A distribution network load characteristics research method and terminal

Abstract

The present invention discloses a method and a terminal for studying the load characteristics of a distribution network. The method comprises the following steps: obtaining sample data to be clustered, determining a characteristic vector for clustering according to the sample data; performing initial classification on the sample data according to the characteristic vector to obtain an initial classification result; selecting the initial classification result to obtain a selected result, and determining the daily load curves of the corresponding power consumption modes according to the selected result; verifying the selected result according to the daily load curve, and if the verification is passed, identifying the daily load curve to be identified according to the daily load curve of each power consumption mode to determine its corresponding power consumption mode, otherwise, adjusting the selected result based on the daily load curve until the verification is passed; the accuracy of the clustering result is ensured by initial classification, selection and verification in sequence, and then determining the power consumption mode of the daily load curve to be identified according to the accurate clustering result, thereby improving the accuracy of the power consumption attributes of the distribution network equipment in the identified distribution network.

Description

Technical Field

[0001] The present invention relates to the field of distribution network data processing, and in particular to a distribution network load characteristic research method and terminal. Background Art

[0002] The distribution network is an important part of the power network system, and it plays an important role in distributing electric energy. Distribution network planning involves the mining and analysis of various massive distribution network data. If useful information can be mined and analyzed from the massive distribution network data, it will play a very important role in improving the level of power grid planning and supporting the planning and construction of distribution networks.

[0003] However, due to the massive heterogeneity and polymorphism of data in massive power grids, the traditional distribution network planning and design model is gradually unable to adapt to the actual planning business needs, and it is difficult to meet the analysis needs of quickly and accurately acquiring knowledge and information from massive data.

[0004] Among them, the analysis of distribution network data involves analyzing the load characteristics of distribution network equipment with unknown attributes, so as to determine the power consumption attribute classification of the distribution network equipment based on the analysis results. After identifying the power consumption attributes of the distribution network equipment, it is possible to predict the change trend of its load characteristics based on its power consumption attributes, providing a basis for the distribution network planning scheme to reasonably arrange power supply capacity and determine capacity configuration. Therefore, it is particularly important to accurately identify the power consumption attributes of distribution network equipment. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method and a terminal for studying the load characteristics of a distribution network, which can improve the accuracy of the power consumption attributes of distribution network equipment in the identified distribution network.

[0006] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0007] A method for studying load characteristics of a distribution network comprises the following steps:

[0008] S1. Obtain sample data to be clustered, and determine a feature vector for clustering based on the sample data;

[0009] S2. Performing initial classification on the sample data according to the feature vector to obtain an initial classification result;

[0010] S3, selecting the initial classification results to obtain selected results, and determining the corresponding daily load curves of each power consumption mode according to the selected results;

[0011] S4, verifying the selected result according to the daily load curve, if the verification is passed, executing S5, otherwise, adjusting the selected result based on the daily load curve until the verification is passed;

[0012] S5. Identify the daily load curve to be identified according to the daily load curves of each power consumption mode, and determine the corresponding power consumption mode.

[0013] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0014] A distribution network load characteristic research terminal comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in the above-mentioned distribution network load characteristic research method is implemented.

[0015] The beneficial effects of the present invention are as follows: when clustering, the characteristic vector of the sample data is first determined, and initial classification is performed based on the determined characteristic vector to obtain an initial classification result, and then the initial classification result is selected to obtain a selected result, and the daily load curve of each corresponding power consumption mode is determined based on the selected result, and finally the selected result is verified according to the daily load curve. If the verification fails, the selected result is adjusted based on the daily load curve until the verification passes. Only after the verification passes, the daily load curve to be identified is identified according to the daily load curve of each power consumption mode to determine its corresponding power consumption mode. The accuracy of the clustering results is ensured by initial classification, selection and verification in sequence, and then the power consumption mode of the daily load curve to be identified is determined according to the accurate clustering result, thereby improving the accuracy of the power consumption attributes of the distribution network equipment in the identified distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of the steps of a method for studying load characteristics of a distribution network according to an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of the structure of a distribution network load characteristic research terminal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0019] Please refer to Figure 1 , a method for studying load characteristics of a distribution network, comprising the steps of:

[0020] S1. Obtain sample data to be clustered, and determine a feature vector for clustering based on the sample data;

[0021] S2. Performing initial classification on the sample data according to the feature vector to obtain an initial classification result;

[0022] S3, selecting the initial classification results to obtain selected results, and determining the corresponding daily load curves of each power consumption mode according to the selected results;

[0023] S4, verifying the selected result according to the daily load curve, if the verification is passed, executing S5, otherwise, adjusting the selected result based on the daily load curve until the verification is passed;

[0024] S5. Identify the daily load curve to be identified according to the daily load curves of each power consumption mode, and determine the corresponding power consumption mode.

[0025] From the above description, it can be seen that the beneficial effect of the present invention is that: when clustering, the characteristic vector of the sample data is first determined, and initial classification is performed based on the determined characteristic vector to obtain an initial classification result, and then the initial classification result is selected to obtain a selected result, and the daily load curve of each corresponding power consumption mode is determined based on the selected result, and finally the selected result is verified according to the daily load curve. If the verification fails, the selected result is adjusted based on the daily load curve until the verification passes. Only after the verification passes, the daily load curve to be identified is identified according to the daily load curve of each power consumption mode to determine its corresponding power consumption mode. The accuracy of the clustering results is ensured by initial classification, selection and verification in sequence, and then the power consumption mode of the daily load curve to be identified is determined based on the accurate clustering result, thereby improving the accuracy of the power consumption attributes of the distribution network equipment in the identified distribution network.

[0026] Furthermore, the determining of the feature vector for clustering according to the sample data includes:

[0027] Determine two samples with the farthest distance in the sample data, and add the two samples to the reference sample set;

[0028] Taking all samples in the reference sample set as reference, determining a sample that is farthest from all samples in the reference sample set and whose distances are all greater than a first threshold, adding the sample to the reference sample set, and returning to execute the step of taking all samples in the reference sample set as reference until no sample that meets the requirement exists;

[0029] All samples in the reference sample set are determined as feature vectors for clustering.

[0030] From the above description, it can be seen that when selecting feature vectors for clustering, different from the existing random selection that causes inaccuracy and thus leads to the problem of slow subsequent convergence time, samples that are far together and have a distance greater than a first threshold are selected based on the distance between samples as feature vectors for clustering. This can ensure the rationality and accuracy of the feature vectors determined for clustering, thereby improving the accuracy of the clustered results and the speed of clustering convergence.

[0031] Furthermore, the initial classification results are selected to obtain the selected results, including:

[0032] Counting the number of samples in each category in the initial classification result, and deleting the categories whose sample numbers do not fall within a preset range;

[0033] Determine the centroid corresponding to each category in the classification result after deleting the category, delete the samples in each category whose distance to the centroid corresponding to the category is greater than the second threshold, and obtain the selected result.

[0034] From the above description, it can be seen that after obtaining the preliminary classification results, the classifications whose sample numbers do not meet the preset range are deleted based on the number of classified samples, so as to first eliminate the classifications with abnormal sample numbers, and then based on the centroid corresponding to each classification after deleting the abnormal classifications, the samples whose distance from the centroid is greater than the second threshold are deleted. By carefully selecting the initial classifications, the rationality of the aggregation degree of the selected classifications is guaranteed, thereby further ensuring the rationality and accuracy of the aggregation results.

[0035] Furthermore, determining the daily load curve of each corresponding power consumption mode according to the selected result includes:

[0036] The sample average value of each category in the selected results is determined, and the sample average value of each category is determined as the daily load curve of the power consumption mode corresponding to each category.

[0037] From the above description, it can be seen that the daily load curve of the power consumption mode corresponding to each category is determined according to the average value of the selected samples of each category, and the value that can reflect the characteristics of each category is determined by statistical methods, which ensures the accuracy of the determined value, thereby further improving the accuracy of power consumption mode identification.

[0038] Further, the verifying the selection result according to the daily load curve includes:

[0039] Determine the data radius of each classification in the selected result and the degree of separation from other classifications according to the daily load curve;

[0040] Verifying the selected result according to the data radius and the separation degree;

[0041] The data radius is the distance between the sample farthest from the corresponding daily load curve in each classification and the daily load curve;

[0042] The separation degree is the distance between each category and the daily load curve corresponding to the other categories closest to it.

[0043] From the above description, it can be seen that after the selection, the selection results are verified based on the data radius and separation of each classification, which can further ensure the convergence and mutual difference of each classification after selection, thereby further improving the accuracy of clustering.

[0044] Further, the verifying the selected result according to the data radius and the separation degree includes:

[0045] Determine the difference between the separation degree and the data radius corresponding to each classification in the selected result;

[0046] It is determined whether the difference values ​​corresponding to each category are greater than or equal to the third threshold value. If so, the verification is passed; otherwise, the verification fails.

[0047] From the above description, it can be seen that by determining the difference between the separation degree and the data radius corresponding to each classification, and comparing the difference with the third threshold, if the difference corresponding to each classification is greater than the third threshold, it means that the differences between the determined classifications are large enough, therefore, the rationality of the clustering results is verified.

[0048] Further, the adjusting the selected result based on the daily load curve until the verification is passed includes:

[0049] The categories whose difference is less than the third threshold are deleted.

[0050] It can be seen from the above description that directly deleting the classifications whose difference is less than the third threshold value can quickly and easily ensure the rationality and accuracy of the clustering results.

[0051] Further, the adjusting the selected result based on the daily load curve until the verification is passed includes:

[0052] The samples at the edge of the classification whose difference is less than the third threshold are deleted to adjust the corresponding data radius, so that the difference corresponding to the classification after deleting the edge samples is greater than or equal to the third threshold.

[0053] From the above description, it can be seen that for the classification whose difference is less than the third threshold, the data radius is reduced by deleting the samples at its edge, so that the difference between it and other classifications can be increased, and the requirement that its difference is greater than or equal to the third threshold is met. The given sample data can be fully utilized for clustering, further ensuring the accuracy and rationality of the clustering results.

[0054] Please refer to Figure 2 A distribution network load characteristic research terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in the above-mentioned distribution network load characteristic research method is implemented.

[0055] The above-mentioned distribution network load characteristic research method and terminal of the present invention are suitable for identifying the power consumption attributes of electrical equipment with unknown power consumption attributes in the distribution network, such as the power consumption attributes of distribution network transformers in different industries, etc. The electrical equipment for power consumption attribute identification has the following characteristics: users with the same power consumption characteristics have similar load characteristics, and the load characteristics of users with different power consumption characteristics have strong distinguishability, which is explained below through specific implementation methods:

[0056] Embodiment 1

[0057] This embodiment is explained by taking the load type of the distribution transformer as an example. Figure 1 , a method for studying load characteristics of a distribution network, comprising the steps of:

[0058] S1. Obtain sample data to be clustered, and determine a feature vector for clustering based on the sample data;

[0059] In this embodiment, the sample data is composed of 24-point load data of the distribution transformer, that is, the daily load curve of the distribution transformer. What is to be identified is the daily load curve of the distribution transformer of different industries. The load power at each time point reflects the electricity consumption of users in different periods of time, and users with the same electricity consumption characteristics have similar load characteristics. Different industries have different electricity consumption patterns. Therefore, the corresponding daily load curves are also quite different. Based on this, it is possible to first cluster the sample data to be clustered to determine the daily load curves of different industries, that is, first determine the typical load curve, and then compare the daily load curve of the distribution transformer to be identified with the typical load curve to determine which typical load curve it is most similar to. The daily load curve of the distribution transformer to be identified and the typical load curve with the highest similarity are classified into one category, so that its corresponding electricity consumption attribute can be determined;

[0060] In this implementation, the industry electricity consumption can be classified based on the daily load curve of the 10kV special transformer to obtain the corresponding typical electricity load, and then the obtained typical electricity daily load curve can be used for the electricity attribute classification of the public transformer;

[0061] When determining the feature vector for clustering, if the types of electricity consumption by industry are known in advance, that is, for example, it is determined that there are 5 categories in total, 5 sample data can be randomly selected from the sample data to be clustered as the feature vector for clustering;

[0062] For users in the same mode, in order to avoid inaccurate classification when the load levels are greatly different, the load power at each measurement time point needs to be normalized;

[0063] Let U i =[u i1 ,u i2 ,…,u ij ,…,u in ] is the power value of the i-th distribution transformer at point n, and U i The corresponding standard value U′ can be obtained by normalizing according to the following formula: i =[u′ i1 , u′ i2 ,…,u′ ij ,…,u′ in ].

[0064]

[0065] Where j = 1, 2, ..., n is the number of the power sampling point of the distribution transformer; u imax is the maximum value of the n-point power of the ith distribution transformer;

[0066] S2. Performing initial classification on the sample data according to the feature vector to obtain an initial classification result;

[0067] Specifically, the feature vector is used as the cluster center, each feature vector represents a category, and the remaining objects in the sample data are clustered. In the clustering process, the distance between the remaining objects in the sample data and each cluster center is calculated respectively, and each object is classified into the category where the cluster center with the closest distance to it is located. For each category, each time a new object is added, the average value of each category is recalculated as the class center of the category, and then the above process of adding objects is repeated until convergence. Convergence can be judged by judging whether the class centers calculated before and after in a category are consistent, that is, whether the difference is less than a preset difference. If so, it means that the category has converged;

[0068] S3, selecting the initial classification results to obtain selected results, and determining the corresponding daily load curves of each power consumption mode according to the selected results;

[0069] The step of selecting the initial classification results to obtain the selected results includes:

[0070] Counting the number of samples in each category in the initial classification result, and deleting the categories whose sample numbers do not fall within a preset range;

[0071] For example, if a total of 5 categories are counted, and the number of samples in each category is 5, 15, 17, 19, and 28 respectively, and the preset range is [10, 20], then the categories with 5 and 28 samples will be deleted;

[0072] In an optional implementation, the deleted sample data may be clustered, and the process may be repeated until a clustering result that meets the sample number requirement is obtained;

[0073] Determine the centroid corresponding to each category in the classification results after the categories are deleted, delete the samples in each category whose distance to the centroid corresponding to the category is greater than the second threshold, and obtain the selected results;

[0074] In an optional implementation, the sample average value may be calculated for each category, and the sample average value is used as the centroid of each sample. Then, the distance between each sample in each category and its corresponding centroid is determined, and samples with a distance greater than a second threshold are deleted; wherein the second threshold reflects the degree of aggregation of samples in each category, and by setting the second threshold, samples in each category that are relatively far from their centroid can be deleted, thereby ensuring that each category has a high degree of aggregation;

[0075] In another optional implementation, the classifications whose number is less than the left boundary of the preset range may be deleted, and for the classifications whose number is greater than the right boundary of the preset range, if there is no sample whose distance from the centroid is greater than the second threshold, the classification is retained;

[0076] Determining the daily load curve of each corresponding power consumption mode according to the selected result includes:

[0077] Determine the sample average value of each category in the selected results, and determine the sample average value of each category as the daily load curve of the power consumption mode corresponding to each category, where one power consumption mode corresponds to one typical daily load curve;

[0078] S4, verifying the selected result according to the daily load curve, if the verification is passed, executing S5, otherwise, adjusting the selected result based on the daily load curve until the verification is passed;

[0079] The daily load curve represents the classification to which it belongs, that is, it represents a power consumption mode. In order to ensure the accuracy of the clustering results, the selected results can be verified based on the daily load curve. In an optional implementation, the distance calculation can be directly performed on the daily load curves corresponding to each power consumption mode, and the differences between different power consumption modes can be determined by distance calculation. It is judged whether the distances between different power consumption modes are all greater than a preset value. If so, it is said that the differences between different power consumption modes are large enough, and the verification is passed. The preset value is set in advance and is used to represent the differences between different power consumption modes. If there are two power consumption modes with a relatively small distance, the two power consumption modes can be merged or one of the power consumption modes can be deleted to meet the difference requirements between different power consumption modes;

[0080] S5. Identify the daily load curve to be identified according to the daily load curves of each power consumption mode, and determine the corresponding power consumption mode;

[0081] After determining the daily load curves of each power consumption mode, for example, five power consumption modes are finally determined: power consumption mode 1, power consumption mode 2, power consumption mode 3, power consumption mode 4, and power consumption mode 5. Then, the distance between the daily load curve to be identified and the daily load curves corresponding to the five power consumption modes is calculated in turn, and the power consumption mode corresponding to the smallest distance is determined as the power consumption mode of the daily load curve to be identified;

[0082] Among them, before performing the distance calculation, it is also necessary to normalize the daily load curve to be identified. When performing the distance calculation, the square of the spatial distance between the normalized daily load curve of the distribution transformer and the typical daily load curve of each power consumption mode is calculated. The smaller the distance, the higher the similarity between the distribution transformer and the power consumption mode. The power consumption mode with the highest similarity is selected as the affiliation of the unknown type of distribution transformer. The calculation formula of the square of the spatial distance is as follows:

[0083]

[0084] Where, k = 1, 2, ..., n is the number of the power sampling point of the distribution transformer; X j =[x j1 ,x j2 ,…,x jk ,…,x jn ] is the n-point power value (normalized) of a typical industry; X i =[x i1 ,x i2 ,…,x ik ,…,x in ] is the n-point power value (normalized) of the i-th distribution transformer;

[0085] After determining the power consumption mode to which the distribution transformer to be identified belongs, the daily load characteristic indicators of the distribution transformer to be identified can be predicted based on the daily load characteristic indicators of the power consumption mode, such as daily maximum load, daily average load, daily peak-to-valley difference, daily peak-to-valley difference rate, daily load rate, and daily minimum load rate, thereby providing a basis for the distribution network planning scheme to reasonably arrange power supply capacity and determine capacity configuration.

[0086] Embodiment 2

[0087] This embodiment provides another method for determining the feature vector for clustering. This embodiment is particularly suitable for application scenarios where the cluster types and locations are known. Of course, it is also applicable to application scenarios where the cluster types are known. Specifically:

[0088] Determine two samples with the farthest distance in the sample data, and add the two samples to the reference sample set;

[0089] Among them, the distance between two samples in the sample data can be calculated, and then the two samples with the largest distance can be selected through statistical analysis;

[0090] Taking all samples in the reference sample set as reference, determining a sample that is farthest from all samples in the reference sample set and whose distances are all greater than a first threshold, adding the sample to the reference sample set, and returning to execute the step of taking all samples in the reference sample set as reference until no sample that meets the requirement exists;

[0091] For example, if there are two samples in the reference sample set at the beginning, then in the sample data, for the remaining sample data, the distance between it and the two samples is calculated in turn. If the distance is less than or equal to the first threshold, it is eliminated from the selection range of the feature vector, that is, it is not necessary to calculate the distance between it and another sample in the reference sample set, nor is it needed in the subsequent element selection of the reference sample set. Only when the distance with the samples in the reference sample set is greater than the first threshold, it can be used as the selection object of the feature vector for clustering;

[0092] On the basis of satisfying the requirement of being greater than the first threshold, when it is determined that the distance to the sample in the reference sample set is the largest, the distance between the object to be selected and each sample in the reference sample set can be added in sequence, and the object to be selected with the largest distance after addition is selected as the element in the reference sample set;

[0093] Determine all samples in the reference sample set as feature vectors for clustering;

[0094] Among them, in an optional embodiment, in order to reduce the computational cost, the clustering process can be implemented synchronously in the process of determining the feature vector for clustering, that is, when calculating the distance with the samples in the reference sample set, the calculated distance can be compared with the second threshold reflecting the clustering degree in Example 1. If the distance is less than the second threshold, it can be directly added to the category where the corresponding sample in the reference sample set is located, and the class center of the corresponding category can be updated synchronously. In this way, clustering is performed while determining the feature vector of the cluster center, which reduces the computational complexity and improves the clustering speed.

[0095] Embodiment 3

[0096] This embodiment provides a method for verifying the selection result according to the daily load curve, specifically:

[0097] Determine the data radius of each classification in the selected result and the degree of separation from other classifications according to the daily load curve;

[0098] Verifying the selected result according to the data radius and the separation degree;

[0099] The data radius is the distance between the sample farthest from the corresponding daily load curve in each classification and the daily load curve;

[0100] For each category, the distance between each sample in the category and the daily load curve corresponding to the category is calculated, and the maximum distance is taken as the data radius of the category;

[0101] The separation degree is the distance between the daily load curve corresponding to each classification and the other classification closest to it;

[0102] Among them, for each category, in order to determine the other categories closest to it, the distance between the daily load curve of the category and the daily load curves of other categories can be calculated respectively, and the other category with the smallest distance is selected as the category closest to it, and then the determined minimum distance value is used as the separation degree of the category;

[0103] Wherein, verifying the selected result according to the data radius and the separation degree includes:

[0104] Determine the difference between the separation degree and the data radius corresponding to each classification in the selected result;

[0105] Determine whether the difference corresponding to each category is greater than or equal to the third threshold, if so, the verification is passed, otherwise, the verification fails;

[0106] In order to ensure the clustering effect, the second threshold and the third threshold should satisfy the following relationship:

[0107] The third threshold value>the second threshold value;

[0108] That is to say, after selection, the data radius of each category is smaller than the second threshold. In order to ensure the difference between different categories, the distance between any two categories should be greater than 2*the second threshold. That is, a circle is drawn with the daily load curve of each category as the center and the data radius as the radius. Each circle should be separated from each other without intersection.

[0109] In order to adjust the selected result based on the daily load curve until the verification is passed, in an optional implementation, specifically:

[0110] Delete the categories whose difference is less than the third threshold;

[0111] Since the intersection state is mutual, we only need to delete one of the categories that does not meet the difference requirement;

[0112] In an optional implementation, if there are more than two deleted categories, clustering can be performed again based on all deleted sample data until the requirements are met;

[0113] In another optional implementation, specifically:

[0114] Deleting samples at the edge of a classification whose difference is less than a third threshold to adjust the corresponding data radius, so that the difference corresponding to the classification after deleting the edge samples is greater than or equal to the third threshold;

[0115] That is to say, for the categories whose difference is less than the third threshold, the categories involved can be drawn with the daily load curve of each category as the center and the data radius as the radius, that is, the categories are represented by circles. If the difference is less than the third threshold, it means that the circles involved are in a tangent or intersecting state. The following operations can be performed on one of the circles:

[0116] Delete the sample data located at the edge and close to the category closest to it, thereby reducing the radius of the corresponding circle, so that the circles in the tangent or intersecting state can be transformed into the separated state;

[0117] In an optional implementation, after deleting the edge sample data, the number of samples in the category may be counted again. If the number meets the preset range in the first embodiment, it is determined that the classification after deleting the edge sample is still valid, and the daily load curve is updated based on the deleted sample data.

[0118] In another optional embodiment, if the number of samples does not meet the requirement after deleting the edge sample data, the classifications involved in the intersecting or tangent circles can be merged, and after the merging, it is determined whether the data radius meets the requirement. If so, the merged classification is valid, and its daily load curve is updated based on the merged sample data. Otherwise, the edge sample data of the merged classification is deleted until its data radius meets the requirement.

[0119] Embodiment 4

[0120] Please refer to Figure 2 A distribution network load characteristic research terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in a distribution network load characteristic research method as described in any one of embodiments one to three is implemented.

[0121] In summary, the present invention provides a method and terminal for studying the load characteristics of a distribution network. When clustering, the characteristic vector of the sample data is first determined based on the distance between the sample data to ensure the difference between the determined characteristic vectors, thereby laying a good data foundation for subsequent clustering. After the initial classification is performed based on the determined characteristic vector and the initial classification result is obtained, the initial classification result is selected based on the sample data and the degree of aggregation to obtain the selected result, thereby ensuring the rationality of the degree of aggregation and the number of samples of each classification in the selected result. Then, the daily load curve of each corresponding power consumption mode is determined based on the selected result. Finally, the daily load curve is selected based on the distance between different classifications and each The aggregation degree of the classification is used to verify the selection results. If the verification fails, the selection results are adjusted based on the daily load curve until the verification passes. Only after the verification passes, the daily load curve to be identified is identified according to the daily load curve of each power consumption mode to determine its corresponding power consumption mode. Based on the distance between different sample data, the distance between different classifications and the aggregation degree of each classification, the initial classification, selection and verification are carried out in sequence to ensure the efficiency of clustering and the accuracy of clustering results. Then, the power consumption mode of the daily load curve to be identified is determined based on the accurate clustering results, thereby improving the accuracy of the power consumption attributes of the distribution network equipment in the identified distribution network.

[0122] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for studying load characteristics of a distribution network, characterized in that: Includes steps: S1. Obtain sample data to be clustered, and determine a feature vector for clustering based on the sample data; S2. Performing initial classification on the sample data according to the feature vector to obtain an initial classification result; S3, selecting the initial classification results to obtain selected results, and determining the corresponding daily load curves of each power consumption mode according to the selected results; S4, verifying the selected result according to the daily load curve, if the verification is passed, executing S5, otherwise, adjusting the selected result based on the daily load curve until the verification is passed; S5, according to the daily load curve of each power consumption mode to be identified daily load curve to be identified, determine the corresponding power consumption mode; Verifying the selected result according to the daily load curve in step S4 includes: Determine the data radius of each classification in the selected result and the degree of separation from other classifications according to the daily load curve; Verifying the selected result according to the data radius and the separation degree: determining the difference between the separation degree and the data radius corresponding to each category in the selected result, and judging whether the difference corresponding to each category is greater than or equal to a third threshold value, if so, the verification is passed, otherwise, the verification is failed; The data radius is the distance between the sample farthest from the corresponding daily load curve in each category and the daily load curve, and the separation degree is the distance between each category and the daily load curve corresponding to the other categories closest to it.

2. A method for studying load characteristics of a distribution network according to claim 1, characterized in that: The determining of the feature vector for clustering according to the sample data comprises: Determine two samples with the farthest distance in the sample data, and add the two samples to the reference sample set; Taking all samples in the reference sample set as reference, determining a sample that is farthest from all samples in the reference sample set and whose distances are all greater than a first threshold, adding the sample to the reference sample set, and returning to execute the step of taking all samples in the reference sample set as reference until no sample that meets the requirement exists; All samples in the reference sample set are determined as feature vectors for clustering.

3. A method for studying load characteristics of a distribution network according to claim 1, characterized in that: The selecting of the initial classification results to obtain the selected results includes: Counting the number of samples in each category in the initial classification result, and deleting the categories whose sample numbers do not fall within a preset range; Determine the centroid corresponding to each category in the classification result after deleting the category, delete the samples in each category whose distance to the centroid corresponding to the category is greater than the second threshold, and obtain the selected result.

4. A method for studying load characteristics of a distribution network according to claim 1, characterized in that: Determining the daily load curve of each corresponding power consumption mode according to the selected result includes: The sample average value of each category in the selected results is determined, and the sample average value of each category is determined as the daily load curve of the power consumption mode corresponding to each category.

5. A method for studying load characteristics of a distribution network according to claim 1, characterized in that: The adjusting the selected result based on the daily load curve until the verification is passed comprises: The categories whose difference is less than the third threshold are deleted.

6. A method for studying load characteristics of a distribution network according to claim 1, characterized in that: The adjusting the selected result based on the daily load curve until the verification is passed comprises: The samples at the edge of the classification whose difference is less than the third threshold are deleted to adjust the corresponding data radius, so that the difference corresponding to the classification after deleting the edge samples is greater than or equal to the third threshold.

7. A distribution network load characteristics research terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, each step of the method for studying load characteristics of a distribution network as described in any one of claims 1 to 6 is implemented.

Images