Method and device for dividing operation scene of power distribution system

Abstract

The invention discloses an operation scene division method and device for a power distribution system. The method comprises the following steps: traversing a multi-dimensional time sequence of the power distribution system and generating a multi-dimensional sub-sequence set with distinguishing characteristics; sequentially carrying out symbolization and integration processing on the multi-dimensional sub-sequence set, and giving an evaluation index in combination with a processing result; screening the multi-dimensional sub-sequence set based on the evaluation index and calculating the distinction degree, and giving a multi-dimensional sub-sequence with the maximum distinction degree; and based on a pre-constructed decision tree, processing the multi-dimensional sub-sequence with the maximum distinction degree, and giving a division result of the operation scene of the power distribution system. According to the method, efficient and explainable division of the multi-dimensional data of the power distribution system can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of integrated energy system scene division, and specifically relates to a method and device for dividing operation scenes of a power distribution system. Background Art

[0002] With the advancement of smart grids and the widespread application of Internet of Things technologies, distribution networks have generated massive amounts of multi-source heterogeneous data, which covers information on thermal power, renewable energy generation, energy storage systems, and multiple energy sources. Multidimensional complex scenario classification is a key technology developed under the background of intelligent and digital transformation of distribution networks. 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. Effectively mining and analyzing these 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.

[0003] Multidimensional time series data refers to a data set that records multiple related variables simultaneously in a time series. There are complex interdependencies and influence relationships between them. When mining time series data, traditional 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. For example, the invention patent CN113972694A discloses a distribution network distributed photovoltaic and energy storage power station investment decision-making method, in the field of distribution network distributed power supply planning, specifically including the following steps: Step 1: Input the historical power generation data set and power load historical data set of the photovoltaic power station, and use the improved K center point clustering algorithm to extract the typical photovoltaic output scene and the typical power load scene; Step 2: According to the photovoltaic output timing characteristics of the typical photovoltaic output scene and the power load timing characteristics of the typical power load scene, construct the photovoltaic and load joint timing scene.

[0004] Based on this, how to provide a method for efficient and interpretable partitioning of multi-dimensional data of power distribution systems is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0005] In view of the defects existing in the above-mentioned prior art, the present invention provides a method and device for dividing the operation scenarios of a power distribution system, which can realize efficient and explainable division of multi-dimensional data of the power distribution system.

[0006] In a first aspect, the present invention provides a method for dividing operation scenarios of a power distribution system, comprising:

[0007] Traversing the multidimensional time series of the power distribution system and generating a set of multidimensional subsequences with distinguishing characteristics;

[0008] The multidimensional subsequence set is symbolized and integrated in turn, and evaluation indicators are given based on the processing results;

[0009] Based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given;

[0010] Based on the pre-built decision tree, the multidimensional subsequences with the greatest discrimination are processed to give the division results of the distribution system operation scenarios.

[0011] Furthermore, the multidimensional time series of the power distribution system is traversed and a set of multidimensional subsequences with distinguishing characteristics is generated, including:

[0012] The multidimensional time series of the distribution system is traversed by the shapelet algorithm, and a set of multidimensional subsequences with distinguishing characteristics is screened in turn.

[0013] Furthermore, the multidimensional subsequence sets are symbolized and integrated in turn, and evaluation indicators are given based on the processing results, including:

[0014] Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set;

[0015] Preprocessing the multi-dimensional symbol sequence set, and comparing the preprocessed symbol sequences of the same dimension, to obtain an integral sequence set corresponding to each dimensional symbol sequence set;

[0016] Based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and the evaluation index is established.

[0017] Furthermore, based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set, including:

[0018] The multidimensional time series is segmented by continuous windows, and the multidimensional time series segments within each continuous window are normalized;

[0019] Perform convolution dimensionality reduction on the normalized multidimensional time series segments;

[0020] Based on the SAX encoding rules, the convolution dimension reduction results are mapped to the corresponding symbols;

[0021] Based on the mapping result and the corresponding relationship between the multidimensional subsequence set and the multidimensional time series, a multidimensional symbol sequence set is given.

[0022] Furthermore, the multi-dimensional symbol sequence set is preprocessed, and the preprocessed symbol sequences of the same dimension are compared to obtain a set of integral sequences corresponding to each dimensional symbol sequence set, including:

[0023] Performing mask processing on each symbol sequence in the multidimensional symbol sequence set;

[0024] The masked symbol sequences in each dimensional symbol sequence set are compared in pairs, and two symbol sequences with the same comparison results are integrated and added by one;

[0025] After the comparison of all dimensional symbol sequence sets is completed, the integral corresponding to each symbol sequence is counted and the final integral sequence set is generated.

[0026] Furthermore, based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and evaluation indicators are established, including:

[0027] Calculate the information entropy for each dimensional symbol sequence set, and calculate the variance for each dimensional integral sequence set;

[0028] Based on the calculated information entropy and variance, the weight coefficient of each subsequence in the corresponding multidimensional subsequence set is determined.

[0029] Furthermore, based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given, including:

[0030] Based on the weight threshold and the weight coefficient of each subsequence in the multidimensional subsequence set, the multidimensional subsequence set is screened to provide a screened multidimensional subsequence set;

[0031] Based on the filtered multi-dimensional subsequence set, the distances with the time series of the corresponding dimension are calculated respectively, and the distance set corresponding to each dimensional subsequence set arranged according to the distance size is given;

[0032] Selecting a distance subset of a predetermined proportion extending from a central dividing line to both sides from each dimension distance set;

[0033] Based on each element in each dimension distance subset as a distinguishing point, the discrimination degree of the corresponding dimension distance set is calculated;

[0034] Compare the discrimination corresponding to each element in each dimensional distance subset and give the multidimensional subsequence with the maximum discrimination.

[0035] Furthermore, based on each element in each dimensional distance subset as a distinguishing point, the discrimination degree of the corresponding dimensional distance set is calculated, including:

[0036] The elements in each dimension distance set that are smaller than the distinguishing point distance and the elements corresponding to the distinguishing point are divided into the first set, and the remaining elements are divided into the second set;

[0037] Calculate the mean and standard deviation for both the first and second sets;

[0038] Based on the mean and standard deviation of the first set and the second set, the discrimination of the corresponding distinguishing points is calculated to satisfy the following relationship:

[0039] Gap=μ b -σ b -(μ a +σ a )

[0040] In the formula, Gap is the discrimination, μ a , μ b are the average values ​​of the first and second sets, σ a , σ b is the standard deviation of the first and second sets.

[0041] Furthermore, the discrimination corresponding to each element in each dimensional distance subset is compared to give a multidimensional subsequence with the greatest discrimination, including:

[0042] Compare the discrimination corresponding to each element in each dimensional distance subset, and give the subsequence corresponding to the element with the largest discrimination in each dimensional distance set;

[0043] Based on the time node of the subsequence corresponding to the element with the greatest discrimination in each dimensional distance set, determine the discrimination of the corresponding elements in other dimensional distance sets;

[0044] Based on the corresponding subsequence of the element with the largest discrimination in each dimensional distance set and the discrimination of the corresponding elements in other dimensional distance sets of the corresponding time node, the average discrimination of the corresponding time node is given;

[0045] Compare multiple average discriminations and give the multidimensional subsequence corresponding to the time node with the largest average discrimination.

[0046] Furthermore, the pre-built decision trees include:

[0047] Take multiple groups of multidimensional historical time series as training sets, and take the multidimensional historical subsequences with the largest degree of distinction corresponding to each group of multidimensional historical time series as nodes;

[0048] Recursion is performed based on the nodes corresponding to the training set until all the stopping conditions are met, and the constructed decision tree is given.

[0049] In a second aspect, the present invention provides an operation scene division device for a power distribution system, which adopts the operation scene division method for the power distribution system, comprising:

[0050] A traversal generation module, which is used to traverse the multidimensional time series of the power distribution system and generate a set of multidimensional subsequences with distinguishing characteristics;

[0051] An index determination module is used to symbolize and integrate the multidimensional subsequence set in turn, and provide evaluation indicators based on the processing results;

[0052] A sequence determination module is used to screen the multidimensional subsequence set based on the evaluation index and calculate the discrimination degree, and give the multidimensional subsequence with the greatest discrimination degree;

[0053] The scenario division module is used to process the multidimensional subsequences with the greatest discrimination based on the pre-built decision tree and provide the division results of the distribution system operation scenarios.

[0054] Furthermore, the traversal generation module includes:

[0055] The multidimensional time series of the distribution system is traversed by the shapelet algorithm, and a set of multidimensional subsequences with distinguishing characteristics is screened in turn.

[0056] Furthermore, the indicator determination module includes:

[0057] Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set;

[0058] Preprocessing the multi-dimensional symbol sequence set, and comparing the preprocessed symbol sequences of the same dimension, to obtain an integral sequence set corresponding to each dimensional symbol sequence set;

[0059] Based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and the evaluation index is established.

[0060] Furthermore, the indicator determination module includes:

[0061] The multidimensional time series is segmented by continuous windows, and the multidimensional time series segments within each continuous window are normalized;

[0062] Perform convolution dimensionality reduction on the normalized multidimensional time series segments;

[0063] Based on the SAX encoding rules, the convolution dimension reduction results are mapped to the corresponding symbols;

[0064] Based on the mapping result and the corresponding relationship between the multidimensional subsequence set and the multidimensional time series, a multidimensional symbol sequence set is given.

[0065] Furthermore, the indicator determination module includes:

[0066] Performing mask processing on each symbol sequence in the multidimensional symbol sequence set;

[0067] The masked symbol sequences in each dimensional symbol sequence set are compared in pairs, and two symbol sequences with the same comparison results are integrated and added by one;

[0068] After the comparison of all dimensional symbol sequence sets is completed, the integral corresponding to each symbol sequence is counted and the final integral sequence set is generated.

[0069] Furthermore, the indicator determination module also includes:

[0070] Calculate the information entropy for each dimensional symbol sequence set, and calculate the variance for each dimensional integral sequence set;

[0071] Based on the calculated information entropy and variance, the weight coefficient of each subsequence in the corresponding multidimensional subsequence set is determined.

[0072] Furthermore, the sequence determination module includes:

[0073] Based on the weight threshold and the weight coefficient of each subsequence in the multidimensional subsequence set, the multidimensional subsequence set is screened to provide a screened multidimensional subsequence set;

[0074] Based on the filtered multi-dimensional subsequence set, the distances with the time series of the corresponding dimension are calculated respectively, and the distance set corresponding to each dimensional subsequence set arranged according to the distance size is given;

[0075] Selecting a distance subset of a predetermined proportion extending from a central dividing line to both sides from each dimension distance set;

[0076] Based on each element in each dimension distance subset as a distinguishing point, the discrimination degree of the corresponding dimension distance set is calculated;

[0077] Compare the discrimination corresponding to each element in each dimensional distance subset and give the multidimensional subsequence with the maximum discrimination.

[0078] Furthermore, the sequence determination module includes:

[0079] The elements in each dimension distance set that are smaller than the distinguishing point distance and the elements corresponding to the distinguishing point are divided into the first set, and the remaining elements are divided into the second set;

[0080] Calculate the mean and standard deviation for both the first and second sets;

[0081] Based on the mean values ​​and standard deviations of the first set and the second set, the discrimination degree of the corresponding discrimination points is calculated.

[0082] Furthermore, the sequence determination module also includes:

[0083] Compare the discrimination corresponding to each element in each dimensional distance subset, and give the subsequence corresponding to the element with the largest discrimination in each dimensional distance set;

[0084] Based on the time node of the subsequence corresponding to the element with the greatest discrimination in each dimensional distance set, determine the discrimination of the corresponding elements in other dimensional distance sets;

[0085] Based on the corresponding subsequence of the element with the largest discrimination in each dimensional distance set and the discrimination of the corresponding elements in other dimensional distance sets of the corresponding time node, the average discrimination of the corresponding time node is given;

[0086] Compare multiple average discriminations and give the multidimensional subsequence corresponding to the time node with the largest average discrimination.

[0087] Furthermore, the scene division module includes:

[0088] Use the multidimensional subsequence with the highest discrimination as the node attribute to build a decision tree model, and set the maximum depth and minimum number of samples;

[0089] Based on the constructed decision tree model, the multidimensional time series of the distribution system is processed to give the division results of the distribution system operation scenarios.

[0090] The present invention provides a method and device for dividing operation scenarios of a power distribution system, which at least have the following beneficial effects:

[0091] (1) By generating a set of multidimensional subsequences with distinguishing characteristics, symbolizing them in turn and giving a set of integral sequences, the continuous time series can be converted into a discrete symbol sequence, simplifying the calculation of statistical similarity in the scene and reducing the impact of noise; at the same time, by using the integral sequence to determine the evaluation index, the subsequence candidate set can be obtained in advance, thereby reducing the amount of calculation when calculating the discrimination and improving the efficiency of data processing. In addition, based on the multidimensional subsequences determined by the discrimination, the construction of a decision tree can be realized, and finally the efficient and interpretable division of the multidimensional data of the distribution system can be achieved, which can provide strong technical support for the intelligent management and decision-making of the distribution system and help improve the operating efficiency and safety level of the distribution system.

[0092] (2) By calculating the frequency of occurrence of each symbol sequence, data features are captured from multiple perspectives, and cross-dimensional co-occurrence patterns and correlations are identified. By finding symbol sequences that appear repeatedly in the same dimension, these repeated patterns often contain the inherent connections and laws of the data. On the basis of multidimensional time series, multidimensional integral sequences are constructed, forming a higher-level data representation.

[0093] (3) Based on the multidimensional integral sequence, the sequence variance is used to quantify the degree of difference between sequences, and the uncertainty of the sequence is measured by calculating the information entropy of the symbol sequence. The two statistics are combined to realize the construction of multidimensional data evaluation indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 A flow chart of a method for dividing operation scenarios of a power distribution system provided by the present invention;

[0095] Figure 2 A flowchart providing evaluation indicators for an embodiment of the present invention;

[0096] Figure 3 A flowchart of obtaining a multi-dimensional symbol sequence set provided by an embodiment of the present invention;

[0097] Figure 4 A flowchart of a set of integral sequences provided for an embodiment of the present invention;

[0098] Figure 5 A flowchart for determining evaluation indicators provided for an embodiment of the present invention;

[0099] Figure 6 A flowchart of providing a multidimensional subsequence is provided for an embodiment of the present invention;

[0100] Figure 7 A flowchart of providing a multidimensional subsequence with the greatest degree of discrimination provided by an embodiment of the present invention;

[0101] Figure 8 A schematic diagram of a symbol sequence and a subsequence provided for an embodiment of the present invention;

[0102] Fig. 9 A schematic diagram of obtaining a score sequence provided by an embodiment of the present invention;

[0103] Fig.10 A schematic diagram of an operating scenario division device for a power distribution system provided by the present invention. DETAILED DESCRIPTION

[0104] 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 creative work are within the scope of protection of the present invention.

[0105] 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.

[0106] 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.

[0107] like Figure 1 As shown, the present invention provides a method for dividing operation scenarios of a power distribution system, comprising:

[0108] Traversing the multidimensional time series of the power distribution system and generating a set of multidimensional subsequences with distinguishing characteristics;

[0109] The multidimensional subsequence set is symbolized and integrated in turn, and evaluation indicators are given based on the processing results;

[0110] Based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given;

[0111] Based on the pre-built decision tree, the multidimensional subsequences with the greatest discrimination are processed to give the division results of the distribution system operation scenarios.

[0112] The multidimensional time series of the power distribution system is traversed and a set of multidimensional subsequences with distinguishing characteristics is generated, including:

[0113] The multidimensional time series of the distribution system is traversed by the shapelet algorithm, and a set of multidimensional subsequences with distinguishing characteristics is screened in turn.

[0114] like Figure 2 As shown, the multidimensional subsequence set is symbolized and integrated in turn, and evaluation indicators are given based on the processing results, including:

[0115] Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set;

[0116] Preprocessing the multi-dimensional symbol sequence set, and comparing the preprocessed symbol sequences of the same dimension, to obtain an integral sequence set corresponding to each dimensional symbol sequence set;

[0117] Based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and the evaluation index is established.

[0118] like Figure 3 As shown, based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set, including:

[0119] The multidimensional time series is segmented by continuous windows, and the multidimensional time series segments within each continuous window are normalized;

[0120] Perform convolution dimensionality reduction on the normalized multidimensional time series segments;

[0121] Based on the SAX encoding rules, the convolution dimension reduction results are mapped to the corresponding symbols;

[0122] Based on the mapping result and the corresponding relationship between the multidimensional subsequence set and the multidimensional time series, a multidimensional symbol sequence set is given.

[0123] like Figure 4 As shown, the multi-dimensional symbol sequence set is preprocessed, and the preprocessed symbol sequences of the same dimension are compared to give the integral sequence set corresponding to each dimensional symbol sequence set, including:

[0124] Performing mask processing on each symbol sequence in the multidimensional symbol sequence set;

[0125] The masked symbol sequences in each dimensional symbol sequence set are compared in pairs, and two symbol sequences with the same comparison results are integrated and added by one;

[0126] After the comparison of all dimensional symbol sequence sets is completed, the integral corresponding to each symbol sequence is counted and the final integral sequence set is generated.

[0127] like Figure 5 As shown, based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and evaluation indicators are established, including:

[0128] Calculate the information entropy for each dimensional symbol sequence set, and calculate the variance for each dimensional integral sequence set;

[0129] Based on the calculated information entropy and variance, the weight coefficient of each subsequence in the corresponding multidimensional subsequence set is determined.

[0130] like Figure 6 As shown, based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given, including:

[0131] Based on the weight threshold and the weight coefficient of each subsequence in the multidimensional subsequence set, the multidimensional subsequence set is screened to provide a screened multidimensional subsequence set;

[0132] Based on the filtered multidimensional subsequence set, the distance dist(S,T) of the time series of the corresponding dimension is calculated respectively, and the distance set corresponding to each dimensional subsequence set arranged according to the distance size is given;

[0133] Selecting a distance subset of a predetermined proportion extending from a central dividing line to both sides from each dimension distance set;

[0134] Based on each element in each dimension distance subset as a distinguishing point, the discrimination degree of the corresponding dimension distance set is calculated;

[0135] Compare the discrimination corresponding to each element in each dimensional distance subset and give the multidimensional subsequence with the maximum discrimination.

[0136] Based on each element in each dimension distance subset as a distinguishing point, the discrimination of the corresponding dimension distance set is calculated, including:

[0137] The elements in each dimension distance set that are smaller than the distinguishing point distance and the elements corresponding to the distinguishing point are divided into the first set, and the remaining elements are divided into the second set;

[0138] Calculate the mean and standard deviation for both the first and second sets;

[0139] Based on the mean and standard deviation of the first set and the second set, the discrimination of the corresponding distinguishing points is calculated to satisfy the following relationship:

[0140] Gap=μ b -σ b -(μ a +σ a )

[0141] In the formula, Gap is the discrimination, μ a , μ b are the average values ​​of the first and second sets, σ a , σ b is the standard deviation of the first and second sets.

[0142] like Figure 7 As shown, the discrimination corresponding to each element in each dimensional distance subset is compared, and the multidimensional subsequence with the largest discrimination is given, including:

[0143] Compare the discrimination corresponding to each element in each dimensional distance subset, and give the subsequence corresponding to the element with the largest discrimination in each dimensional distance set;

[0144] Based on the time node of the subsequence corresponding to the element with the greatest discrimination in each dimensional distance set, determine the discrimination of the corresponding elements in other dimensional distance sets;

[0145] Based on the corresponding subsequence of the element with the largest discrimination in each dimensional distance set and the discrimination of the corresponding elements in other dimensional distance sets of the corresponding time node, the average discrimination of the corresponding time node is given;

[0146] Compare multiple average discriminations and give the multidimensional subsequence corresponding to the time node with the largest average discrimination.

[0147] Pre-built decision trees, including:

[0148] Multiple groups of multidimensional historical time series are used as training sets, and the multidimensional historical subsequences with the greatest degree of discrimination corresponding to each group of multidimensional historical time series are used as nodes; wherein the training set also includes actual classification results of operation scenarios corresponding to the multidimensional historical time series;

[0149] Recursion is performed based on the nodes corresponding to the training set until all the stopping conditions are met, and the constructed decision tree is given.

[0150] After taking the multidimensional historical subsequence with the greatest discrimination corresponding to each group of multidimensional historical time series as a node, the subnodes are generated by splitting the node into two, and the process is continuously recursive until all samples corresponding to the training set in the node are of one category or the attribute set formed by the multidimensional historical subsequence with the greatest discrimination is zero, then the construction of the decision tree is completed;

[0151] Decision trees are prone to overfitting during their establishment, and pruning can be used to reduce the structure and size of the tree;

[0152] Specifically, a random sampling method is used to extract 10% of the samples (multidimensional historical time series) as the test set, and the rest are used as the training set. A decision tree model is established based on the training set data, and a cross-validation method is used in parameter selection;

[0153] The GINI value is used to measure the purity of the node during the classification process. The more impure the node, the larger the GINI value. The calculation formula is as follows:

[0154] GINI=1-∑P i 2

[0155] Where P i is the proportion of samples belonging to the i-th category in the node.

[0156] Sample variance is used to calculate node purity during the regression process. The greater the purity, the better the regression effect. The regression variance calculation formula is as follows:

[0157]

[0158] Where σ is the regression variance, xi is the i-th sample, n is the total number of samples, and μ is the mean of the samples.

[0159] Select the attribute that minimizes the GINI value or regression variance of the child node as the splitting solution;

[0160] At the same time, the CCP (cost complexity) pruning method is used to select the node surface error rate gain value to evaluate the pruning effect. The calculation formula of the surface error rate gain value is as follows:

[0161]

[0162] Where R(t) represents the error cost with t as the root node, that is, judging whether node t should be pruned, R(t) = r(t)*p(t), r(t) is the error rate of the node, p(t) is the proportion of the node data volume, and N(T) represents the number of subtree nodes.

[0163] R(T) represents the error cost of the leaf node with t as the root node, satisfying the following relationship:

[0164]

[0165] where r i (t) is the error rate of the child node, p i (t) represents the data node ratio of node i.

[0166] Based on the pre-built decision tree, the multidimensional subsequences with the highest degree of discrimination are processed to give the division results of the distribution system operation scenarios. Since each dimension has its own shapelet (the subsequence with the highest degree of discrimination) in the multidimensional time series, these shapelets can capture the key patterns and features in their respective dimensions; by analyzing the shapelets on each dimension, the relationships and influences between different dimensions can be identified; in the decision tree, each node represents a shapelet, and each branch represents a classification decision based on the shapelet; in this way, the decision tree can reflect how different shapelets (i.e., key features of different dimensions) jointly affect the final classification results; each shapelet is initially extracted independently from its own dimension, but in the decision tree, they together form a multidimensional classification framework that can reflect the relationship between data of different dimensions, and the final classification decision is based on the comprehensive consideration of all relevant shapelets.

[0167] The present invention converts the continuous time series into a discrete symbol sequence through SAX changes, simplifies the calculation amount of statistical shapelet similarity in the scene, and reduces the impact of noise. At the same time, the integral sequence is used for multi-dimensional data fusion to obtain the shapelet candidate set in advance, reduce the calculation amount when calculating the shapelet discrimination, and improve the efficiency of data processing.

[0168] The multidimensional time series is divided based on shapelet discrimination, so that the complex time series data is converted into a form that is easy to understand and interpret. The basis for the distribution network scenario division and the direct characteristics of each scenario are intuitively explained, enhancing the interpretability of the data.

[0169] The present invention provides strong technical support for intelligent management and decision-making of distribution networks by realizing interpretable partitioning of multi-dimensional power data, which helps to improve the operating efficiency and safety level of distribution networks and promote intelligent management.

[0170] In an actual application scenario, the present invention provides a method for dividing operation scenarios of a power distribution system, comprising:

[0171] Step 1: Shapelet traversal and generation based on complex scenarios of power distribution system;

[0172] Step 2: Transform the shapelet obtained in step 1 into a symbolic sequence through the symbolic approximation (SAX) technology;

[0173] Step 3: Based on the symbol sequence, count the similarity information contained in the symbol sequence in each dimension and establish a multidimensional integral sequence;

[0174] Step 4: Establish evaluation indicators based on sequence variance and information entropy in the multidimensional integral sequence to perform multidimensional data fusion;

[0175] Step 5: Introduce discrimination as the shapelet selection criterion and construct a multidimensional shapelet evaluation system;

[0176] Step 6: Establish a decision tree based on the obtained shapelet set, and interpretably divide the multidimensional complex scenarios of the distribution system according to the decision tree.

[0177] Among them, the original power data is traversed and shapelets are screened in turn. Shapelets represent unique fragments of system behavior under specific conditions. As a key local pattern in complex distribution network time series data, they contain specific information. The present invention improves the efficiency of data processing based on SAX technology; optimizes multidimensional time series fusion by establishing evaluation indicators by combining sequence variance and information entropy; and uses shapelets to enhance the interpretability of distribution network / distribution system operation scenario division.

[0178] Specifically, a time series is a complete period of data that changes with time, such as a daily photovoltaic power generation curve, a load time series change curve, a wind turbine power curve, etc., which is represented by T. Clustering methods based on time series often treat a single time series as a whole, and evaluate the representative characteristics of the time series based on similarity metrics. The purpose of the shapelet method used in the present invention is to discover fragments with distinguishing characteristics in the time series, which has stronger interpretability. A shapelet is a subsequence of a time series containing specific information, and its length is less than or equal to the original time series, represented by S. The distance between a time series and a certain subsequence is defined as a sliding window assuming the subsequence on the time series. When the window slides, the shortest distance between the two is the distance between the subsequence and the original time series, expressed as dist(S,T). Among them, S q is the qth shapelet, T is the original time series, J is the number of sliding windows, and (a, b) are the starting and ending positions of the shapelet.

[0179]

[0180] The present invention uses symbolic aggregate approXimation (SAX) to convert shapelets, which are originally continuous, numerical time series fragments, into easy-to-process symbol sequences. SAX is a process of converting complex, continuous or abstract information into a set of simplified, recognizable symbols or codes. The SAX method first standardizes the shapelet and then maps it to a limited alphabet using equal-height binning, such as Figure 8 As shown, each numerical interval is assigned a specific symbol. SAX transformation is performed on data of different dimensions to obtain a multi-dimensional SAX symbol sequence.

[0181] Specifically, the symbolic approximation technology converts the time series into a discrete symbol sequence, which is used to simplify data representation and improve the efficiency of data mining and analysis. Each dimension of data adopts a coding method, and each shapelet becomes a symbol sequence after encoding. For each potential shapelet, SAX technology is used for preprocessing. SAX technology segments the time series and assigns a representative symbol to each segment, thereby converting each shapelet into a set of discrete symbols. First, the time series data is normalized, and then convolution is used for dimensionality reduction. In general, the length of the symbol sequence is less than the length of the shapelet, so adjacent fusion and averaging are required to obtain SAX encoding. The present invention uses linear fusion in the form of average values. SAX technology uses continuous windows to segment the actual sequence and takes the average value in each window. The convolved values ​​are mapped to a set of symbols, and different symbols are used to represent different intervals. Using discrete symbols to replace the original data features helps to identify repeated patterns or anomalies in time series, and provides a way to quickly compare similarities between different data.

[0182] The integral sequence set actually captures data features from multiple perspectives by calculating the frequency of occurrence of each symbol sequence, and identifies cross-dimensional co-occurrence patterns and correlations. By finding symbol combinations that appear repeatedly in the same dimension, these repeated patterns often contain the inherent connections and laws of the data. On the basis of multidimensional time series, a multidimensional integral sequence is constructed to form a higher-level data representation.

[0183] Specifically, based on the symbol sequence, the similarity information contained in the symbol sequence in each dimension is counted to establish a multidimensional integral sequence. The symbol sequences represented by SAX are compared two by two. If the symbol sequences of two shapelets are the same, one point is added to each. If a time series contains S symbol sequences, then the one-dimensional data needs to be compared with S 2 times. The present invention uses ζ(N,S) to represent the integral sequence, where N is the dimension of the time series. Each value represents the integral of a shapelet.

[0184] In order to increase the robustness of the algorithm and reduce the impact of random events on data categories, the present invention uses a mask to pre-process the SAX symbol sequence as follows Fig. 9 As shown. First, randomly ignore two bits of encoding, and compare all the remaining SAX symbol sequences with each other. If two symbol sequences are completely consistent, each adds one to the corresponding position of ζ(N, S). According to the accuracy requirements, multiple masking operations can be performed to increase the number of statistics. Considering the computational efficiency, the present invention uses two masks for similarity statistics. A multidimensional integral sequence is established in each dimension, and each bit in the integral sequence represents the unique information contained in a shapelet.

[0185] The present invention is based on multidimensional integral sequence, uses sequence variance to quantify the degree of difference between sequences, and measures the uncertainty of sequences by calculating the information entropy of symbolic sequences, and combines the two statistics to construct a multidimensional data evaluation index. Based on the obtained evaluation index, multidimensional sequences are fused and sorted to screen out candidate shapelet sets. Multidimensional sequence fusion is performed based on integral sequences and weight coefficients to enhance the representativeness of data, reveal deep-level associations and patterns under the operating state of the distribution network, and provide more accurate data support for system state evaluation, fault prediction and optimization control.

[0186] Specifically, MTS fusion is performed based on the obtained evaluation indicators. The goal of MTS fusion is to select a shapelet combination that contains shapelets of the same time period in each dimension, such as the coordinated changes in photovoltaic, load, wind power output, and market electricity prices in a certain area from 13:00 to 18:00. In the MTS fusion process, it is necessary to consider the complex dependencies that may exist between multiple variables, and the characteristics of different variables may differ greatly. The original time series and the difference series are analyzed for similarity to obtain the integral sequence ζ(N,S). Each row in the integral sequence ζ(N,S) represents a one-dimensional time series, such as load demand, regional light intensity, etc. Each column represents the number of repetitions of a set of shapelets, including the repetition information of the original time series and the repetition information in the first-order difference sequence. In the data fusion stage, the weight coefficient of data fusion is given based on the similarity characteristics of the shapelets contained in the integral sequence, combined with information entropy and sequence variance.

[0187] The present invention introduces discrimination as a shapelet selection criterion. The core of discrimination is to quantify the ability of shapelet in identifying and distinguishing different operating states or modes of the system. Specifically, by measuring the distance between the time series and the shapelet, a shapelet with significant discrimination is selected. An ideal shapelet should be very close to some scenes and very far away from other scenes, so that it has good discrimination in time series scenes. The core of discrimination is to quantify the ability of shapelet in identifying and distinguishing different operating states of the system. By measuring the distance between the time series and the shapelet, significantly different scenarios are identified, and the difference values ​​between the scenarios are calculated. Through the difference value calculation, a shapelet with higher discrimination that can better divide the distribution network operation scenarios is selected.

[0188] Specifically, by evaluating the discriminability of the shapelet itself, the effectiveness of the evaluation index obtained by the multidimensional sequence fusion algorithm is evaluated. The premise that a shapelet has discriminability is that it can significantly distinguish the time series into two parts, namely, dist(S, T a )< <dist(S,Tb ). At this time, T a is the set of sequences that shapelet can represent (the first set), T b It is the set that shapelet can distinguish significantly (the second set).

[0189] Representative shapelets need to meet two criteria, namely, universality and uniqueness. Universality means that the shapelet needs to have certain common features with other shapelets, while uniqueness means that the shapelet cannot be completely consistent with other shapelets. Therefore, the criterion for determining a representative shapelet is that the similarity with other shapelets is within a certain range.

[0190] Define orderline as a set of distances between all sequences and shapelet, where the elements are sorted from small to large. Then there is a distinguishing point dt that divides the set into two parts, represented by O a , O b .O a The elements in the shapelet represent the distance between the time series and the shapelet. b The elements in mean the distance between the sequence and the shapelet that can be significantly distinguished by the shapelet.

[0191] The present invention uses gap to describe the temporal differentiation capability of shapelet, satisfying the following relationship:

[0192] Gap=μ b -σ b -(μ a +σ a )

[0193] In the formula, Gap is the discrimination, μ a , μ b are the average values ​​of the first and second sets, σ a , σ b is the standard deviation of the first set and the second set. The meaning of this definition is: Assume that the close set O a is the set that shapelet can represent, then O a The characteristics of O should be smaller average value, more concentrated data, and lower volatility. b The elements of set O should be more dispersed, with a significantly higher mean and standard deviation than those of set O. a Therefore, the goal of shapelet should be to maximize the gap value of orderline and maximize the temporal differentiation ability.

[0194] The core of calculating the gap is to find the best distinguishing point dt. Specifically, the traversal method is used to find the best distinguishing point dt in the orderline. For a one-dimensional time series, traverse all points in the orderline, obtain the means and variances of the two categories in turn, and find dt when the gap is maximum. Considering that the maximized dt is often somewhere in the middle of the orderline, it is possible to traverse the middle part to increase the calculation efficiency. In actual application scenarios, only the middle 80% of the values ​​in the orderline can be considered as potential values ​​of dt.

[0195] In a multidimensional time series, considering that shapelets of N dimensions correspond to the same time, dt should also be divided at the same point. For a group of multidimensional shapelets, a certain proportion of the indexes in the middle of their orderlines are obtained as potential index sets. The orderlines in each dimension have different rankings, so the intersection of the multidimensional sequence orderline index sets is taken, dt is traversed in the intersection, the gap value of each point is calculated, and the best distinguishing point is obtained. This index set acquisition method can significantly reduce the number of calculations for traversing dt. For a group of multidimensional shapelets, the average value of each gap is taken as the distinguishing point evaluation criterion.

[0196] The present invention mines a series of highly discriminative shapelets from historical data, and then uses these shapelets as node attributes to construct a decision tree. In the decision tree, each shapelet is used as a discriminative condition to determine whether a data point conforms to a specific pattern or abnormal state, and each leaf node corresponds to an operating state or category of a distribution network, establishing an interpretable data category division.

[0197] Specifically, a decision tree is established based on the obtained shapelet set, and the multi-dimensional complex scene of the distribution system is interpretably divided according to the decision tree. Each shapelet is used as a division criterion to divide all sequences into two categories. The shapelet with the highest gap value is selected in turn as the distinction criterion. Taking the division of 365 typical days in a year as an example, two categories will be obtained in the process of calculating the gap. After determining the distinguishing point dt, all typical days can be divided into two categories according to the distance from the shapelet. When selecting dt and calculating the gap, the present invention determines that the weight coefficients between multi-dimensional time series are the same. 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 shapelets. However, although the complexity of multi-dimensional time series brings challenges to the analysis of the characteristics of a certain dimension alone, the application of shapelets in multi-dimensional time series can reveal the cross-influence and compound effects that are difficult to capture in single-dimensional analysis. In the distribution network analysis, it can be targeted at wind power, photovoltaics, load demand, and changes in electricity price levels. For example, in the scenario described by the shapelet of the first level 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 described by the shapelet of the second level classification, 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 system, the capacity of photovoltaic configuration is small and has not had a significant impact on the system; in the scenario described by the shapelet of the third level classification, 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 scenario 1 of the first level classification. By analogy, multiple scenarios are selected and classified according to the distance between typical days and typical scenarios. Shapelet analysis helps to identify hidden patterns in multidimensional data, so as to more accurately predict system trends or make decisions.

[0198] The decision tree of the present invention needs to set the maximum and minimum sample segmentation number constraints. If no constraints are imposed, 15 shapelets will eventually be traversed, and the classification efficiency after the ninth layer will be less than 3%, that is, a single segmentation will be less than 10 typical days. As the classification level increases, since the new shapelet only classifies unclassified typical days, the subsequent classification efficiency will gradually decrease. Therefore, the generation of the decision tree is constrained by the minimum sample segmentation number, and it is stipulated that each node must have at least 20 samples before it can be segmented. Without this constraint, the classification efficiency will be less than 10 typical days obtained at the fourth and fifth layers. The last category is the number of typical days remaining after the classification is completed.

[0199] In time series analysis, shapelet is a key feature, and its performance in each classification level has a decisive influence on the effectiveness of the entire classification system. On the one hand, when the number of typical days classified by shapelet at a certain level is large, it shows that the shapelet has a wide range of applicability and can cover and distinguish more data samples. However, the improvement of classification efficiency does not always mean the improvement of classification quality. The excessive number of typical days classified by shapelet at a certain level means that the discrimination of the shapelet is weak, and there are no outstanding features that can be used to accurately identify and distinguish data. Therefore, the maximum number of segmentation constraints is further set, and the typical day classification strategy is finally given after traversing 21 shapelets. The average segmentation efficiency under this strategy is about 10%, which is approximately equal to the median under the classification strategy. It can be used as a segmentation strategy for typical days.

[0200] like Fig.10 As shown, the present invention provides an operation scene division device of a power distribution system, which adopts the operation scene division method of the power distribution system, including:

[0201] A traversal generation module, which is used to traverse the multidimensional time series of the power distribution system and generate a set of multidimensional subsequences with distinguishing characteristics;

[0202] An index determination module is used to symbolize and integrate the multidimensional subsequence set in turn, and provide evaluation indicators based on the processing results;

[0203] A sequence determination module is used to screen the multidimensional subsequence set based on the evaluation index and calculate the discrimination degree, and give the multidimensional subsequence with the greatest discrimination degree;

[0204] The scenario division module is used to process the multidimensional subsequences with the greatest discrimination based on the pre-built decision tree and provide the division results of the distribution system operation scenarios.

[0205] Furthermore, the traversal generation module includes:

[0206] The multidimensional time series of the distribution system is traversed by the shapelet algorithm, and a set of multidimensional subsequences with distinguishing characteristics is screened in turn.

[0207] Furthermore, the indicator determination module includes:

[0208] Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set;

[0209] Preprocessing the multi-dimensional symbol sequence set, and comparing the preprocessed symbol sequences of the same dimension, to obtain an integral sequence set corresponding to each dimensional symbol sequence set;

[0210] Based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and the evaluation index is established.

[0211] Furthermore, the indicator determination module includes:

[0212] The multidimensional time series is segmented by continuous windows, and the multidimensional time series segments within each continuous window are normalized;

[0213] Perform convolution dimensionality reduction on the normalized multidimensional time series segments;

[0214] Based on the SAX encoding rules, the convolution dimension reduction results are mapped to the corresponding symbols;

[0215] Based on the mapping result and the corresponding relationship between the multidimensional subsequence set and the multidimensional time series, a multidimensional symbol sequence set is given.

[0216] Furthermore, the indicator determination module includes:

[0217] Performing mask processing on each symbol sequence in the multidimensional symbol sequence set;

[0218] The masked symbol sequences in each dimensional symbol sequence set are compared in pairs, and two symbol sequences with the same comparison results are integrated and added by one;

[0219] After the comparison of all dimensional symbol sequence sets is completed, the integral corresponding to each symbol sequence is counted and the final integral sequence set is generated.

[0220] Furthermore, the indicator determination module also includes:

[0221] Calculate the information entropy for each dimensional symbol sequence set, and calculate the variance for each dimensional integral sequence set;

[0222] Based on the calculated information entropy and variance, the weight coefficient of each subsequence in the corresponding multidimensional subsequence set is determined.

[0223] Furthermore, the sequence determination module includes:

[0224] Based on the weight threshold and the weight coefficient of each subsequence in the multidimensional subsequence set, the multidimensional subsequence set is screened to provide a screened multidimensional subsequence set;

[0225] Based on the filtered multi-dimensional subsequence set, the distances with the time series of the corresponding dimension are calculated respectively, and the distance set corresponding to each dimensional subsequence set arranged according to the distance size is given;

[0226] Selecting a distance subset of a predetermined proportion extending from a central dividing line to both sides from each dimension distance set;

[0227] Based on each element in each dimension distance subset as a distinguishing point, the discrimination degree of the corresponding dimension distance set is calculated;

[0228] Compare the discrimination corresponding to each element in each dimensional distance subset and give the multidimensional subsequence with the maximum discrimination.

[0229] Furthermore, the sequence determination module includes:

[0230] The elements in each dimension distance set that are smaller than the distinguishing point distance and the elements corresponding to the distinguishing point are divided into the first set, and the remaining elements are divided into the second set;

[0231] Calculate the mean and standard deviation for both the first and second sets;

[0232] Based on the mean values ​​and standard deviations of the first set and the second set, the discrimination degree of the corresponding discrimination points is calculated.

[0233] Furthermore, the sequence determination module also includes:

[0234] Compare the discrimination corresponding to each element in each dimensional distance subset, and give the subsequence corresponding to the element with the largest discrimination in each dimensional distance set;

[0235] Based on the time node of the subsequence corresponding to the element with the greatest discrimination in each dimensional distance set, determine the discrimination of the corresponding elements in other dimensional distance sets;

[0236] Based on the corresponding subsequence of the element with the largest discrimination in each dimensional distance set and the discrimination of the corresponding elements in other dimensional distance sets of the corresponding time node, the average discrimination of the corresponding time node is given;

[0237] Compare multiple average discriminations and give the multidimensional subsequence corresponding to the time node with the largest average discrimination.

[0238] Furthermore, the scene division module includes:

[0239] Use the multidimensional subsequence with the highest discrimination as the node attribute to build a decision tree model, and set the maximum depth and minimum number of samples;

[0240] Based on the constructed decision tree model, the multidimensional time series of the distribution system is processed to give the division results of the distribution system operation scenarios.

[0241] 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 method for dividing operation scenarios of a power distribution system, characterized in that: include: Traversing the multidimensional time series of the power distribution system and generating a set of multidimensional subsequences with distinguishing characteristics; The multidimensional subsequence set is symbolized and integrated in turn, and evaluation indicators are given based on the processing results; Based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given; Based on the pre-built decision tree, the multidimensional subsequences with the greatest discrimination are processed to give the division results of the distribution system operation scenarios.

2. The method for dividing the operation scenarios of the power distribution system according to claim 1, characterized in that: The multidimensional subsequence sets are symbolized and integrated in turn, and evaluation indicators are given based on the processing results, including: Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set; Preprocessing the multi-dimensional symbol sequence set, and comparing the preprocessed symbol sequences of the same dimension to obtain an integral sequence set corresponding to each dimensional symbol sequence set; Based on a set of multidimensional symbol sequences and a set of multidimensional integral sequences, the sequence variance and information entropy are determined and evaluation indicators are established.

3. The method for dividing the operation scenarios of the power distribution system according to claim 2, characterized in that: Based on the multidimensional time series, the multidimensional subsequence set is symbolized to obtain a multidimensional symbol sequence set, including: The multidimensional time series is segmented by continuous windows, and the multidimensional time series segments within each continuous window are normalized; Perform convolution dimensionality reduction on the normalized multidimensional time series segments; Based on the SAX encoding rules, the convolution dimension reduction results are mapped to the corresponding symbols; Based on the mapping result and the corresponding relationship between the multidimensional subsequence set and the multidimensional time series, a multidimensional symbol sequence set is given.

4. The method for dividing the operation scenarios of the power distribution system according to claim 2, characterized in that: Preprocess the multidimensional symbol sequence set, and compare the preprocessed symbol sequences of the same dimension to obtain the integral sequence set corresponding to each dimensional symbol sequence set, including: Performing mask processing on each symbol sequence in the multidimensional symbol sequence set; The masked symbol sequences in each dimensional symbol sequence set are compared in pairs, and two symbol sequences with the same comparison results are integrated and added by one; After the comparison of all dimensional symbol sequence sets is completed, the integral corresponding to each symbol sequence is counted and the final integral sequence set is generated.

5. The method for dividing the operation scenarios of the power distribution system according to claim 2 or 4, characterized in that: Based on the multidimensional symbol sequence set and the multidimensional integral sequence set, the sequence variance and information entropy are determined and evaluation indicators are established, including: Calculate the information entropy for each dimensional symbol sequence set, and calculate the variance for each dimensional integral sequence set; Based on the calculated information entropy and variance, the weight coefficient of each subsequence in the corresponding multidimensional subsequence set is determined.

6. The method for dividing the operation scenarios of the power distribution system according to claim 5, characterized in that: Based on the evaluation index, the multidimensional subsequence set is screened and the discrimination is calculated, and the multidimensional subsequence with the greatest discrimination is given, including: Based on the weight threshold and the weight coefficient of each subsequence in the multidimensional subsequence set, the multidimensional subsequence set is screened to provide a screened multidimensional subsequence set; Based on the filtered multi-dimensional subsequence set, the distances with the time series of the corresponding dimension are calculated respectively, and the distance set corresponding to each dimensional subsequence set arranged according to the distance size is given; Selecting a distance subset of a predetermined proportion extending from a central dividing line to both sides from each dimension distance set; Based on each element in each dimension distance subset as a distinguishing point, the discrimination degree of the corresponding dimension distance set is calculated; Compare the discrimination corresponding to each element in each dimensional distance subset and give the multidimensional subsequence with the maximum discrimination.

7. The method for dividing the operation scenarios of the power distribution system according to claim 6, characterized in that: Based on each element in each dimension distance subset as a distinguishing point, the discrimination of the corresponding dimension distance set is calculated, including: The elements in each dimension distance set that are smaller than the distinguishing point distance and the elements corresponding to the distinguishing point are divided into the first set, and the remaining elements are divided into the second set; Calculate the mean and standard deviation for both the first and second sets; Based on the mean and standard deviation of the first set and the second set, the discrimination of the corresponding distinguishing points is calculated to satisfy the following relationship: Gap=μ b -s b -(m a +s a ) In the formula, Gap is the discrimination, μ a , μ b are the average values ​​of the first and second sets, σ a , σ b is the standard deviation of the first and second sets.

8. The method for dividing the operation scenarios of the power distribution system according to claim 6 or 7, characterized in that: Compare the discrimination of each element in each dimensional distance subset, and give the multidimensional subsequence with the largest discrimination, including: Compare the discrimination corresponding to each element in each dimensional distance subset, and give the subsequence corresponding to the element with the largest discrimination in each dimensional distance set; Based on the time node of the subsequence corresponding to the element with the greatest discrimination in each dimensional distance set, determine the discrimination of the corresponding elements in other dimensional distance sets; Based on the corresponding subsequence of the element with the largest discrimination in each dimensional distance set and the discrimination of the corresponding elements in other dimensional distance sets of the corresponding time node, the average discrimination of the corresponding time node is given; Compare multiple average discriminations and give the multidimensional subsequence corresponding to the time node with the largest average discrimination.

9. The method for dividing the operation scenarios of the power distribution system according to claim 8, characterized in that: Pre-built decision trees, including: Take multiple groups of multidimensional historical time series as training sets, and take the multidimensional historical subsequences with the largest degree of distinction corresponding to each group of multidimensional historical time series as nodes; Recursion is performed based on the nodes corresponding to the training set until all the stopping conditions are met, and the constructed decision tree is given.

10. A device for dividing operation scenarios of a power distribution system, characterized in that: The method for dividing the operation scenarios of the power distribution system according to any one of claims 1 to 9 comprises: A traversal generation module, which is used to traverse the multidimensional time series of the power distribution system and generate a set of multidimensional subsequences with distinguishing characteristics; An index determination module is used to symbolize and integrate the multidimensional subsequence set in turn, and provide evaluation indicators based on the processing results; A sequence determination module is used to screen the multidimensional subsequence set based on the evaluation index and calculate the discrimination degree, and give the multidimensional subsequence with the greatest discrimination degree; The scenario division module is used to process the multidimensional subsequences with the greatest discrimination based on the pre-built decision tree and provide the division results of the distribution system operation scenarios.

Images