A clustering method and system for power user load curves considering adaptive fast search for density peaks

Abstract

The present invention discloses a power user load curve clustering method and system considering adaptive fast search for density peaks. The method includes the following steps: preprocessing user load data, calculating the distance matrix, local density, and relative distance between data samples; using the calculation results of local density and relative distance, adopting a power user adaptive CFSFDP load curve clustering method based on local outlier factor and silhouette coefficient to obtain the local optimal clustering centers, cut-off distance, and clustering results of the power consumption load data set under standard quantization. The method proposed by the present invention helps to reduce the deviation of clustering results caused by human factors and has certain practical significance for realizing adaptive and accurate clustering under a large amount of user data.

Description

Technical Field

[0001] The present invention relates to the technical field of power data mining, and particularly relates to a power user load curve clustering method and system considering adaptive fast search for density peaks. Background Art

[0002] Mastering the types of user electricity loads is an important prerequisite for power system dispatching and operation planning. At the same time, accurate clustering of load curves is the key basis for power load type classification. The different electricity consumption characteristics of various industries will have important impacts on aspects such as power grid dispatching arrangements, operation modes, and peak shaving capabilities. Therefore, by analyzing and mining the electricity consumption laws and demands of different industries through data, it can not only have a positive impact on aspects such as accurate load prediction and dispatching management of the power grid, but also provide support for aspects such as demand response on the user side.

[0003] The fast search for density peak clustering algorithm has relatively few preset parameters and a low time complexity, so it has strong adaptability to various types of data distributions. The selection of the clustering center and the cut-off distance are important factors affecting the clustering effect, and they play a key role in dividing different load curve clusters and determining the shape and quantity of typical load curves. Selecting the clustering center in the decision graph requires human interaction, and the artificial selection increases the subjective factors of the clustering result, creating a great obstacle to the spontaneous analysis of industry load data, having poor adaptability to the distribution of original load data, and reducing the accuracy of the clustering result. When there are multiple density peaks in the decision graph, it is difficult for humans to identify the accurate clustering center, resulting in poor clustering performance. In addition, the clustering center needs to be artificially selected each time the algorithm runs. This process not only increases the running time of the algorithm, but also the increase in human factors makes the selection of the clustering center inconsistent, and the selection quantity and selected samples at different times will vary, resulting in a decrease in the accuracy of the clustering result. Therefore, further improvement and optimization of the above two defects existing in the prior art are key issues. Summary of the Invention

[0004] The main technical problem to be solved by the present invention is to provide a power user load curve clustering method and system considering adaptive fast search for density peaks by constructing a calculation model for the local optimal clustering center and cut-off distance of a load sample data set.

[0005] The technical solution of the present invention is as follows:

[0006] A power user load curve clustering method considering adaptive fast search for density peaks, comprising:

[0007] Step 1: Preprocess the user power load data, construct a distance matrix between data samples, and calculate the local density and relative distance of the samples;

[0008] Step 2: Using the calculation results of local density and relative distance, an adaptive CFSFDP load curve clustering method for power users based on local outlier factor and silhouette coefficient is adopted to obtain the local optimal clustering centers, cut-off distance and clustering results of the power consumption load data set under standard quantization.

[0009] In the above technical solution, further, in the step 1, the power load data of users is preprocessed, a distance matrix between data samples is constructed, and the local density and relative distance of the samples are calculated, specifically including:

[0010] (1) For the power consumption load data set of power users, the sequences of missing data are found and invalid data is removed;

[0011] (2) The data is normalized;

[0012] (3) Calculate the Euclidean distance d ij between the power consumption load data samples i and j, and construct a sample distance matrix;

[0013] (4) Calculate the local density ρ i and relative distance δ i of the sample i in the power consumption load data set:

[0014]

[0015]

[0016] In the formula: d ij is the Euclidean distance between the load sample i and the load sample j in the data set, and d c is the cut-off distance. When the data point x i has the maximum local density (ρ i = max{ρ j}), δ i represents the distance between the data with the maximum distance from x i in the data set and x i ; when the data point x i does not have the maximum local density (ρ i < max{ρ j}), δ i represents the distance between the data point with the minimum distance from x i among all data points with local density greater than x i and x i .

[0017] Further, for the sample data density, a Gaussian kernel function is used for density calculation to reduce the influence of the cut-off distance on the sample data density.

[0018] Furthermore, using the calculation results of the local density ρ and the relative distance δ, an adaptive CFSFDP load curve clustering method for power users based on the local outlier factor and the silhouette coefficient is adopted to obtain a clustering center decision diagram, and the calculation of the local optimal clustering center, the cut-off distance, and the clustering result of the electricity load data set under standard quantization is completed, including:

[0019] Based on the local density ρ of the sample i in the calculated electricity load data set i and the relative distance δ i Calculate the local outlier factor of the load data sample i, which is the average value of the ratio of the local reachability density between all load samples in the k-distance neighborhood N k (i) of the load sample i and the sample i; normalize the local outlier factor of the obtained sample to obtain the normalized local outlier factor result;

[0020] According to the normalized local outlier factor result, determine the load sample set selected as the clustering center, and take this clustering center sample set as the object to classify the remaining load curve samples to obtain a preliminary clustering result;

[0021] Then, further optimize the preliminary clustering result based on the adaptive solution of the local optimal cut-off distance of the silhouette coefficient.

[0022] Furthermore, the local outlier factor of the load data sample i and its normalization are obtained by the following method:

[0023] dist(i,j) = |δ i -δ j | (3)

[0024]

[0025] |N k (i)| ≥ k (5)

[0026] dist k-reach (i,j) = max{dist k (i), dist(i,j)} (6)

[0027]

[0028]

[0029]

[0030] In the formula, dist(i,j) is the distance difference between the typical load curve samples i and j, which can be obtained from the relative distance δ; dist k(i) is the k-distance of the load sample i under the condition of satisfying the shown constraints, where the set J represents that there are at least k points j′ such that dist(i, j′) ≤ dist(i, j), and the set R represents that there are at most k - 1 points j′ such that dist(i, j′) < dist(i, j); N k (i) is the set of the k-distance neighborhoods of i, including the load curves with distances less than or equal to the k-distance between load samples i; |N k (i)| is N k (i) is the number of elements contained in the set; dist k-reach (i, j) is the reachable distance between two load samples; d lrd (i) is the local reachability density of the load sample i, which represents the reciprocal of the reachable distances of all load samples within the k-distance neighborhood of the load sample i; LOF(i) is the local outlier factor of the load sample i, which is expressed as the average value of the ratios of the local reachability densities of all load samples within the k-distance neighborhood N k (i) to i; LOF n (i) is the result of the normalized local outlier factor, LOF max (i) is the maximum value among the local outlier factors of all samples, LOF min (i) is the minimum value among the local outlier factors of all samples.

[0031] Further, by calculating the local density ρ and the relative distance δ corresponding to the result of the normalized local outlier factor, the load sample set selected as the clustering center is obtained. Specifically: after normalizing the local outlier factor, according to a pre-determined threshold, the normalized local outlier factors exceeding the threshold are screened, and the local density ρ and the relative distance δ corresponding to these selected samples are obtained, and their positions in the decision graph are determined to identify the clustering center samples

[0032] Further, the preliminary clustering result is further optimized based on the adaptive solution of the local optimal truncation distance of the silhouette coefficient, where the silhouette coefficient is an evaluation index for selecting the clustering result, and by continuously changing the size of d c the results of each clustering are evaluated using the silhouette coefficient, and according to the best result of the silhouette coefficient, the optimal size of d c and the best clustering result are determined; specifically: c the size of and the best clustering result; specifically:

[0033]

[0034] a(i) = average j∈A,j≠i (d ij ) (11)

[0035] b(i) = min(averagei∈A,j∈B (d ij )) (12)

[0036] sc(i) ∈ [-1, 1] (13)

[0037]

[0038] Wherein, sc(i) is the silhouette coefficient of load sample i; A is the cluster to which load sample i belongs; a(i) is the average value of the dissimilarity within the cluster, reflecting the cohesion of the load samples in this load cluster, and is expressed as the average distance between i and other load samples in the cluster A to which it belongs; b(i) is the minimum value of the average inter-cluster dissimilarity, reflecting the separation of the load samples, and is expressed as the minimum value of the average distance between i and the load samples in other load clusters B; Ω represents the load sample set; |Ω| represents the number of load samples in the load sample set; is the average value of the silhouette coefficients of all load samples; for different d c For the obtained clustering results, the clustering result corresponding to the maximum average silhouette coefficient is taken as the optimal clustering result, and the corresponding truncation distance is the optimal truncation distance.

[0039] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the power user load curve clustering method considering adaptive fast search for density peaks as described in any one of the above.

[0040] The present invention also provides a power user load curve clustering system considering adaptive fast search for density peaks, and the system includes:

[0041] One or more processors;

[0042] A memory for storing one or more programs;

[0043] When the one or more programs are executed by the one or more processors, the one or more processors implement the power user load curve clustering method considering adaptive fast search for density peaks as described in any one of the above.

[0044] The beneficial effects of the present invention are:

[0045] Aiming at the accurate clustering of electricity consumption characteristics of users in various industries and the precise classification of electricity consumption types of users, the present invention proposes a method for clustering power user load curves that takes into account adaptive rapid search for density peaks. Compared with the traditional power clustering method based on density peaks, the adaptive rapid search for density peak user load clustering method proposed in the present invention can simultaneously improve the influence of artificially selected input parameters and clustering centers on clustering accuracy and analysis time, and is helpful to formulate a more reasonable and effective power user load clustering strategy, which has certain practical significance for realizing adaptive and accurate clustering under a large amount of user data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the overall step flow of a specific implementation method of the present invention.

[0047] Figure 2 This is a typical curve of the clustering results of the industry measured load data of the method of the present invention.

[0048] Figure 3 The present invention compares the performance indicators of clustering results of measured load data in the industry between the method of the present invention and other methods. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the invention.

[0050] The present invention proposes a power user load curve clustering method for adaptively and quickly searching for density peaks. Figure 1 As shown, the implementation process includes the following steps:

[0051] Step 1: Preprocess the user's power load data and calculate the distance matrix, local density and relative distance between data samples. The specific implementation method of this step is as follows:

[0052] (1) Input the power load data set of power users, find the sequence of missing data, and eliminate invalid data;

[0053] (2) Normalize the data;

[0054] (3) Calculate the Euclidean distance d between power load data samples i and j ij , construct the sample distance matrix.

[0055] (4) Calculate the local density ρ of sample i in the power load data set i and the relative distance δ i , which can be expressed as the following formula:

[0056]

[0057]

[0058] Where: d ij is the Euclidean distance between the load sample i and the load sample j in the dataset, and d c is the truncation distance. To reduce the influence of the truncation distance on the sample data density, a Gaussian kernel function is used for density calculation. It can be seen that when calculating the relative distance, when the data point x i has the maximum local density (ρ i = max{ρ j}), δ i is the distance between the data point with the maximum distance from x i in the dataset and x i ; when the data point x i does not have the maximum local density (ρ i < max{ρ j}), δ i is the distance between the data point with the minimum distance from x i among all data points with local density greater than x i and x i .

[0059] Step 2: Using the calculation results of the local density and the relative distance, propose an adaptive CFSFDP load curve clustering method for power users based on the local outlier factor and the silhouette coefficient to obtain the local optimal clustering center, truncation distance, and clustering result of the electricity load dataset under standard quantization. The specific implementation method of this step is as follows:

[0060] The calculation process of the local outlier factor of the load data can be expressed by the following formula:

[0061] dist(i,j) = |δ i - δ j | (31)

[0062]

[0063] |N k (i)| ≥ k (33)

[0064] dist k-reach (i,j) = max{dist k (i), dist(i,j)} (34)

[0065]

[0066]

[0067]

[0068] Wherein, dist(i,j) is the distance difference between the typical load curve samples i and j, which can be obtained from the relative distance δ; dist k (i) is the k-distance of the load sample i under the condition of satisfying the shown constraints, and it is equal to the distance value dist(i,j) between i and j when the following two conditions are satisfied: there are at least k points j' in the set J such that dist(i,j') ≤ dist(i,j), and there are at most k - 1 points j' in the set R such that dist(i,j') < dist(i,j); N k (i) is the k-distance neighborhood set of i, including the load curves with distances less than or equal to the k-distance between the load samples i; |N k (i)| is the number of elements contained in the set N k (i); dist k-reach (i,j) is the reachable distance between two load samples; d lrd (i) is the local reachability density of the load sample i, which represents the reciprocal of the reachable distances of all load samples within the k-distance neighborhood of the load sample i; LOF(i) is the local outlier factor of the load sample i, which is expressed as the average value of the ratio of the local reachability density of all load samples within the k-distance neighborhood N k (i) of the load sample i to i; LOF n (i) is the result of the normalized local outlier factor.

[0069] Through the calculation result of the normalized local outlier factor, the LOF distribution is obtained. A threshold (this threshold can be set according to experience) is specified, and the samples exceeding this threshold are screened. Corresponding to the local density ρ and the relative distance δ, their positions in the decision diagram and the set of load samples selected as the clustering centers are obtained. Taking this set of clustering center samples as the object, the remaining load curve samples are classified to obtain the preliminary clustering result. The preliminary clustering result is further optimized through subsequent adaptive solution of the local optimal truncation distance based on the silhouette coefficient.

[0070] The adaptive solution process of the local optimal truncation distance based on the silhouette coefficient can be expressed by the following formula:

[0071]

[0072] a(i) = average j∈A,j≠i (d ij ) (39)

[0073] b(i) = min(average i∈A,j∈B (d ij )) (40)

[0074] sc(i) ∈ [-1, 1] (41)

[0075]

[0076] Wherein, sc(i) is the silhouette coefficient of load sample i; A is the cluster to which load sample i belongs; a(i) is the average value of the dissimilarity degree within the cluster, reflecting the cohesion of load samples in this load cluster, and is expressed as the average distance between i and other load samples in the belonging cluster A; b(i) is the minimum value of the average inter-cluster dissimilarity degree, reflecting the separation degree of load samples, and is expressed as the minimum value of the average distance between i and load samples in other load clusters B; Ω represents the load sample set; |Ω| represents the number of load samples in the load sample set; is the average value of the silhouette coefficients of all load samples.

[0077] Example 1

[0078] To verify the effectiveness of the proposed power user load curve clustering method considering adaptive fast search for density peaks, the present invention takes the metal processing machinery manufacturing industry in a certain city of Zhejiang Province as the analysis object, selects 1000 load curves of 100 load users in this industry from September 10th to 19th, 2021 as input sample data for analysis. The data has a sampling interval of 15 minutes, and there are 96 sampling points per day. The typical load curves of the clustering results of this user load data are obtained as Figure 2 shown. The results in the figure show that the number of clustering centers automatically captured by the proposed method is 4, and the load types in this industry are mainly divided into high-power stable load, peak-shaving load, and double-peak load. Therefore, the feasibility of the proposed method in power user load clustering can be reflected, and this method can effectively reduce the repetition of clustering cluster types, making the clustering results more accurate and the curve types within the clusters more concentrated.

[0079] Example 2

[0080] To further verify the superiority of the proposed power user load curve clustering method considering adaptive fast search for density peaks compared with other clustering methods, using the above user load data as input samples, the clustering results of the present invention (i.e., the optimized CFSFDP method) are compared with those of the k-means clustering method, DBSCAN clustering method, and traditional CFSFDP clustering method. Table 1 shows the number of clustering clusters and running time of the results of four different algorithms. It can be seen that the number of clustering clusters of the DBSCAN algorithm is too large, indicating that the division of its clustering center types is not clear and there are many repeated electricity consumption types, which will greatly reduce the accuracy of the clustering results.

[0081] Table 1

[0082]

[0083] The clustering results of the four methods are evaluated using the results of the clustering metrics CHI (Calinski-Harabasz index) and DBI (Davies-Bouldin index). Among them, CHI uses the within-cluster scatter matrix to measure the within-cluster compactness, and the larger its value, the better the clustering effect; DBI represents the ratio of the average distance within the clustering clusters to the shortest distance between clusters. Therefore, the smaller its value, the better the clustering effect. The performance metrics of the load curve data for the clustering results of different methods are as Figure 3 shown, where Figure 3 (a) is the CHI metric of the clustering result, Figure 3 (b) is the DBI metric of the clustering result. As can be Figure 3 seen, both the CHI value and the DBI value of the optimized clustering method proposed in the present invention are better than the index results of the other three clustering methods, indicating that the classified load curve clusters have more superior within-cluster similarity and between-cluster difference, and the clustering results are more accurate. The CHI metric value of the DBSCAN algorithm is close to that of the optimized algorithm proposed in the present invention. However, the number of central clusters selected by the DBSCAN algorithm is too large, and the difference in the types of clustering centers is not clear. This can also be seen from the relatively high DBI metric value compared to the clustering algorithm of the present invention. The CHI and DBI metric results of the k-means algorithm are both poor, indicating that the similarity of the curve morphological features within the load clusters obtained by this method is low, and the classification effect between clusters is not obvious, and the accuracy of classifying the load curve categories is not high. There is a large gap between the CHI metric and the DBI metric of the traditional density peak clustering algorithm and the index values of the clustering results of the optimized algorithm of the present invention, indicating that the traditional density peak clustering algorithm has obvious defects in identifying the morphological differences of different load types. At the same time, the running time is significantly longer than that of the other three algorithms, indicating that manual participation in clustering is the main limiting factor in the running time of this algorithm. Through comprehensive analysis of various aspects such as the overall running time of load clustering and the clustering effectiveness metrics, the method proposed in the present invention is superior to the clustering results of traditional algorithms in actual case performance.

[0084] In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the power user load curve clustering method considering adaptive fast search for density peaks as described in any one of the above.

[0085] The present invention also provides a power user load curve clustering system considering adaptive fast search for density peaks, and the system includes:

[0086] One or more processors;

[0087] A memory for storing one or more programs;

[0088] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above power user load curve clustering methods that consider adaptive fast search for density peaks.

[0089] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] The above has introduced in detail the power user load curve clustering method that considers adaptive fast search for density peaks provided by the present invention. Specific examples are used herein to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A clustering method for power user load curves considering adaptive fast search for density peaks, characterized in that: include: Step 1: Preprocess the user's power load data, construct a distance matrix between data samples, and calculate the local density and relative distance of the samples; Step 2: Using the calculation results of local density and relative distance, the power user adaptive CFSFDP load curve clustering method based on local outlier factor and silhouette coefficient is used to obtain the local optimal clustering center, cutoff distance and clustering results of the power load data set under standard quantization; Using the calculation results of local density ρ and relative distance δ, the power user adaptive CFSFDP load curve clustering method based on local outlier factor and silhouette coefficient is adopted to obtain the cluster center decision diagram, and complete the calculation of the local optimal cluster center, truncation distance and clustering results of the power load data set under standard quantization, including: Based on the locally density ρ of the sample i in the calculated power consumption load dataset i and the relative distance δ i Calculate the local outlier factor of the load data sample i, which is the average value of the ratios of the local reachability densities of all load samples in the k-distance neighborhood N k (i) to the sample i; normalize the local outlier factor of the obtained sample to obtain the normalized local outlier factor result; According to the result of normalized local outlier factor, the load sample set selected as the cluster center is determined, and the remaining load curve samples are classified based on the cluster center sample set to obtain the preliminary clustering result; Then, the preliminary clustering results are further optimized based on the adaptive solution of the local optimal cutoff distance of the silhouette coefficient; Among them, the load sample set selected as the cluster center is obtained by calculating the local density ρ and relative distance δ corresponding to the normalized local outlier factor calculation result. Specifically, after normalizing the local outlier factor, according to a predetermined threshold, the normalized local outlier factor exceeding the threshold is screened, and the local density ρ and relative distance δ of the corresponding screened samples are obtained to obtain their position in the decision diagram and determine the cluster center sample.

2. The power user load curve clustering method for considering adaptive fast search density peaks according to claim 1, characterized in that: In the step 1, the user power load data is preprocessed, a distance matrix between data samples is constructed, and the local density and relative distance of the samples are calculated, which specifically includes: (1) For the power load data set of power users, find the sequence of missing data and eliminate invalid data; (2) Normalize the data; (3) Calculate the Euclidean distance d between the electricity load data samples i and j ij , and construct a sample distance matrix; (4) Calculate the local density ρ of sample i in the power load dataset i and the relative distance δ i : Where: d ij is the Euclidean distance between load sample i and load sample j in the dataset, and d c is the truncation distance.

3. The power user load curve clustering method for considering adaptive fast search of density peaks according to claim 2, characterized in that: For the sample data density, Gaussian kernel function is used for density calculation.

4. The power user load curve clustering method considering adaptive fast search for density peaks according to claim 1, characterized in that: The local outlier factor of load data sample i and its normalization are obtained by the following method: dist(i,j) = |δ i - δ j | (3) dist k (i) = dist(i, j) stdist(i,j′)≤dist(i,j) (j′≠i,j′∈J) |J|≥k (4) dist(i,r′)<dist(i,j) (r′≠i,r′∈R) |R|≤k-1 |N k (i)|≥k (5) dist k-reach (i,j) = max{dist k (i), dist(i,j)} (6) Where dist(i,j) is the distance difference between typical load curve samples i and j, which can be obtained from the relative distance δ; dist k (i) is the k-distance of the load sample i under the condition of satisfying the shown constraints, where the set J represents that there are at least k points j′ such that dist(i, j′) ≤ dist(i, j), and the set R represents that there are at most k - 1 points j′ such that dist(i, j′) < dist(i, j); N k (i) is the set of the k-distance neighborhoods of i, including the load curves with distances less than or equal to the k-distance between load samples i; |N k (i)| is the number of elements contained in the set N k (i); dist k-reach (i, j) is the reachability distance between two load samples; d lrd (i) is the local reachability density of the load sample i, which represents the reciprocal of the reachability distances of all load samples within the k-distance neighborhood of the load sample i; LOF(i) is the local outlier factor of the load sample i, which is expressed as the average value of the ratios of the local reachability densities of all load samples within the k-distance neighborhood N k (i) of the load sample i; LOF n (i) is the result of the normalized local outlier factor, LOF max (i) is the maximum value among the local outlier factors of all samples, LOF min (i) is the minimum value among the local outlier factors of all samples.

5. The method for clustering power user load curves by considering adaptive fast search for density peaks according to claim 1, wherein: The preliminary clustering results are further optimized by adaptively solving the local optimal cutoff distance based on the silhouette coefficient, specifically: a(i) = average j∈A,j≠i (d ij ) (11) b(i) = min(average i∈A,j∈B (d ij )) (12) sc(i)∈[-1,1](13) In the formula, sc(i) is the silhouette coefficient of load sample i; A is the cluster to which load sample i belongs; a(i) is the average value of the dissimilarity within the cluster, reflecting the cohesion of load samples in the load cluster, expressed as the average distance between i and other load samples in the cluster A to which it belongs; b(i) is the minimum value of the average dissimilarity between clusters, reflecting the separation of load samples, expressed as the minimum value of the average distance between i and load samples in other load clusters B; Ω represents the load sample set; |Ω| represents the number of load samples in the load sample set; is the average of the silhouette coefficients of all load samples; for different d c For the obtained clustering results, take the clustering result corresponding to the maximum average silhouette coefficient as the best clustering result, and the corresponding truncation distance is the optimal truncation distance.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the power user load curve clustering method considering adaptive fast search for density peaks as described in any one of claims 1-5.

7. A power user load curve clustering system considering adaptive fast search for density peaks, characterized in that, The system includes: One or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the power user load curve clustering method considering adaptive fast search for density peaks as described in any one of claims 1-5.

Images