably setting the initial clustering center; it is used to divide low-voltage users into three categories using the k-means phase sequence recognition clustering model through the segmented feature sequence of low-voltage users, and combine the similarity measurement between the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation to label the three categories of low-voltage users with phase labels of A, B, and C; it is also used to predict low-voltage users through the k-means phase sequence recognition clustering model for multiple consecutive days, and take the phase labels that appear multiple times as the final prediction.

[0072] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0073] This scheme systematically obtains the continuous multi-day voltage data of the substation and low-voltage users from the electricity consumption information collection system, uses the segmented aggregation approximation method to extract the mean, coefficient of variation, skewness and kurtosis of each segment, and uses the segmented central difference sequence to establish a similarity measurement function. Combined with the improved k-means clustering method, this scheme finally constructs a low-voltage substation phase sequence recognition model.

[0074] After experimental verification, the low-voltage phase sequence identification model of the present invention has shown excellent performance and significant effect improvement in the low-voltage phase sequence identification task. The model makes full use of the characteristic characteristics of voltage data, significantly improves the accuracy and efficiency of the phase sequence identification task, and makes a positive contribution to the development of the low-voltage phase sequence identification field.

[0075] The phase sequence identification method adopted by the present invention is highly automated, greatly reducing the reliance on manual operation, thereby effectively improving the operating efficiency and safety performance of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic flow chart of a low voltage station phase sequence identification method based on voltage curve segment feature clustering disclosed in an embodiment of the present invention;

[0077] Figure 2 A confusion matrix diagram of a low-voltage substation phase sequence identification method based on voltage curve segment feature clustering disclosed in an embodiment of the present invention on an example prediction set;

[0078] Figure 3 A histogram of the index evaluation of the low-voltage phase sequence identification method based on voltage curve segmentation feature clustering in the example prediction set disclosed in the embodiment of the present invention;

[0079] Figure 4A diagram showing the dimensionality reduction after classification of a segmented feature data model of low-voltage users in a substation area disclosed in an embodiment of the present invention;

[0080] Figure 5 The present invention discloses a schematic diagram of the structure of a low-voltage phase sequence identification system based on voltage curve segment feature clustering. DETAILED DESCRIPTION

[0081] The technical solution of the present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings of the specification.

[0082] Example 1

[0083] like Figure 1 As shown, a low voltage phase sequence identification method based on voltage curve segmentation feature clustering of the present invention comprises the following steps:

[0084] S1. Data acquisition: From the electricity consumption information collection system, obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, as well as the continuous single-phase voltage data of low-voltage single-phase users for multiple days to form a voltage curve data set.

[0085] S2. Data preprocessing: Preprocess the voltage curve data set to eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and abnormal values. Specifically, the following steps are included:

[0086] S2.1. Delete irrelevant and duplicate data in the voltage curve dataset, and use the function in Python to identify the special identifiers NaN, NULL, and NA in the dataset to identify missing values;

[0087] S2.2. Calculate the median absolute deviation of the voltage curve data set and use the median absolute deviation method to identify outliers. The calculation formula is as follows:

[0088] MAD=median(|v0-median(V)|,|v1-median(V)|,...,|v 95 -median(V)|)

[0089] Among them, MAD represents the median absolute deviation of the voltage curve data set; median represents the median; v0, v1...v 95 represents each numerical point in the voltage curve; V represents the voltage curve data set; therefore,

[0090] |v i -median(V)|>3×MAD

[0091] Any numerical point satisfies the above v i All of them were identified as outliers;

[0092] S2.3. Use Newton interpolation to fill in the missing values ​​and outliers identified. The formula of Newton interpolation is as follows:

[0093]

[0094] Where P(x) represents the polynomial function obtained by Newton interpolation method, which is used to estimate the interpolation result at a given position x of the measured data point; v0 represents the initial voltage value point; x j Indicates the position of the jth voltage value point; Δ k v0 represents the k-order difference quotient of v0, and the difference quotient is defined as follows:

[0095] Δ k v i =Δ k-1 v i+1 -Δ k-1 v i

[0096] Δ 1 v i =v i+1 -v i

[0097] Among them, Δ k-1 v i+1 Indicates v i+1 k-1 order difference quotient; v i represents the voltage value at the i-th position; v i+1 Indicates the voltage value at the i+1th position.

[0098] S3, segment feature sequence generation: segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segment data set; then, extract the average value of each segment to form an average feature sequence; extract the coefficient of variation of each segment to form a coefficient of variation feature sequence; extract the skewness of each segment to form a skewness feature sequence; extract the kurtosis of each segment to form a kurtosis feature sequence; use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes. Specifically, it includes the following steps:

[0099] S3.1. Segment the preprocessed voltage curve data set as follows:

[0100] Based on the preprocessed voltage curve dataset [v0,v1,...v i ,...,v 95 ],v i Represents the voltage value at the i-th position. The voltage curve data set is divided into 8 segments, each with 12 data, to obtain the segmented data set:

[0101] [[v0,...,v 11 ],[v 12 ,...,v 23 ],...[v 84 ,...,v 95 ]].

[0102] S3.2. Extract the average value for each segment to form an average feature sequence, as follows:

[0103] Take the average value of each segment sequence in the above segment data set to form the average feature sequence in, represents [v0,...,v 11 ], and so on.

[0104] S3.3. Extract the coefficient of variation for each segment to form a characteristic sequence of the coefficient of variation, as follows:

[0105] Calculate the coefficient of variation value for each segment sequence in the above segment data set to form a coefficient of variation feature sequence in, represents [v0,...,v 11 The specific calculation process of the coefficient of variation is as follows:

[0106] First, calculate the average value of each segment of data Then, calculate the standard deviation S of each segment of data i , the formula is as follows:

[0107]

[0108] Among them, v j represents the voltage value of the jth point, and n represents the data length of the segment.

[0109] Finally, according to the standard deviation S i With the average Calculate the coefficient of variation for each segment of data The calculation formula is as follows:

[0110]

[0111] S3.4. Extract the skewness of each segment to form a skewness feature sequence, as follows:

[0112] Calculate the skewness value for each segmented sequence in the above segmented data set to form a skewness feature sequence in, represents [v0,...,v 11 The specific calculation process of skewness is as follows:

[0113] First, calculate the average value of each segment of data Then, find the median of each segment of data (V i ), calculate the standard deviation S of each segment of data according to S3.3 i ;

[0114] Finally, according to the average value of each segment of data The median of each data segment (V i ) and the standard deviation S of each segment of data i , calculate the skewness characteristics of each segment of data The calculation formula is as follows:

[0115]

[0116] S3.5. Extract the kurtosis of each segment to form a kurtosis feature sequence, as follows:

[0117] Calculate the kurtosis value for each segmented sequence in the above segmented data set to form a kurtosis characteristic sequence in, represents [v0,...,v 11 The specific calculation process of kurtosis is as follows:

[0118] First, according to the above, calculate the fourth power of the standard deviation of each data segment The calculation formula is as follows:

[0119]

[0120] Then, based on the average value of each segment of data Find the fourth-order center distance of each segment of data The calculation formula is as follows:

[0121]

[0122] Finally, according to the fourth-order center distance of each segment of data and the fourth power of the standard deviation of each segment of data Calculate the kurtosis characteristic of each segment of data The calculation formula is as follows:

[0123]

[0124] S3.6. For each segment, the central difference method is used to form a central difference feature sequence set that reflects the local trend changes, as follows:

[0125] Based on the segmented dataset [v0,...,v 11 ], the formula for obtaining the central difference is as follows:

[0126]

[0127] Among them, f(x+h) and f(xh) are the values ​​at a distance h to the right and left of point x; according to the above formula, when h is equal to 1, the central differential feature sequence of the segmented data set is obtained in,

[0128]

[0129] Finally, the above operations are performed on all segmented data sets to obtain the central differential feature sequence set reflecting the local trend changes. The details are as follows:

[0130] [[v′1,v′2,...,v′ 10 ],[v′ 13 ,v′ 14 ,...,v′ 22 ],....,[v′ 85 ,v′ 86 ,...,v′ 94 ]].

[0131] S4. Establishment of similarity measurement function: Establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change. The specific process is as follows:

[0132] First, there is a user's average feature sequence Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence

[0133] Secondly, there is the average feature sequence of b users Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence

[0134] The formula for the similarity measurement function between user a and user b is as follows:

[0135]

[0136] in, express express express express The Euclidean distance of the characteristic series of mean, coefficient of variation, skewness and kurtosis was calculated through these four variables. express The DTW distance between two sequences, such as and The DTW distance is, find the sequence [v′ a,1 ,v′ a,2 ,...,v′ a,10 ] and [v′ b,1 ,v′ b,2 ,...,v′ b,10 The purpose of obtaining the DTW distance is to find the set of all curved paths between two time series W = {w1,w2,…,w s ,…,w r} to find the best bending path so that the total bending cost is minimized. The specific calculation is:

[0137]

[0138] Among them, w s is the coordinate of the sth element in the curved path; r is the total number of elements in the path; d(w s ) is w s Corresponding coordinates v′ a,i and v′ b,j The Euclidean distance between for and Specifically, the DTW distance between each segment can be calculated by a dynamic programming method.

[0139] S5. Model establishment and prediction: First, through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, the k-means phase sequence recognition clustering model is established by reasonably setting the initial cluster center; then, through the segmented feature sequence of low-voltage users, the k-means phase sequence recognition clustering model is used to divide the low-voltage users into three categories, and the three categories of low-voltage users are marked with phase labels of A, B, and C based on the similarity measurement between the segmented feature sequence of the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation; finally, the low-voltage users are predicted by the k-means phase sequence recognition clustering model for many consecutive days, and the phase labels that appear multiple times are taken as the final prediction.

[0140] In this embodiment, the process of establishing the k-means phase sequence recognition clustering model is as follows:

[0141] S5.1, select the three-phase voltage segment feature sequence of the substation as the initial cluster center;

[0142] S5.2. Calculate the distance between the three-phase voltage segment feature sequence of the substation and the voltage segment feature sequence of the user using the similarity measurement function. The smaller the distance, the higher the similarity, that is, the low-voltage user is classified into this cluster center.

[0143] S5.3. After processing the data of all low-voltage users, for the low-voltage users in the same cluster, take the average value of their segmented feature sequences as the new cluster center, and recalculate the distance between the segmented feature sequence of the cluster center and the voltage segmented feature sequence of the low-voltage user. Similarly, the smaller the distance, the higher the similarity, that is, the user is classified into this cluster center;

[0144] S5.4. Repeat step S5.3 until the phase categories of all low-voltage users no longer change. Finally, calculate the similarity measure between the cluster center and the segmented feature sequence of the three-phase voltage in the substation, pair the cluster center with the three-phase voltage in the substation one by one, and obtain the k-means phase sequence recognition clustering model.

[0145] The technical solution of the present invention collects the voltage data of the substation and low-voltage users for multiple consecutive days; pre-processes the data to eliminate noise and outliers; adopts the segmented aggregation approximation method to segment all the voltage data, extracts the average value, coefficient of variation, skewness and kurtosis for each segment, and forms a characteristic sequence reflecting the overall distribution; uses the central difference method to process each segmented data to form a characteristic sequence reflecting the local trend change; combines the Euclidean distance reflecting the overall distribution characteristic sequence and the time dynamic bending DTW distance reflecting the local trend change characteristic sequence to establish a similarity measurement function. The similarity of the voltage segmented characteristic sequence is comprehensively determined by the similarity measurement function, and the similarity measurement of the traditional k-means clustering is improved to construct a low-voltage substation phase sequence recognition model. According to the characteristic sequence of the low-voltage user, the user is divided into three categories by using the clustering method, and the phase of the substation to which the low-voltage user belongs is divided into three phases A, B and C by combining the similarity measurement between the class center and the characteristic sequence data of the three-phase voltage of the substation. The low-voltage user is processed by the model for multiple consecutive days, and the phase that appears most times is taken as the final prediction. The present invention generates characteristic sequences based on voltage curve segmentation, combines similarity measurement functions, and utilizes an improved k-means clustering method to significantly improve the accuracy of a low-voltage station phase sequence recognition method.

[0146] The low-voltage substation phase sequence identification method and system based on voltage curve segmentation clustering of the present invention are further introduced below through specific embodiments.

[0147] like Figure 1 As shown, a low voltage area phase sequence identification method based on voltage curve segmentation clustering provided by an embodiment of the present invention specifically includes the following steps:

[0148] S1. Data acquisition: Obtain the continuous multi-day voltage data of 50 groups of substations and their 4936 low-voltage users, including 3087 three-phase low-voltage users and 1849 single-phase low-voltage users, so there are a total of 11110 single-phase voltage data, including 3711 A-phase voltages, 3699 B-phase voltages, and 3700 C-phase voltages.

[0149] S2. Data preprocessing: Delete irrelevant data and duplicate data in the data set, check the special identifiers in the data set to identify missing values: such as NaN, NULL, NA, etc., use the median absolute deviation method to identify outliers, and use Newton interpolation to fill in the identified missing values ​​and outliers.

[0150] S3, Segmented feature sequence generation: I. Segment the voltage curve data, fix the number of data in each segment, and form a segmented data set. II. Extract the average value for each segment to form an average feature sequence. III. Extract the coefficient of variation for each segment to form a coefficient of variation feature sequence. IV. Extract the skewness for each segment to form a skewness feature sequence. V. Extract the kurtosis for each segment to form a kurtosis feature sequence. VI. Use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes.

[0151] S4. Establishment of similarity measurement function: Establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence reflecting the local trend change.

[0152] S5. Model establishment and prediction: For the segmented feature sequences of each substation and its low-voltage users, the idea of ​​k-means algorithm is used, combined with the similarity measurement function, and by setting the three-phase voltage segmented feature sequence of the substation as the initial clustering center, an improved k-means phase sequence recognition clustering model is established. The voltage segmented feature sequence of low-voltage users is divided into three categories using the improved k-means phase sequence recognition clustering model mentioned above. Combined with the similarity measurement between the final class center and the feature sequence data of the three-phase voltage of the substation, the categories are labeled with three-phase labels A, B, and C. The voltage segmented feature sequence of low-voltage users is processed by this model for many consecutive days, and the phase labels that appear multiple times are taken as the final prediction.

[0153] Finally, of the original 3711 A-phase voltages, 3654 were predicted to be A-phase, 43 were predicted to be B-phase, and 14 were predicted to be C-phase; of the original 3699 B-phase voltages, 21 were predicted to be A-phase, 3657 were predicted to be B-phase, and 21 were predicted to be C-phase; of the original 3700 C-phase voltages, 17 were predicted to be A-phase, 20 were predicted to be B-phase, and 3663 were predicted to be C-phase. After calculation, the average precision of the model on each label in the test set is 98.78%, the average recall rate is 98.78%, and the average F1 score is 98.78%. For details, see Figure 2 and Figure 3 It can be seen that the phase recognition model of this model has high accuracy. The t-SNE dimension reduction display diagram of the low-voltage user segmentation feature data model classification in one area is shown in Figure 4 .

[0154] Example 2

[0155] like Figure 5 As shown, a low voltage area phase sequence identification system based on voltage curve segment feature clustering of the present invention comprises:

[0156] The data acquisition module is used to obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, and the continuous single-phase voltage data of low-voltage single-phase users for multiple days from the power consumption information collection system to form a voltage curve data set;

[0157] A data preprocessing module is used to preprocess the voltage curve data set, eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and abnormal values;

[0158] The segmented feature sequence generation module is used to segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segmented data set; to extract the average value of each segment to form an average value feature sequence; to extract the coefficient of variation of each segment to form a coefficient of variation feature sequence; to extract the skewness of each segment to form a skewness feature sequence; to extract the kurtosis of each segment to form a kurtosis feature sequence; to use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes;

[0159] A similarity measurement function establishment module is used to establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change;

[0160] The model building and prediction module is used to establish a k-means phase sequence recognition clustering model through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, by reasonably setting the initial clustering center; it is used to divide low-voltage users into three categories using the k-means phase sequence recognition clustering model through the segmented feature sequence of low-voltage users, and combine the similarity measurement between the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation to label the three categories of low-voltage users with phase labels of A, B, and C; it is also used to predict low-voltage users through the k-means phase sequence recognition clustering model for multiple consecutive days, and take the phase labels that appear multiple times as the final prediction.

[0161] In an optional embodiment, the phase sequence identification method of low-voltage substations based on segmented clustering of voltage curves includes: a) obtaining voltage curve data of substations and low-voltage users in the power consumption information collection system; b) performing data cleaning and value supplementation on the obtained user data; c) segmenting the voltage curve data, extracting the mean value, coefficient of variation, skewness and kurtosis feature sequences, and forming a central difference sequence using the central difference method; d) establishing a similarity measurement function by calculating the Euclidean distance of the mean value, coefficient of variation, skewness and kurtosis feature sequences, combined with the time dynamic bending DTW distance of the central difference feature sequence; e) establishing an improved k-means phase sequence identification clustering model by reasonably setting the initial clustering center through the idea of ​​the k-means algorithm, combining the segmented feature sequence and the similarity measurement function. Finally, the low-voltage users are labeled with the phase labels of A, B, and C phases respectively through the model and multi-day prediction results.

Claims

1. A low voltage phase sequence identification method based on voltage curve segmentation feature clustering, characterized in that: include: Data acquisition: From the electricity consumption information collection system, obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, as well as the continuous single-phase voltage data of low-voltage single-phase users for multiple days to form a voltage curve data set; Data preprocessing: preprocess the voltage curve data set to eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and outliers; Segment feature sequence generation: segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segment data set; extract the average value of each segment to form an average feature sequence; Extract the coefficient of variation for each segment to form a characteristic sequence of coefficient of variation; Extract skewness from each segment to form a skewness feature sequence; extract kurtosis from each segment to form a kurtosis feature sequence; use the central difference method for each segment to form a central difference feature sequence set that reflects local trend changes; Establishment of similarity measurement function: Establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change; Establish a similarity measurement function, including: Get the average feature sequence of user a Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence Get the average feature sequence of user b Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence The formula for the similarity measurement function between user a and user b is as follows: in, express express express express The Euclidean distance of the characteristic series of mean, coefficient of variation, skewness and kurtosis was calculated through these four variables; express The DTW distance between two sequences is calculated based on the set of all curved paths between two time series. s ,…,w r } to find the best bending path so that the total bending cost is minimized. The specific calculation is: Among them, w s is the coordinate of the sth element in the curved path, r is the total number of elements in the path, d(w s ) is w s Corresponding coordinates v ′ a,i With v ′ b,j The Euclidean distance between for and The DTW distance between each segment is obtained by dynamic programming method; Model establishment and prediction: First, through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, the k-means phase sequence recognition clustering model is established by reasonably setting the initial cluster center; then, through the segmented feature sequence of low-voltage users, the k-means phase sequence recognition clustering model is used to divide low-voltage users into three categories, and the three categories of low-voltage users are marked with phase labels of A, B, and C based on the similarity measurement between the segmented feature sequence of the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation; finally, the low-voltage users are predicted by the k-means phase sequence recognition clustering model for many consecutive days, and the phase labels that appear multiple times are taken as the final prediction.

2. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: The data preprocessing process is as follows: Delete irrelevant and duplicate data in the voltage curve dataset, and use the function in Python to identify the special identifiers NaN, NULL, and NA in the dataset to identify missing values; Calculate the median absolute deviation of the voltage curve data set and use the median absolute deviation method to identify outliers. The calculation formula is as follows: MAD=median(|v0-median(V)|,|v1-median(V)|,...,|v 95 -median(V)|) Among them, MAD represents the median absolute deviation of the voltage curve data set; median represents the median; v0, v1...v 95 represents each numerical point in the voltage curve; V represents the voltage curve data set; therefore, |v i -median(V)|>3×MAD Any numerical point satisfies the above v i All of them were identified as outliers; The Newton interpolation method is used to fill the identified missing values ​​and outliers. The formula of the Newton interpolation method is as follows: Where P(x) represents the polynomial function obtained by Newton interpolation method, which is used to estimate the interpolation result at a given position x of the measured data point; v0 represents the initial voltage value point; x j Indicates the position of the jth voltage value point; Δ k v0 represents the k-order difference quotient of v0, and the difference quotient is defined as follows: D k v i =D k-1 v i+1 -D k-1 v i Δ 1 v i =v i+1 -v i Among them, Δ k-1 v i+1 Indicates v i+1 k-1 order difference quotient; v i represents the voltage value at the i-th position; v i+1 Indicates the voltage value at the i+1th position.

3. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: The average value is extracted for each segment to form an average feature sequence, including: Calculate the average value of each segment sequence in the segmented data set to form an average feature sequence in, represents [v0,...,v 11 ], and so on.

4. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: Extract the coefficient of variation for each segment to form a characteristic sequence of coefficient of variation, including: Calculate the coefficient of variation value for each segment sequence in the segmented data set to form a characteristic sequence of the coefficient of variation in, represents [v0,...,v 11 The specific calculation process of the coefficient of variation is as follows: First, calculate the average value of each segment of data Then, calculate the standard deviation S of each segment of data i , the formula is as follows: Among them, v j represents the voltage value of the jth point, and n represents the length of the segmented data; Finally, according to the standard deviation S i With the average Calculate the coefficient of variation for each segment of data The calculation formula is as follows:

5. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 4 is characterized in that: The skewness is extracted for each segment to form a skewness feature sequence, including: Calculate the skewness value for each segment sequence in the segmented data set to form a skewness feature sequence in, represents [v0,...,v 11 ]; the specific calculation process of skewness is as follows: First, calculate the average value of each segment of data Then, find the median of each segment of data (V i ); Finally, according to the average value of each segment of data The median of each data segment (V i ) and the standard deviation S of each segment of data i , calculate the skewness characteristics of each segment of data The calculation formula is as follows:

6. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: Extract the kurtosis of each segment to form a kurtosis feature sequence, including: Calculate the kurtosis of each segmented sequence in the segmented data set to form a kurtosis characteristic sequence in, represents [v0,...,v 11 The specific calculation process of kurtosis is as follows: First, find the fourth power of the standard deviation of each segment of data. The calculation formula is as follows: Then, based on the average value of each segment of data Find the fourth-order center distance of each segment of data The calculation formula is as follows: Finally, according to the fourth-order center distance of each segment of data and the fourth power of the standard deviation of each segment of data Calculate the kurtosis characteristic of each segment of data The calculation formula is as follows:

7. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: The central difference method is used for each segment to form a central difference feature sequence set that reflects the local trend changes, including: Based on the segmented dataset [v0,...,v 11 ], the formula for obtaining the central difference is as follows: Among them, f(x+h) and f(xh) are the values ​​at a distance h to the right and left of point x; according to the above formula, when h is equal to 1, the central differential feature sequence of the segmented data set is obtained in, Finally, the above operations are performed on all segmented data sets to obtain the central differential feature sequence set reflecting the local trend changes. The details are as follows: [[v1 ′ ,v2 ′ ,...,v1 ′ 0],[v1 ′ 3,v1 ′ 4,...,v2 ′ 2],....,[v ′ 85 ,v ′ 86 ,...,v9 ′ 4]]。 8. The low voltage substation phase sequence identification method based on voltage curve segment feature clustering according to claim 1 is characterized in that: The process of establishing the k-means phase sequence recognition clustering model is as follows: The three-phase voltage segment feature sequence of the substation is selected as the initial cluster center; The distance between the three-phase voltage segment feature sequence of the substation and the voltage segment feature sequence of the user is calculated using the similarity measurement function. The smaller the distance, the higher the similarity, that is, the low-voltage user is classified into this cluster center. After processing the data of all low-voltage users, for the low-voltage users in the same cluster, take the average value of their segmented feature sequences as the new cluster center, and recalculate the distance between the segmented feature sequence of the cluster center and the voltage segmented feature sequence of the low-voltage user. Similarly, the smaller the distance, the higher the similarity, that is, the user is classified into this cluster center; Repeat the previous step until the phase category of all low-voltage users no longer changes. Finally, calculate the similarity measure between the cluster center and the segmented feature sequence of the three-phase voltage in the substation area, pair the cluster center with the three-phase voltage in the substation area one by one, and obtain the k-means phase sequence recognition clustering model.

9. A low voltage area phase sequence identification system based on voltage curve segment feature clustering, characterized in that: include: The data acquisition module is used to obtain the continuous three-phase voltage data of the substation and low-voltage three-phase users for multiple days, and the continuous single-phase voltage data of low-voltage single-phase users for multiple days from the power consumption information collection system to form a voltage curve data set; A data preprocessing module is used to preprocess the voltage curve data set, eliminate errors, redundancy, and noise in the voltage curve data set, and identify and process missing values ​​and abnormal values; The segmented feature sequence generation module is used to segment the preprocessed voltage curve data set, fix the number of data in each segment, and form a segmented data set; it is used to extract the average value of each segment to form an average feature sequence; Used to extract the coefficient of variation for each segment and form a characteristic sequence of the coefficient of variation; It is used to extract the skewness of each segment to form a skewness feature sequence; it is used to extract the kurtosis of each segment to form a kurtosis feature sequence; it is used to use the central difference method for each segment to form a central difference feature sequence set that reflects the local trend change; A similarity measurement function establishment module is used to establish a similarity measurement function by calculating the Euclidean distance of the feature sequence reflecting the mean value, coefficient of variation, skewness and kurtosis of the overall distribution, combined with the time dynamic bending DTW distance of the central difference feature sequence set reflecting the local trend change; Establish a similarity measurement function, including: Get the average feature sequence of user a Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence Get the average feature sequence of user b Coefficient of variation characteristic sequence Skewness characteristic sequence Kurtosis characteristic sequence And the central difference characteristic sequence The formula for the similarity measurement function between user a and user b is as follows: in, express express express express The Euclidean distance of the characteristic series of mean, coefficient of variation, skewness and kurtosis was calculated through these four variables; express The DTW distance between two sequences is calculated based on the set of all curved paths between two time series. s ,…,w r } to find the best bending path so that the total bending cost is minimized. The specific calculation is: Among them, w s is the coordinate of the sth element in the curved path, r is the total number of elements in the path, d(w s ) is w s Corresponding coordinates v ′ a,i With v ′ b,j The Euclidean distance between for and The DTW distance between each segment is obtained by dynamic programming method; The model building and prediction module is used to establish a k-means phase sequence recognition clustering model through the k-means clustering algorithm, combined with the segmented feature sequence and the similarity measurement function, by reasonably setting the initial clustering center; it is used to divide low-voltage users into three categories using the k-means phase sequence recognition clustering model through the segmented feature sequence of low-voltage users, and combine the similarity measurement between the cluster center to which the low-voltage users finally belong and the segmented feature sequence data of the three-phase voltage of the substation to label the three categories of low-voltage users with phase labels of A, B, and C; it is also used to predict low-voltage users through the k-means phase sequence recognition clustering model for multiple consecutive days, and take the phase labels that appear multiple times as the final prediction.

Images