Virtual power plant flexible load historical operation data high-fidelity compression method and system

Abstract

The invention discloses a high-fidelity compression method and system for flexible load historical operation data of a virtual power plant. The method specifically comprises the following steps: S1, acquiring various data included in an operation scene of the virtual power plant; s2, preprocessing various data included in the virtual power plant operation scene to form an input matrix; s3, taking a cycle as a unit, taking a running cycle contained in the input matrix as an initial class, and then performing clustering until the number of the running cycles is reduced to a target number of representative cycles by a clustering result; s4, on the basis of the processing result of S3, taking each operation point in each representative cycle as an initial class, and then performing clustering until the number of the operation points is reduced to the target number of the representative operation points by the clustering result; and S5, restoring the processed data, and outputting the processed data. The virtual power plant information density can be compressed, and the system operation condition can be accurately reflected.

Description

Technical Field

[0001] The present invention relates to a high-fidelity compression method and system for virtual power plant flexibility load historical operation data used in the field of virtual power plant data processing. Background Art

[0002] Virtual power plants (VPPs) include adjustable loads, micropower sources, conventional loads, energy storage systems, and control devices. These load characteristics are closely tied to power production schedules and may exhibit significant natural cycle characteristics. Furthermore, renewable energy sources exhibit generation characteristics that are strongly correlated with natural cycles such as daily, weekly, and monthly, but the randomness and volatility of sources and loads are inconsistent across these cycles. This presents challenges for characterizing VPP operating scenarios and also poses significant challenges to methods for optimizing source and storage configuration based on typical cycles (e.g., daily and weekly).

[0003] The existing data compression method is the scenario approach. This approach uses multiple deterministic, typical scenarios to represent uncertain variables, enabling optimal planning for highly stochastic systems. Increasing the amount of typical scenario data can effectively improve configuration accuracy, but excessive operating points not only create redundancy but also render the problem computationally cumbersome, potentially rendering it unsolvable. Therefore, the key to optimizing virtual power plant configuration is how to leverage massive amounts of data on load, wind speed, and light intensity to extract key information and generate representative scenarios containing a limited number of operating points that accurately reflect system performance. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a high-fidelity compression method and system for the historical operating data of the flexibility load of a virtual power plant, which can compress the information density of the virtual power plant and accurately reflect the system operation status.

[0005] A technical solution to achieve the above-mentioned purpose is: a high-fidelity compression method for historical operating data of flexibility loads of a virtual power plant, comprising the following steps:

[0006] S1, obtain various data contained in the virtual power plant operation scenario;

[0007] S2, preprocesses various data contained in the virtual power plant operation scenario to form an input matrix;

[0008] S3, taking the running weeks contained in the input matrix as the initial clusters, and then clustering them until the clustering results reduce the number of running weeks to the target number of representative weeks;

[0009] S4, based on the processing result of S3, each operating point in each representative week is used as an initial cluster, and then clustering is performed until the clustering result reduces the number of operating points to the target number of representative operating points;

[0010] S5, restore the processed data and output the processed data.

[0011] Furthermore, S2 is specifically:

[0012] S201, supplement the missing data and process the data in the entire time period into the form of hourly units;

[0013] S202, using time as one dimension and different data types of different nodes as another dimension, to obtain a two-dimensional data matrix of the operating points;

[0014] S203, performing data standardization processing for each operating point, as shown in formula (1);

[0015]

[0016] Where: α nor is the data after standardization; α is the original data; min is the minimum value of this type of data; max is the maximum value of this type of data.

[0017] Furthermore, in S201, the method of supplementing data is a systematic sampling method or an interpolation method.

[0018] Furthermore, S3 is specifically:

[0019] S301, the Euclidean distance is used to measure the differences between the data of each operating week, as shown in formula (2);

[0020]

[0021] Where: E(A,B) is the Euclidean distance between running week A and running week B; x A1 to x An is the data in running week A; x B1 to x Bn This is the data from running week B;

[0022] S302, using the Ward minimum variance method as the criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class, as shown in formula (3);

[0023]

[0024] Where: is the center point of class I; A represents all the operating cycles contained in class I; |I| is the number of operating cycles in class I;

[0025] S303, calculating the sum of squares of the intra-class deviations of each class, that is, the sum of squares of the Euclidean distances of all running points in the class to the class center, as shown in formula (4);

[0026]

[0027] Where: DSS I is the sum of squares of intra-class deviations of class I;

[0028] S304, iteratively calculate the total sum of squares of deviations of the data, as shown in formula (5);

[0029]

[0030] Where: SDSS is the total deviation sum of squares of the data; data represents all the data in the data processing process;

[0031] S304, selecting two classes that minimize the total sum of squared deviations and merging them, and selecting the operating week closest to the center point of each class as the representative week of the class;

[0032] Repeat the above steps until the clustering results reduce the number of running weeks to the target number of representative weeks.

[0033] Furthermore, the number of running weeks contained in each class in the clustering result is the weight of the representative week.

[0034] Furthermore, S4 is specifically:

[0035] S401, using Euclidean distance to measure the differences between the operating point data of each representative operating week;

[0036] S402, using the Ward minimum variance method as a criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class;

[0037] S403, using weighted Euclidean distance to calculate the distance between the center points, as shown in formula (6);

[0038]

[0039] Where: WE(X,Y) is the weighted Euclidean distance between class X and class Y; σ A is the weight representing week A; |X| and |Y| are the number of elements in class X and class Y respectively;

[0040] S404: Merge two clusters that are adjacent and closest in time sequence, and select the center point of each cluster as the representative running point of the cluster;

[0041] The above steps are repeated until the clustering results reduce the number of operating points to the target number of representative operating points.

[0042] Furthermore, the number of operating points contained in each cluster in the clustering result is the weight of the representative operating point.

[0043] Furthermore, S5 is specifically as shown in formula (7);

[0044] α=α nor (max-min)+min (7)

[0045] Where: α is the restored data. nor The standardized data are obtained after data processing.

[0046] Furthermore, the data types in the virtual power plant operation scenario include sunlight levels, customs, load levels and electricity price data.

[0047] Another technical solution for achieving the above-mentioned purpose relates to a high-fidelity compression system for historical operating data of flexibility loads of a virtual power plant, comprising a data preprocessing module, a clustering module, and an output module;

[0048] The data preprocessing module obtains various data contained in the virtual power plant operation scenario and forms an input matrix after preprocessing;

[0049] The clustering module includes a representative week clustering module and a representative operating point clustering module. The representative week clustering module reduces the number of operating weeks in the input matrix to the target number of representative weeks; the representative operating point clustering module reduces the number of operating points in the representative week to the target number of representative operating points.

[0050] The output module restores the clustered data and outputs the restored data.

[0051] The advantages of this invention lie in the following: It proposes a high-fidelity compression method and system for historical operating data of flexible loads in virtual power plants. Building on existing results of dynamic self-optimization of flexible loads, it can dynamically adjust the operating point time scale according to operating characteristics. This method reduces data size by leveraging both the similarities between operating weeks and the similarities between adjacent operating points, preserving the properties of the original data and the correlations between different data types to the greatest extent possible, improving data information density and enabling the pooling and encapsulation of flexible load energy resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a high-fidelity compression method for historical operating data of a virtual power plant's flexibility load according to the present invention;

[0053] Figure 2 This is a schematic diagram of the architecture of a high-fidelity compression system for historical operating data of a virtual power plant's flexibility load according to the present invention. DETAILED DESCRIPTION

[0054] In order to better understand the technical solution of the present invention, the following is a detailed description through specific embodiments:

[0055] See also Figure 1A high-fidelity compression method for historical operating data of a virtual power plant's flexibility load according to the present invention comprises the following steps:

[0056] S1, obtain various data contained in the virtual power plant operation scenario.

[0057] The types of data in the virtual power plant operation scenario include sunlight levels, customs, load levels and electricity price data.

[0058] S2, preprocesses various types of data contained in the virtual power plant operation scenario to form an input matrix.

[0059] The data contained in the results of flexible load self-optimization may have differences in time scale and data type. Before applying the data, it is necessary to extract and integrate the various types of data contained in the scenario. Specifically, it includes:

[0060] S201: Supplement missing data by processing and supplementing the data in the entire time period into hourly units. The data supplementation method is a systematic sampling method or an interpolation method.

[0061] S202 uses time as one dimension and the different data types of different nodes as another dimension to obtain a two-dimensional data matrix of operating points for input. For example, for light level, customs, load level, and electricity price data, the four types of data containing three nodes correspond to 3*4=12. Since a year corresponds to 8760 hours, the input matrix format should be 8760×12.

[0062] S203, different types of data at different nodes may have large numerical differences. During data processing, the distance between data corresponding to different operating points needs to be calculated. In order to give all data the same weight, the data needs to be standardized. This project uses the Min-max standardization method. For each operating point, data standardization is performed as shown in formula (1);

[0063]

[0064] Where: α nor is the data after standardization; α is the original data; min is the minimum value of this type of data; max is the maximum value of this type of data.

[0065] S3, in units of weeks, takes the running weeks contained in the input matrix as initial clusters and then performs clustering until the clustering results reduce the number of running weeks to the target number of representative weeks.

[0066] In power grid operation data, data such as electricity and heat loads may not exhibit obvious daily patterns. To capture the fluctuation characteristics of the data over a longer period of time, it is necessary to use the similarities between operating weeks to reduce the number of operating points. First, all operating weeks included in the operation scenario are used as an initial class. Then, hierarchical clustering is used to reduce the number of classes until the number of remaining classes meets the desired target. Specifically:

[0067] S301, the Euclidean distance is used to measure the differences between the data of each operating week, as shown in formula (2);

[0068]

[0069] Where: E(A,B) is the Euclidean distance between running week A and running week B; x A1 to x An is the data in running week A; x B1 to x Bn This is the data from running week B.

[0070] S302, using the Ward minimum variance method as the criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class, as shown in formula (3);

[0071]

[0072] Where: is the center point of class I; A represents all the operating cycles contained in class I; |I| is the number of operating cycles in class I.

[0073] S303, calculating the sum of squares of the intra-class deviations of each class, that is, the sum of squares of the Euclidean distances of all running points in the class to the class center, as shown in formula (4);

[0074]

[0075] Where: DSS I is the sum of squares of the intra-class deviations of class I.

[0076] S304, the sum of squared deviations within a class can represent the degree of dispersion between the running cycles in a class and measure the clustering of elements in each class. When the number of running cycles of a class increases, its sum of squared deviations within the class will be greater than or equal to the original sum of squared deviations within the class. During the clustering process, the total sum of squared deviations of the data is iteratively calculated, as shown in formula (5);

[0077]

[0078] Where: SDSS is the total deviation sum of squares of the data; data represents all the data in the data processing process.

[0079] In step S304, because a smaller sum of squared deviations indicates a higher degree of clustering, the two clusters with the smallest total sum of squared deviations are merged. After clustering, this project uses the operating week closest to each cluster's center as the representative week for that cluster, rather than the cluster's center. This approach avoids reducing fluctuations between adjacent operating points within a representative week by taking the average of multiple weeks.

[0080] During data processing, each clustering algorithm run reduces the number of operating weeks by one. If the entire operating scenario contains i operating weeks and needs to be reduced to j representative weeks, clustering is performed ij times. The number of operating weeks included in each cluster in the clustering result is the weight of the representative week. Repeat this step until the clustering results reduce the number of operating weeks to the target number of representative weeks.

[0081] S4, based on the processing result of S3, each operating point in each representative week is used as an initial class, and then clustering is performed until the clustering result reduces the number of operating points to the target number of representative operating points.

[0082] The operation status of the power grid is closely related to the production operation behavior. When the operation task remains unchanged, the same operation state will last for a long time. Therefore, in order to further reduce the data size of typical scenarios, the similarity between the operation points can be used to dynamically adjust the time scale of each operation point to achieve the reconstruction of the operation scenario. For example, when the operation status does not change much, the time scale of the operation point is increased so that one operation point represents multiple hours to reduce the data size; when the operation status changes rapidly, the time scale of the operation point is reduced so that one operation point represents one hour to retain key operation information. Based on the data processing results of the previous stage, the continuous hierarchical clustering method is used to cluster the adjacent operation points in each representative week. Each operation point data contained in each representative week is used as an initial class. The classes with the smallest difference and adjacent in time series will be merged, and classes belonging to different representative weeks are not allowed to be merged. The difference between classes is measured by the distance between the class center points. The calculation formula of the class center point is the same as S3. Specifically:

[0083] S401, using Euclidean distance to measure the differences between the operating point data of each representative operating week.

[0084] S402, using the Ward minimum variance method as a criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class.

[0085] S403, using weighted Euclidean distance to calculate the distance between the center points, as shown in formula (6);

[0086]

[0087] Where: WE(X,Y) is the weighted Euclidean distance between class X and class Y; σ Ais the weight representing week A; |X| and |Y| are the number of elements in class X and class Y respectively.

[0088] S404 , by merging two clusters that are adjacent and closest in time sequence, select the center point of each cluster as the representative running point of the cluster.

[0089] As can be seen from S403, the weight of the representative week and the number of elements within the class are introduced in the distance calculation process. This method, on the one hand, amplifies the distance between the classes in the high-weight representative week, making it more cautious when reducing the operating points in the high-weight representative week; on the other hand, it avoids too many elements in each class. After the distance between each class is calculated, the two classes that are adjacent and closest in time are merged, and the center point of each class is taken as the representative operating point of the class. Repeat the above process until the number of representative operating points reaches the expected target, and the operating point reconstruction is completed. The number of operating points contained in each class is the weight of the representative operating point of the class. Repeat the above steps until the clustering result reduces the number of operating points to the target number of representative operating points.

[0090] S5, restore the processed data and output the processed data.

[0091] In the data preprocessing stage, the data in the running scene has been standardized, so the data processing results need to be restored to their original form to represent the correct meaning. The data restoration method corresponds to the Min-max normalization method, as shown in formula (7);

[0092] α=α nor (max-min)+min (7)

[0093] Where: α is the restored data. nor The standardized data are obtained after data processing.

[0094] The system of the present invention for high-fidelity compression of historical operating data of a virtual power plant's flexibility load is used to implement the above method, and includes a data preprocessing module 1, a clustering module 2 and an output module 3.

[0095] The data preprocessing module 1 obtains various data contained in the virtual power plant operation scenario and forms an input matrix after preprocessing;

[0096] Clustering module 2 includes a representative week clustering module and a representative operating point clustering module. The representative week clustering module reduces the number of operating weeks in the input matrix to the target number of representative weeks; the representative operating point clustering module reduces the number of operating points in the representative week to the target number of representative operating points.

[0097] The output module 3 performs restoration processing on the clustered data and outputs the restored data.

[0098] This invention reduces data size by leveraging the similarities between operating cycles and adjacent operating points, preserving the properties of the original data and the correlations between different data types to the greatest extent possible, thereby increasing data information density and achieving pooled packaging of flexible load energy resources.

[0099] Those skilled in the art should recognize that the above embodiments are merely intended to illustrate the present invention and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A high-fidelity compression method for historical operating data of flexibility loads of a virtual power plant, characterized in that: The steps include: S1, obtain various data contained in the virtual power plant operation scenario; S2, preprocesses various data contained in the virtual power plant operation scenario to form an input matrix; S3, taking the running weeks contained in the input matrix as the initial clusters, and then clustering them until the clustering results reduce the number of running weeks to the target number of representative weeks; S4, based on the processing result of S3, each operating point in each representative week is used as an initial cluster, and then clustering is performed until the clustering result reduces the number of operating points to the target number of representative operating points; S5, restore the processed data and output the processed data.

2. A high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 1, characterized in that: S2 is specifically: S201, supplement the missing data and process the data in the entire time period into the form of hourly units; S202, using time as one dimension and different data types of different nodes as another dimension, to obtain a two-dimensional data matrix of the operating points; S203, performing data standardization processing for each operating point, as shown in formula (1); Where: α nor is the data after standardization; α is the original data; min is the minimum value of this type of data; max is the maximum value of this type of data.

3. A high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 2, characterized in that: In S201, the method of supplementing data is a systematic sampling method or an interpolation method.

4. The high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 1 is characterized in that: S3 specifically: S301, the Euclidean distance is used to measure the differences between the data of each operating week, as shown in formula (2); Where: E(A,B) is the Euclidean distance between running week A and running week B; x A1 to x An is the data in running week A; x B1 to x Bn The data in running week B; S302, using the Ward minimum variance method as the criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class, as shown in formula (3); Where: is the center point of class I; A represents all the operating cycles contained in class I; |I| is the number of operating cycles in class I; S303, calculating the sum of squares of the intra-class deviations of each class, that is, the sum of squares of the Euclidean distances of all running points in the class to the class center, as shown in formula (4); Where: DSS I is the sum of squares of intra-class deviations of class I; S304, iteratively calculate the total sum of squares of deviations of the data, as shown in formula (5); Where: SDSS is the total deviation sum of squares of the data; data represents all the data in the data processing process; S304, selecting two classes that minimize the total sum of squared deviations and merging them, and selecting the operating week closest to the center point of each class as the representative week of the class; Repeat the above steps until the clustering results reduce the number of running weeks to the target number of representative weeks.

5. A high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 4, characterized in that: The number of running weeks contained in each class in the clustering result is the weight of the representative week.

6. The high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 1 is characterized in that: S4 is specifically: S401, using Euclidean distance to measure the differences between the operating point data of each representative operating week; S402, using the Ward minimum variance method as a criterion for merging classes in the hierarchical clustering process, and calculating the center point of each class; S403, using weighted Euclidean distance to calculate the distance between the center points, as shown in formula (6); Where: WE(X,Y) is the weighted Euclidean distance between class X and class Y; σ A is the weight representing week A; |X| and |Y| are the number of elements in class X and class Y respectively; S404: Merge two clusters that are adjacent and closest in time sequence, and select the center point of each cluster as the representative running point of the cluster; The above steps are repeated until the clustering results reduce the number of operating points to the target number of representative operating points.

7. A high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 6, characterized in that: The number of operating points contained in each class in the clustering result is the weight of the representative operating point.

8. The high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 1 is characterized in that: S5 is as shown in formula (7); a = a nor (max-min)+min (7) Where: α is the restored data. nor The standardized data are obtained after data processing.

9. The high-fidelity compression method for historical operating data of flexible load of a virtual power plant according to claim 1 is characterized in that: The types of data in the virtual power plant operation scenario include sunlight levels, customs, load levels and electricity price data.

10. A system for implementing the high-fidelity compression method for historical operating data of flexibility loads of a virtual power plant according to any one of claims 1 to 9, characterized in that: Includes data preprocessing module, clustering module and output module; The data preprocessing module obtains various data contained in the virtual power plant operation scenario and forms an input matrix after preprocessing; The clustering module includes a representative week clustering module and a representative operating point clustering module. The representative week clustering module reduces the number of operating weeks of the input matrix to the target number of representative weeks. The representative operating point clustering module reduces the number of operating points in the representative week to the target number of representative operating points; The output module restores the clustered data and outputs the restored data.

Images