Shaplet-based power distribution operation data zoning method and device

Abstract

The invention discloses a Shaplet-based power distribution operation data zoning method and device, and the method comprises the steps: obtaining power distribution operation data, carrying out the dimension reduction fitting processing, obtaining a target operation sequence, carrying out the Shaplet traversal of the target operation sequence, obtaining a Shaplet feature, and enabling the Shaplet feature to be a subsequence of the target operation sequence; based on each Shape feature, determining a distinction degree reflecting the Shape feature; and based on the distinction degree, determining a shaplet set, giving nodes for dividing the power distribution operation data, and realizing zoning of the power distribution operation data. The power distribution operation data is processed into a target operation sequence, so that the data processing difficulty is reduced; and traversing the target operation sequence Shaplet, determining nodes of data division according to the distinction degree of the Shaplet characteristics, completing division of the power distribution operation data, and realizing interpretable division of the multi-dimensional data in the power distribution network.

Description

Technical Field

[0001] The present invention belongs to the technical field of data mining, and in particular relates to a distribution operation data zoning method based on shapelet. Background Art

[0002] With the advancement of smart grids and the widespread application of IoT technology, distribution networks have generated massive amounts of multi-source heterogeneous data, which covers information from thermal power, renewable energy generation, energy storage systems, and multiple energy sources. The operating data of distribution networks has the characteristics of high frequency, large volume, and multi-dimensionality, including but not limited to grid load, power generation, equipment status, environmental parameters, user behavior, etc.

[0003] Multidimensional time series data is an inevitable product of the intelligent and digital transformation of distribution networks. Multidimensional time series data refers to a data set that records multiple related variables simultaneously in a time series, and there are complex interdependencies and influence relationships between them. Effectively mining and analyzing this data is of great significance for improving the operating efficiency of distribution networks, optimizing resource allocation, predicting system failures, and improving the safety and reliability of power grids. How to achieve interpretable division of multidimensional data in distribution networks is a problem that needs to be solved at present.

[0004] Patent CN116930739A discloses a shapelet disconnector fault diagnosis method based on an automatic optimization mechanism. The method takes into account the motor current frequency conversion timing characteristics and fundamental wave timing characteristics, and obtains the operating torque and motor current relationship model by analyzing the relationship between the disconnector operating lever and the motor current. At the same time, under different faults, the sub-timing of the fault feature timing is obtained by analyzing the frequency conversion timing and fundamental wave timing characteristics; under the fault diagnosis model of the self-optimization mechanism, the sub-timing characteristics are continuously optimized and learned to obtain the most discriminative sub-timing to determine the fault type and perform rapid fault diagnosis.

[0005] In addition, patent CN114881077A discloses a voltage sag source classification method based on time series trajectory features, and the method is specifically as follows: after preprocessing the original data set, randomly sample some voltage sag effective values ​​as data set X, and the rest as data set Y; use the FLAG algorithm to extract several special subsequence shapelets with the most category recognition ability from the labeled sag data to form a shapelet set; based on the learned shapelets, shapelet conversion is performed on data sets X and Y to obtain training set X' and test set Y'; use the training set X' to train the random forest to obtain a "white box" voltage sag classification model with good classification performance; use the trained random forest classifier to classify the test set Y' to obtain the classification result; fully consider the time series trajectory characteristics, improve the classification quality and the interpretability of the classification model, and can accurately identify various voltage sag sources in the power system. At the same time, an explainable classification basis is provided, which makes up for the defects of the "black box" classification model in the research related to voltage sag source identification.

[0006] The two solutions disclosed above consider the data division of one dimension of current-related data or voltage-related data, and do not involve the division of multi-dimensional data in the distribution network. Therefore, how to achieve interpretable division of multi-dimensional data in the distribution network, improve the operating efficiency of the distribution network, optimize resource allocation, predict system failures, and improve the safety and reliability of the power grid are the current problems that need to be solved. Summary of the invention

[0007] In view of the defects existing in the above-mentioned prior art, the present invention provides a distribution operation data zoning method and device based on shapelet, the method comprising acquiring distribution operation data, performing dimensionality reduction fitting processing, and obtaining a target operation sequence, wherein the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network; performing shapelet traversal on the target operation sequence to obtain shapelet features, and the shapelet features are subsequences of the target operation sequence; based on each shapelet feature, determining the discrimination reflecting the shapelet feature; based on the discrimination, determining the shapelet set, providing nodes for dividing the distribution operation data, and realizing the zoning of the distribution operation data.

[0008] The present invention obtains the target operation sequence by processing the distribution operation data, thereby reducing the difficulty of data processing; by performing shapelet traversal on the target operation sequence, the shapelet feature is extracted, and the nodes for data partitioning are determined according to the discrimination of the shapelet feature, the partition of the distribution operation data is completed, and the interpretable partition of multidimensional data in the distribution network is realized.

[0009] A distribution operation data zoning method based on shapelet, comprising:

[0010] Obtain distribution operation data, perform dimensionality reduction fitting processing, and obtain a target operation sequence, wherein the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network;

[0011] Perform shapelet traversal on the target running sequence to obtain shapelet features, which are subsequences of the target running sequence;

[0012] Based on each shapelet feature, determining the discrimination degree reflecting the shapelet feature;

[0013] Based on the discrimination, the shapelet set is determined, and the nodes for dividing the distribution operation data are given to realize the division of the distribution operation data.

[0014] It can be seen from the above technical scheme that the target operation sequence is obtained by performing dimensionality reduction fitting processing on the distribution operation data, which reduces the difficulty of data partitioning; by performing shapelet traversal on the target operation sequence, the shapelet features are extracted, and the nodes for data partitioning are determined according to the discrimination of the shapelet features, the partition of the distribution operation data is completed, and the interpretable partitioning of multidimensional data in the distribution network is realized.

[0015] In a possible implementation, power distribution operation data is acquired, dimension reduction fitting is performed, and a target operation sequence is obtained, including:

[0016] Use the first fitting algorithm and the second fitting algorithm to fit the distribution operation data respectively to obtain a first fitting sequence and a second fitting sequence;

[0017] Determine a first fitting distance between the first fitting sequence and the power distribution operation data, and a second fitting distance between the second fitting sequence and the power distribution operation data, wherein the first fitting distance is used to reflect the fitting effect of the first fitting sequence, and the second fitting distance is used to reflect the fitting effect of the second fitting sequence;

[0018] The first fitting distance and the second fitting distance are compared, and a target running sequence is determined from the first fitting sequence and the second fitting sequence.

[0019] In a possible implementation, the first fitting algorithm is to divide the power distribution operation data into intervals and perform analysis and processing by using a preset compression ratio to form a first fitting sequence;

[0020] The first fitting sequence is determined by:

[0021]

[0022] Among them, s′ i The distribution operation data S = (s1, s2, ..., s m ) The sequence after dimensionality reduction S'=(s'1,s'2,...,s' w ), and w≤m,n k =m / w,n k is the compression ratio of data dimensionality reduction.

[0023] In a possible implementation, the second fitting algorithm is a fitting algorithm integrating distribution operation data trends;

[0024] The second fitting sequence is determined by:

[0025] Set the objective function:

[0026]

[0027] Where x=[x1,x2,…,x n ] T ∈R n is the distribution operation data, w=[w1,w2,…,w n ] T is the fitting parameter, L() is the objective function, f() is the second fitting function, k is the total amount of data of y, and x, w, y satisfy the function y = f(x, w);

[0028] According to the value of the fitting parameter when the objective function obtains the minimum value, a second fitting function is given and a second fitting sequence is determined.

[0029] In a possible implementation, the first fitting distance and / or the second fitting distance are determined by:

[0030]

[0031] d(a i ,b j )=|a i -b j | 2

[0032] Among them, D(A i ,B j ) is the superposition of the fitting distances between the first i data in sequence A and the first j data in sequence B, D(A i ,B j-1 ) is the superposition of the fitting distances between the first i data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j-1) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j ) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j data in sequence B, d(a i ,b j ) is the fitting distance between ai in sequence A and bj in sequence B, sequence A = {a1, a2, ..., a t} is the distribution operation data, sequence B = {b1, b2, ..., b p} is the first fitting sequence or the second fitting sequence.

[0033] In a possible implementation, determining the discrimination degree reflecting the shapelet feature based on each shapelet feature includes:

[0034] According to the shapelet feature, the target run sequence in the same dimension is divided into sets T a and the set T b , where dist(S,T a )<λdist(S,T b ), λ is the preset parameter, Where dist() is the distance function, S is the shapelet feature, T is the target run sequence, J is the number of shapelet features in the target run sequence, a is the starting position of the shapelet feature in the target run sequence, b is the ending position of the shapelet feature in the target run sequence, S (a , b) S (a,b) is a shapelet feature with a starting position a and an ending position b, S q is the shapelet feature with the starting position q, T j+q-1 Run the target sequence with the starting position (j+q-1);

[0035] According to the distance between the target running sequence and the shapelet feature, the discrimination degree reflecting the shapelet feature is determined.

[0036] In a possible implementation, determining the discrimination degree reflecting the shapelet feature according to the distance between the target running sequence and the shapelet feature includes:

[0037] Sort the distances between the target run sequence and the shapelet feature in ascending order to obtain the orderline set;

[0038] According to the distinguishing point dt in the orderline set, the orderline set is divided into set Oa and set O b ;

[0039] According to the set O a The mean, standard deviation and set O b The mean and standard deviation are used to determine the discrimination.

[0040] In a possible implementation manner, the discrimination degree is specifically expressed as:

[0041] Gap=μ b -σ b -(μ a +σ a )

[0042] Among them, μ a For the set O a The average value, μ b For the set O b The average value, σ a For the set O a The standard deviation, σ b For the set O b The standard deviation of .

[0043] In a possible implementation, based on the discrimination, a shapelet set is determined, nodes for dividing the power distribution operation data are given, and the division of the power distribution operation data is achieved, including:

[0044] Traverse the orderline collection of each dimension separately;

[0045] Take the points in the orderline set as the distinguishing points dt, and determine the degree of distinction corresponding to the distinguishing points dt;

[0046] Add the distinguishing point dt corresponding to the distinguishing degree within the preset range to the shapelet set;

[0047] The distribution operation data is divided according to the intersection of shapelet sets in each dimension.

[0048] The present invention also discloses a shapelet-based distribution operation data partitioning device, comprising:

[0049] The data acquisition module is used to acquire the distribution operation data, perform dimensionality reduction fitting processing, and obtain the target operation sequence, wherein the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network;

[0050] The feature extraction module is used to perform shapelet traversal on the target running sequence to obtain shapelet features, which are subsequences of the target running sequence;

[0051] A discrimination determination module, for determining the discrimination reflecting the shapelet feature based on each shapelet feature;

[0052] The data partitioning module is used to determine the shapelet set based on the discrimination, provide the nodes for partitioning the power distribution operation data, and realize the partitioning of the power distribution operation data.

[0053] The shapelet-based distribution operation data partitioning method and device provided by the present invention have at least the following beneficial effects:

[0054] The target operation sequence is obtained by performing dimensionality reduction fitting processing on the distribution operation data, which reduces the complexity of data analysis for data partitioning. The shapelet features of the target operation sequence are extracted by shapelet traversal, and the nodes for data partitioning are determined according to the discrimination of the shapelet features. The partition of the distribution operation data is completed, and the interpretable partition of multidimensional data in the distribution network is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of a method for zoning power distribution operation data based on shapelet provided in an embodiment of the present invention;

[0056] Figure 2 A flowchart of determining a target operation sequence provided by an embodiment of the present invention;

[0057] Figure 3 A flowchart for determining a first fitting sequence provided by an embodiment of the present invention;

[0058] Figure 4 A flowchart for determining a multi-dimensional shapelet set provided by an embodiment of the present invention;

[0059] Figure 5 A schematic diagram of a decision tree provided by an embodiment of the present invention;

[0060] Figure 6 A schematic diagram of some effects of power distribution operation data zoning provided by an embodiment of the present invention;

[0061] Figure 7 A structural diagram of a shapelet-based power distribution operation data zoning device provided in an embodiment of the present invention.

[0062] In the figure, 201 is a data acquisition module; 202 is a feature extraction module; 203 is a discrimination determination module; 204 is a data partitioning module. DETAILED DESCRIPTION

[0063] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0064] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0065] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0066] Multidimensional time series data refers to a data set that records multiple related variables simultaneously in a time series, and there are complex interdependencies and influence relationships between them. Shapelet is a feature extraction technology. The core of the technology is to find the most representative subsequence for classification, and it plays a prominent role in the field of time series data mining. The discovery of Shapelet is crucial for tasks such as time series classification, clustering, and anomaly detection.

[0067] The current prior art has a shapelet disconnector fault diagnosis method with an automatic optimization mechanism. This method takes into account the frequency conversion timing characteristics and fundamental wave timing characteristics of the motor current. By analyzing the relationship between the disconnector operating rod and the motor current, the operating torque and motor current relationship model is obtained. At the same time, under different faults, the sub-sequences of the fault feature timing are obtained through the frequency conversion timing and fundamental wave timing characteristics; under the fault diagnosis model of the self-optimization mechanism, the sub-sequence characteristics are continuously optimized and learned to obtain the most discriminative sub-sequence to determine the fault type and perform rapid fault diagnosis. This solution uses shapelet to analyze the relationship between the disconnector operating rod and the motor current, which serves as the basis for fault judgment. However, this method uses the shapelet method to analyze the relevant data of the current dimension, and does not involve the analysis of data between multiple dimensions.

[0068] In addition, a voltage sag source classification method based on time series trajectory features is specifically as follows: after preprocessing the original data set, randomly sample some voltage sag effective values ​​as data set X, and the rest as data set Y; use the FLAG algorithm to extract several special subsequence shapelets with the most category recognition ability from the labeled sag data to form a shapelet set. Based on the learned shapelet, shapelet conversion is performed on data sets X and Y to obtain training set X' and test set Y'; the training set X' is used to train the random forest to obtain a "white box" voltage sag classification model with good classification performance; the trained random forest classifier is used to classify the test set Y' to obtain the classification result; the time series trajectory features are fully considered to improve the classification quality and the interpretability of the classification model, and various voltage sag sources in the power system can be accurately identified. At the same time, an interpretable classification basis is provided, which makes up for the defects of the "black box" classification model in the research related to voltage sag source identification. This scheme uses shapelet to analyze the correlation of voltage sag sources, considers the time series trajectory features, improves the classification quality and the interpretability of the classification model, and provides a data basis for the classification model. Similarly, this method does not involve the analysis of data between multiple dimensions. Therefore, how to achieve interpretable division of multi-dimensional data in the distribution network, improve the operating efficiency of the distribution network, optimize resource allocation, predict system failures, and improve the safety and reliability of the power grid are the current problems that need to be solved.

[0069] Compared with shapelets, traditional feature extraction methods often rely on global statistical features or complex distance metrics, such as dynamic time bending, but these methods face challenges in efficiency and interpretability when processing large-scale data sets. In order to overcome these challenges, the present invention provides a distribution operation data zoning method based on shapelets. This method first performs dimensionality reduction fitting on the distribution operation data to reduce the complexity of the data, and then performs shapelet traversal and generation to identify the shapelet features in the data. By counting and fusing the shapelet features of multidimensional data, the multidimensional features contained in the data are coordinated, and a representative sequence containing the similarity of multidimensional time series features is established. Based on the obtained sequence, a shapelet subset is screened. Then, by evaluating the discrimination of each shapelet in the subset, the most representative shapelet features are screened, and a decision tree is established based on the shapelet features, thereby achieving an interpretable division of the distribution operation data.

[0070] like Figure 1 As shown, the present invention provides a distribution operation data zoning method based on shapelet, and the specific steps are as follows:

[0071] S101: Obtain distribution operation data, perform dimensionality reduction fitting processing, and obtain a target operation sequence.

[0072] Specifically, the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network, and the target operation sequence has the same trend characteristics as the distribution operation data. In a specific example, the above-mentioned multiple dimensions include photovoltaic power generation dimension, distribution network load dimension, wind power generation dimension and market electricity price dimension. In other implementations, the dimensions may also include thermal power generation dimension, new energy power generation dimension, energy storage system dimension, etc.

[0073] Reference Figure 2 ,The specific process of determining the target operation sequence is as follows:

[0074] Further, the distribution operation data is fitted using the first fitting algorithm and the second fitting algorithm respectively to obtain the first fitting sequence and the second fitting sequence. The first fitting algorithm is to divide the distribution operation data into intervals and analyze the data by a preset compression ratio to form a first fitting sequence. The second fitting algorithm is a fitting algorithm that integrates the trend of the distribution operation data. In a specific embodiment, the first fitting algorithm is a piecewise aggregate approximation (PAA) algorithm, which can achieve time series dimensionality reduction and feature extraction. The piecewise aggregate approximation algorithm has high computational efficiency and can reduce the complexity of time series operations. It is suitable for processing time series data with long time spans and obvious trend characteristics. The PAA algorithm divides the time series into equal intervals and uses the mean of each segment to represent the original sequence to achieve dimensionality reduction. The second fitting algorithm is a trend feature based piecewise aggregate approximation (TFPAA). The PAA method uses the mean of each segment to represent the original time series, but in some cases, the original time series cannot be effectively fitted by only the single data representation of the mean. If the mean values ​​of two time series are similar but the trend characteristics are obviously different, the PAA method will determine that the two sequence segments are extremely similar after dimensionality reduction. Therefore, only using the original time series mean for dimensionality reduction may cause the original time series trend characteristics to be lost, and the fluctuation characteristics of the voltage curve cannot be effectively fitted. When using the PAA method, it should be noted that when using the PAA algorithm to approximate the distribution operation data, the appropriate compression ratio should be selected by comprehensively considering the degree of fitting of the distribution operation data and the amount of calculation. When the compression ratio is too large, it is difficult to effectively extract the fluctuation characteristics of the distribution operation data; when the compression ratio is set too small, although the distribution operation data after approximate representation has a higher accuracy and can retain more original sequence information, the calculation time will also increase.

[0075] In order to accurately measure the fluctuation of power distribution operation data, the present invention also adopts the TFPAA algorithm as the second fitting algorithm, using the mean and slope indicators to retain the trend characteristics to the greatest extent.

[0076] Reference Figure 3 , the first fitting sequence mentioned above is determined by:

[0077] Step 1: Standardize the power distribution operation data to obtain standard power distribution data;

[0078] Step 2: Determine the compression ratio n of the standard power distribution data k ;

[0079] Step 3: According to the compression ratio n k The standard power distribution data is divided into equal intervals, and the mean of each segment is used to represent the standard power distribution data, thereby obtaining the first fitting sequence.

[0080] The above first fitting sequence is specifically expressed as:

[0081]

[0082] Among them, s′ i The distribution operation data S = (s1, s2, ..., s m ) The sequence after dimensionality reduction S'=(s'1,s'2,...,s' w ), and w≤m,n k =m / w,n k is the compression ratio of data dimensionality reduction.

[0083] The second fitting sequence is determined as follows:

[0084] Set the objective function:

[0085] Where x=[x1,x2,...,x n ] T ∈R n is the distribution operation data, w=[w1,w2,...,w n ] T is the fitting parameter, L() is the objective function, f() is the second fitting function, k is the total amount of data of y, x, w, y satisfy the function y = f(x, w), the second fitting function is determined by the fitting parameter, and the second fitting parameter can be a linear function, a quadratic function, a polynomial, etc., without limitation.

[0086] According to the value of the fitting parameter when the objective function obtains the minimum value, a second fitting function is given and a second fitting sequence is determined.

[0087] Before using the TFPAA algorithm to fit the distribution operation data, the distribution operation data needs to be standardized so that the distribution operation data obeys a standard normal distribution with a mean of 0 and a variance of 1.

[0088] In order to compare the approximate representation effects of different algorithms on distribution operation data, it is necessary to select a suitable distance measurement function. The time series similarity measurement method usually uses distance measurement functions such as Euclidean distance, Manhattan distance and dynamic time warping distance to compare the similarity of each sequence.

[0089] The Euclidean distance (ED) is faster to calculate, but can only be used to calculate the distance between sequences of the same length. The dynamic time warping (DTW) algorithm takes a long time to calculate, but can be used to calculate sequences of unequal lengths, with more accurate results and a wider range of applications. The principle of the DTW algorithm is to find the best alignment path between sequences of unequal lengths so that the distance cost between the two sequences is minimized.

[0090] In order to evaluate the fitting effect of the first fitting sequence and the second fitting sequence on the distribution operation data, the DTW algorithm is used in this example for evaluation. The DTW algorithm can realize the similarity measurement between sequences with inconsistent time intervals and different lengths, and has been widely used in the field of time series similarity. The DTW algorithm uses the idea of ​​dynamic programming to correspond sequence elements with different time intervals and lengths, find an optimal curved path, and make the two sequences along the path achieve the best match in shape, solving the problem of matching feature points of time series of different lengths. The DTW distance is used to measure the similarity between the distribution operation data and the first fitting sequence and the second fitting sequence obtained after dimensionality reduction. The smaller the DTW distance, the higher the degree of similarity.

[0091] Further, a first fitting distance between the first fitting sequence and the power distribution operation data and a second fitting distance between the second fitting sequence and the power distribution operation data are determined, wherein the first fitting distance is used to reflect the fitting effect of the first fitting sequence and the second fitting distance is used to reflect the fitting effect of the second fitting sequence. The first fitting distance and the second fitting distance are compared, and a target operation sequence is determined from the first fitting sequence and the second fitting sequence.

[0092] The first fitting distance and / or the second fitting distance are determined in the following manner:

[0093]

[0094] d(a i ,b j )=|a i -b j | 2

[0095] Among them, D(A i ,B j ) is the superposition of the fitting distances between the first i data in sequence A and the first j data in sequence B, D(A i ,B j-1 ) is the superposition of the fitting distances between the first i data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j-1 ) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j ) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j data in sequence B, d(a i ,b j ) is a in sequence A i With sequence B in b j The fitting distance between the sequence A = {a1, a2, ..., a t} is the distribution operation data, sequence B = {b1, b2, ..., b p} is the first fitting sequence or the second fitting sequence.

[0096] S102: Perform shapelet traversal on the target running sequence to obtain shapelet features.

[0097] Specifically, the shapelet feature is a subsequence of the target running sequence. A subsequence of a corresponding length of each time series object T in the target running sequence is intercepted as a candidate shapelet set. For example, if there is a time series object T of [1,1,2,2,3,3,5], then the candidate shapelet set includes [1,1,2], [1,2,2], ..., [1,1,2,2], [1,2,2,3], [2,2,3,3,5], that is, the candidate shapelet set is a set composed of any subsequence in the time series object T. Traverse the subsequences in the candidate shapelet set to find the optimal shapelet. In a specific embodiment, the optimal shapelet feature can be found in the candidate shapelet set by information gain. In other embodiments, the length of the shapelet feature can be set according to actual needs to improve the efficiency of shapelet traversal. The length of the shapelet feature is not specifically limited here.

[0098] S103: Based on each shapelet feature, determine the discrimination degree reflecting the shapelet feature.

[0099] Specifically, according to the shapelet feature, the target run sequence in the same dimension is divided into sets Ta and the set T b , where dist(S,T a )<λdist(S,T b ), λ is the preset parameter,

[0100] Where dist() is the distance function, S is the shapelet feature, T is the target run sequence, J is the number of shapelet features in the target run sequence, a is the starting position of the shapelet feature in the target run sequence, b is the ending position of the shapelet feature in the target run sequence, S (a,b) is a shapelet feature with a starting position a and an ending position b, S q is the shapelet feature with the starting position q, T j+q-1 is the target running sequence with the starting position (j+q-1). In a specific example, λ can be 1 / 10, 1 / 100, or 1 / 1000. The value of λ is determined by the user's selection of the set T. a and the set T b The user requirement set T is related to the differentiated needs of a and the set T b The greater the difference between them, the smaller the value of λ. Conversely, the user requires the set T a and the set T b The smaller the difference between them, the larger the value of λ. a is the set of target run sequences that the shapelet feature can represent, T b Run the collection of sequences for targets that are significantly discriminative by shapelet features.

[0101] Furthermore, according to the distance between the target running sequence and the shapelet feature, the discrimination degree reflecting the shapelet feature is determined, and the specific contents are as follows:

[0102] Sort the distances between the target run sequence and the shapelet feature in ascending order to obtain the orderline set. According to the distinguishing point dt in the orderline set, the orderline set is divided into set O a and set O b According to the set O a The mean, standard deviation and set O b The mean and standard deviation are used to determine the discrimination.

[0103] Specifically, the discrimination is expressed as:

[0104] Gap=μ b -σ b -(μa +σ a )

[0105] Among them, μ a For the set O a The average value, μ b For the set O b The average value, σ a For the set O a The standard deviation, σ b For the set O b The standard deviation of .

[0106] The meaning of discrimination is: Assume that the set O a is the set that shapelet features can represent, and the set O b For all except set O a The set of elements other than the elements in a The characteristics of the elements in set O should be smaller average value, more concentrated data, and lower volatility. b The elements in the set should be more dispersed, with a significantly higher mean and standard deviation than those in the set O. a Therefore, the larger the gap value corresponding to the shapelet feature is, the stronger the ability of the shapelet feature to distinguish time series is.

[0107] S104: Based on the discrimination, determine the shapelet set, provide nodes for dividing the power distribution operation data, and realize the division of the power distribution operation data.

[0108] Reference Figure 4 Specifically, the orderline sets of each dimension are traversed respectively, and the points in the orderline set are used as the distinguishing points dt, and the discrimination corresponding to the distinguishing points dt is determined. The distinguishing points dt corresponding to the discrimination within the preset range are added to the shapelet set. According to the intersection of the shapelet sets in each dimension, the distribution operation data is zoned. The above preset range is set according to the actual requirements of the user for the discrimination, and is not limited here.

[0109] Furthermore, after obtaining the intersection of the shapelet sets in each dimension, the discrimination degree of each distinguishing point in the intersection is determined, and the discrimination degree is the average of the discrimination degrees of the distinguishing points in each dimension. The distinguishing points in the intersection are sorted in descending order according to the discrimination degree. Then, the distinguishing points in the intersection are obtained in turn, and a decision tree is constructed to realize the zoning of the distribution operation data.

[0110] In a specific implementation, in a multidimensional time series data curve, the time corresponding to the shapelet features of N dimensions is the same, and the distinguishing point dt should also be divided at the same point. In a specific example, time is decomposed into 356 time points, and each shapelet corresponds to 365 distances in the same dimension. The set of distances constitutes the orderline set corresponding to the shapelet. Due to the discrimination corresponding to each distinguishing point dt in the orderline set in each dimension, the intersection of the orderline sets in multiple dimensions is taken, and the distinguishing point dt is traversed in the intersection, the discrimination of each distinguishing point is calculated, and a decision tree is constructed. By finding the intersection, the number of calculations for traversing the distinguishing point dt can be significantly reduced. For a set of multidimensional shapelets, the average value of the discrimination in each dimension is taken as the final discrimination.

[0111] In another specific example, there are four dimensions A1, A2, A3, and A4. The shapelet set corresponding to A1 is [1,3,5,7,9,11], the shapelet set corresponding to A2 is [2,3,5,6,9,11], the shapelet set corresponding to A3 is [1,3,5,6,9,10], and the shapelet set corresponding to A4 is [1,3,5,6,8,11]. The intersection of the four shapelet sets is [3,5], and the distinguishing point 3 corresponds to The discrimination degree of the distinguishing point 5 is (a1+a2+a3+a4) / 4, and the discrimination degree of the distinguishing point 5 is (b1+b2+b3+b4) / 4, where a1 is the discrimination degree of the distinguishing point 3 in the A1 dimension, a2 is the discrimination degree of the distinguishing point 3 in the A2 dimension, a3 is the discrimination degree of the distinguishing point 3 in the A3 dimension, a4 is the discrimination degree of the distinguishing point 3 in the A4 dimension, b1 is the discrimination degree of the distinguishing point 5 in the A1 dimension, b2 is the discrimination degree of the distinguishing point 5 in the A2 dimension, b3 is the discrimination degree of the distinguishing point 5 in the A3 dimension, and b4 is the discrimination degree of the distinguishing point 5 in the A4 dimension. According to the discrimination degree of the distinguishing point 3 and the distinguishing point 5, a decision tree is constructed to realize the division of the distribution operation data.

[0112] In the above implementation, by traversing the orderline set of each dimension respectively and calculating the discrimination, the distribution of the power distribution operation data is completed. It can be understood that the distinguishing point dt is selected from the orderline set according to the discrimination, and the distribution of the distinguishing point dt will be in the middle part of the orderline set, but not at both ends of the orderline set. Therefore, in order to improve the computing efficiency, in other implementations, the orderline set is further screened, and after removing part of the data at both ends of the orderline set of each dimension, the new orderline set is traversed. In a specific example, 10% of the data at both ends of the orderline set is removed, and only 80% of the values ​​in the middle of the orderline set are considered as candidate data for the distinguishing point dt.

[0113] In a specific example, a decision tree is established based on the obtained shapelet set, and the distribution operation data is interpretably divided according to the decision tree. Each element in the shapelet set is used as a division criterion to divide all the distribution operation data into two categories. The elements with the highest discrimination are selected from the shapelet set as the distinction criteria. For example, 365 typical days in a year are divided. The typical day in the power system is also called the typical load calculation day. The determination of the typical day is to calculate the line loss of each node in the power system, calculate the peak and valley time of each day, and understand the distribution of load in the power system. The determination of the typical day is based on the division of winter and summer power production into thermal power and hydropower, generally in the middle and end of the month in winter and summer.

[0114] Two categories will be obtained in the process of calculating the discrimination. After determining the discrimination point dt, all typical days will be divided into two categories based on the distance from the shapelet. In this example, since the time series change characteristics of four dimensions are considered at the same time, it is difficult to explain the sequence change characteristics of a certain dimension when analyzing the shapelet. However, although the complexity of multidimensional time series poses challenges to analyzing the characteristics of a certain dimension alone, the application of shapelets in multidimensional time series can reveal cross-influences and compound effects that are difficult to capture in single-dimensional analysis. In the distribution network analysis, it can be used for wind power, photovoltaics, load demand, and changes in electricity price levels.

[0115] For example, in the scenario corresponding to the shapelet of the first classification, while the load demand increases, the photovoltaic power generation also increases, and the system electricity price rises slightly at this time; in the scenario corresponding to the second shapelet, when the load demand is stable, even if the photovoltaic output drops rapidly, the system electricity price level only fluctuates slightly. From this, it can be inferred that in this power system, the capacity of photovoltaic configuration is small and has not had a significant impact on the system; in the scenario corresponding to the third shapelet, the electricity price and load changes are very stable, but the levels are lower than the first two scenarios, and the wind power output is greater than the first scenario. And so on, select multiple scenarios and classify them according to the distance between the typical day and the typical scenario. Shapelet analysis helps to identify hidden patterns in multidimensional data, so as to more accurately predict system trends or make decisions.

[0116] When using a decision tree for segmentation, set a constraint on the maximum and minimum number of sample segmentations. If no constraint is imposed, 15 shapelets will eventually be traversed, and the classification efficiency after the ninth level will be less than 3%, that is, less than 10 typical days will be obtained in a single segmentation. As the classification level increases, since the new shapelet only classifies unclassified typical days, the subsequent classification efficiency will gradually decrease. Therefore, this example constrains the generation of the decision tree by the minimum number of sample segmentations, stipulating that each node must have at least 20 samples before it can be segmented. Without this constraint, the classification efficiency will obtain less than 10 typical days at the fourth and fifth levels. The last category is the number of typical days remaining after the classification is completed. The final classification results of typical days in a year are shown in Table 1, and the decision tree corresponding to the classification results is shown in Table 1. Figure 5 shown.

[0117] Table 1

[0118] level one two three Four five six seven eight Nine typical day 65 48 37 46 53 60 20 28 16

[0119] In time series analysis, shapelet is a key feature, and its performance at each classification level has a decisive impact on the effectiveness of the entire classification system. On the one hand, when a shapelet classifies a large number of typical days at a certain level, this indicates that the shapelet has a wide range of applicability and can cover and distinguish more data samples. However, improved classification efficiency does not always mean improved classification quality. If a shapelet classifies too many typical days at a certain level, it means that the shapelet has weak discrimination and no outstanding features can be used to accurately identify and distinguish data. Some of the effects of distribution operation data zoning are shown in Figure 2. Figure 6 shown.

[0120] The present invention uses shapelet feature extraction to divide multi-dimensional distribution operation data, so that complex time series data is converted into a form that is easy to understand and explain, intuitively explains the basis for distribution network scene division and the characteristics between each scene, and enhances the interpretability of the data. At the same time, through the interpretable division of multi-dimensional distribution operation data, it provides strong technical support for the intelligent management and decision-making of the distribution network, helps to improve the operating efficiency and safety level of the distribution network, and promotes the intelligent management of distribution network data.

[0121] Reference Figure 7 The embodiment of the present application provides a shapelet-based distribution operation data partitioning device, including:

[0122] The data acquisition module 201 is used to acquire the power distribution operation data, perform dimensionality reduction fitting processing, and obtain the target operation sequence, wherein the power distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network;

[0123] A feature extraction module 202 is used to perform shapelet traversal on the target running sequence to obtain shapelet features, where the shapelet features are subsequences of the target running sequence;

[0124] A discrimination determination module 203, for determining a discrimination reflecting a shapelet feature based on each shapelet feature;

[0125] The data partitioning module 204 is used to determine the shapelet set based on the discrimination, provide nodes for partitioning the power distribution operation data, and realize the partitioning of the power distribution operation data.

[0126] The shapelet-based distribution operation data partitioning method and device provided by the present invention have at least the following beneficial effects:

[0127] The target operation sequence is obtained by performing dimensionality reduction fitting processing on the distribution operation data, which reduces the complexity of data analysis for data partitioning. The shapelet features of the target operation sequence are extracted by shapelet traversal, and the nodes for data partitioning are determined according to the discrimination of the shapelet features. The partition of the distribution operation data is completed, and the interpretable partition of multidimensional data in the distribution network is realized.

[0128] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0129] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0130] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0132] The units involved in the embodiments of the present disclosure may be implemented by software or hardware, wherein the name of a unit does not limit the unit itself in some cases.

[0133] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A distribution operation data partitioning method based on shapelet, characterized in that: include: Obtain distribution operation data, perform dimensionality reduction fitting processing, and obtain a target operation sequence, wherein the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network; Perform shapelet traversal on the target running sequence to obtain shapelet features, which are subsequences of the target running sequence; Based on each shapelet feature, determining the discrimination degree reflecting the shapelet feature; Based on the discrimination, the shapelet set is determined, and the nodes for dividing the distribution operation data are given to realize the division of the distribution operation data.

2. The method for zoning power distribution operation data based on shapelet according to claim 1, characterized in that: The step of acquiring the power distribution operation data and performing dimensionality reduction fitting processing to obtain the target operation sequence includes: Use the first fitting algorithm and the second fitting algorithm to fit the distribution operation data respectively to obtain a first fitting sequence and a second fitting sequence; Determine a first fitting distance between the first fitting sequence and the power distribution operation data, and a second fitting distance between the second fitting sequence and the power distribution operation data, wherein the first fitting distance is used to reflect the fitting effect of the first fitting sequence, and the second fitting distance is used to reflect the fitting effect of the second fitting sequence; The first fitting distance and the second fitting distance are compared, and a target running sequence is determined from the first fitting sequence and the second fitting sequence.

3. The shapelet-based distribution operation data partitioning method according to claim 2, characterized in that: The first fitting algorithm is to divide the distribution operation data into intervals and perform analysis and processing by using a preset compression ratio to form a first fitting sequence; The first fitting sequence is determined by: Among them, s i ' is the distribution operation data S = (s1, s2, ..., s m ) The sequence after dimensionality reduction S'=(s1',s'2,...,s' w ), and w≤m,n k =m / w,n k is the compression ratio of data dimensionality reduction.

4. The method for zoning power distribution operation data based on shapelet according to claim 2, characterized in that: The second fitting algorithm is a fitting algorithm that integrates the distribution operation data trend; The second fitting sequence is determined by: Set the objective function: Where x=[x1,x2,...,x n ] T ∈R n is the distribution operation data, w=[w1,w2,...,w n ] T is the fitting parameter, L() is the objective function, f() is the second fitting function, k is the total amount of data of y, and x, w, y satisfy the function y = f(x, w); According to the value of the fitting parameter when the objective function obtains the minimum value, a second fitting function is given and a second fitting sequence is determined.

5. The method for zoning power distribution operation data based on shapelet according to claim 2, characterized in that: The first fitting distance, and / or the second fitting distance are determined by: d(a i ,b j )=|a i -b j | 2 Among them, D(A i ,B j ) is the superposition of the fitting distances between the first i data in sequence A and the first j data in sequence B, D(A i ,B j-1 ) is the superposition of the fitting distances between the first i data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j-1 ) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j-1 data in sequence B, D(A i-1 ,B j ) is the superposition of the fitting distances between the first i-1 data in sequence A and the first j data in sequence B, d(a i ,b j ) is a in sequence A i With sequence B in b j The fitting distance between the sequence A = {a1, a2, ..., a t } is the distribution operation data, sequence B = {b1, b2, ..., b p } is the first fitting sequence or the second fitting sequence.

6. The method for zoning power distribution operation data based on shapelet according to claim 1, characterized in that: The determining, based on each shapelet feature, the discrimination degree reflecting the shapelet feature comprises: According to the shapelet feature, the target run sequence in the same dimension is divided into sets T a and the set T b , where dist(S,T a )<λdist(S,T b ), λ is the preset parameter, Where dist() is the distance function, S is the shapelet feature, T is the target run sequence, J is the number of shapelet features in the target run sequence, a is the starting position of the shapelet feature in the target run sequence, b is the ending position of the shapelet feature in the target run sequence, S (a,b) is a shapelet feature with a starting position a and an ending position b, S q is the shapelet feature with the starting position q, T j+q-1 Run the target sequence with the starting position (j+q-1); According to the distance between the target running sequence and the shapelet feature, the discrimination degree reflecting the shapelet feature is determined.

7. The method for zoning power distribution operation data based on shapelet according to claim 6, characterized in that: Determining the discrimination degree reflecting the shapelet feature according to the distance between the target running sequence and the shapelet feature includes: Sort the distances between the target run sequence and the shapelet feature in ascending order to obtain the orderline set; According to the distinguishing point dt in the orderline set, the orderline set is divided into set O a and set O b ; According to the set O a The mean, standard deviation and set O b The mean and standard deviation are used to determine the discrimination.

8. The method for zoning power distribution operation data based on shapelet according to claim 7, characterized in that: The specific discrimination is expressed as: Gap=μ b -s b -(m a +s a ) Among them, μ a For the set O a The average value, μ b For the set O b The average value, σ a For the set O a The standard deviation, σ b For the set O b The standard deviation of .

9. The method for zoning power distribution operation data based on shapelet according to claim 7, characterized in that: The method of determining a shapelet set based on the discrimination, providing nodes for dividing the power distribution operation data, and realizing the division of the power distribution operation data includes: Traverse the orderline collection of each dimension separately; Take the points in the orderline set as the distinguishing points dt, and determine the degree of distinction corresponding to the distinguishing points dt; Add the distinguishing point dt corresponding to the distinguishing degree within the preset range to the shapelet set; The distribution operation data is divided according to the intersection of shapelet sets in each dimension.

10. A distribution operation data partitioning device based on shapelet, characterized in that: The method for zoning distribution operation data based on shapelet according to any one of claims 1 to 9 comprises: The data acquisition module is used to acquire the distribution operation data, perform dimensionality reduction fitting processing, and obtain the target operation sequence, wherein the distribution operation data is a multi-dimensional time series data curve generated during the operation of the distribution network; The feature extraction module is used to perform shapelet traversal on the target running sequence to obtain shapelet features, which are subsequences of the target running sequence; A discrimination determination module, for determining the discrimination reflecting the shapelet feature based on each shapelet feature; The data partitioning module is used to determine the shapelet set based on the discrimination, provide the nodes for partitioning the power distribution operation data, and realize the partitioning of the power distribution operation data.

Images