Hierarchical clustering method for operation data of high-proportion new energy power system
By using a hierarchical clustering method to classify operating mode data in the new energy power system, the problem of artificial experience subjectivity and unclear clustering results in the existing technology is solved, and more efficient identification and classification of power grid operation modes is achieved. , enhances the representativeness and interpretability of clustering results.
Patent Information
- Application Number
- CN202410994133.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-22
AI Technical Summary
The classification and sample extraction of existing technology in the classification and sample extraction of operating modes in new energy power systems have strong subjectivity based on artificial experience, insufficient consideration of multi-factor correlation, and the physical meaning of clustering centers and sample categories based on mean clustering algorithms. Clear problems lead to poor classification results.
A high-proportion new energy power system operation data hierarchical clustering method is proposed, including topological partitioning, feature quantity selection, data equal-value dimensionality reduction and hierarchical mean clustering algorithm, and the operating mode data set is divided through the system-led operation feature sequence. Combining manual experience, screen running points to form a representative operating mode classification set.
It effectively improves the shortcomings of the traditional mean clustering method, enhances the interpretability and representativeness of the clustering results, can better reflect the complex operating conditions of the new energy power system, and provide guarantees for the safety of the power grid operation.
Smart Images

Figure CN119004144A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the application field of power system operation mode sample data analysis, and in particular relates to a hierarchical clustering method for operation data of a high-proportion new energy power system. Background Art
[0002] The current research on typical extraction and sample generation of power systems can be divided into two categories: knowledge-driven methods based on expert experience and data-driven methods: Knowledge-driven extraction methods are often seen in engineering applications, such as planning thermal power installed capacity based on the annual maximum load. Therefore, scenario extraction includes hydropower generation, thermal power generation, maximum or minimum load operation modes, and other possible grid safety boundary operation modes due to seasonal changes. However, with the increase in the proportion of new energy and the complexity of the grid structure, the scenarios extracted based on manual experience are highly subjective, with fewer factors that can be considered, and it is difficult to simultaneously associate the operation scenarios caused by multiple factors. The data-driven methods for extracting typical operation scenarios of power grids include sample-driven random analysis methods and artificial intelligence methods such as machine learning. Among them, although the K-means and its improved algorithms have the advantages of high computational efficiency and easy implementation in operation mode extraction, they are sensitive to the initial center point, cluster shape and outliers, and lack high-dimensional data processing capabilities. When faced with data characteristics such as high dimensionality, nonlinearity, uncertainty and combinatorial explosion of power systems, their application effect is limited. As a result, when using K-means and its improved algorithms to classify operation mode samples, there are shortcomings such as unclear physical meaning of classification, lack of pertinence, and poor typicality of generated samples. Summary of the invention
[0003] In view of the fact that the operation mode in the new energy power system mentioned in the above background technology shows a de-typical trend, it is difficult for the typical mode selected based on artificial experience to fully reflect the impact of the complex operation conditions of new energy on the power grid. In addition, the operation mode sample extraction and classification method based on the mean clustering algorithm in the conventional data-driven method research has the problems of unclear physical meanings of cluster centers and sample categories. The present invention proposes a hierarchical clustering method for operation data of a high-proportion new energy power system, including the following steps:
[0004] Step 1. Topological partitioning: partition the power system topological network to form a power flow section;
[0005] Step 2. Feature selection: Select key characteristics that can characterize the operation of the power grid from the original data collected from the power grid company to construct the feature quantity under the operation mode;
[0006] Step 3. Data equivalence dimensionality reduction: Based on the results of the power system topology partitioning, the features with strong correlation are merged to achieve dimensionality reduction of the features;
[0007] Step 4. Use the hierarchical mean clustering algorithm to hierarchically cluster the power system operation data;
[0008] Step 5. Select typical operating modes: Perform N-1 verification on each sample in the operating mode to be verified sample set under the system static safety analysis framework, evaluate the risk level of system operation at the operating point, extract generalized typical operating modes, and use visualization methods to identify and analyze the operating characteristics of risk areas.
[0009] Furthermore, in step 1, the rule is set as follows: the initial direction of the active power flow of the tie line in the normal operation mode is set to be 1 in the positive direction, and -1 in the opposite direction, and 0 when there is no power flow on the line.
[0010] Furthermore, the key characteristics selected in step 2 include: active power on the load side, power flow on the grid side, output of synchronous machines on the source side, and output of new energy units on the source side.
[0011] Furthermore, the raw data collected from the power grid company in step 2 needs to be cleaned and preprocessed to achieve preliminary dimensionality reduction of the operation mode in the new feature space;
[0012] Furthermore, in step 3, the plant load data in each partition is linearly added, and the independent plant load characteristics are replaced by the overall regional load characteristics, so as to improve the clustering algorithm's extraction effect on key features and complete further dimensionality reduction of the operation mode data;
[0013] Furthermore, the hierarchical mean clustering algorithm includes the following process:
[0014] (1) For a data set formed by n types of labeled data, select one type of labeled data, reset its data weight to 1, and reset the weights of the remaining labeled data to 0, and perform the first-level mean clustering;
[0015] (2) Based on the clustering results of (1), the data set is divided into k first-level sub-data sets. For each sub-data set, the labels with a first-level clustering weight of 1 are set to zero, and the second-level data labels are selected for second-level mean clustering.
[0016] (3) Based on the clustering results of (2), continue to split the data set. For each second-level sub-data set, set the labels with a second-level clustering weight of 1 to zero, select the third data label for third-level mean clustering, and repeat this process until all n labels are exhausted;
[0017] (4) Extract the result of the nth layer of clustering as the final result of hierarchical clustering of the data set.
[0018] Through the above design scheme, the present invention can bring the following beneficial effects:
[0019] The present invention incorporates expert knowledge into the identification framework of the operating mode by the data-driven method. By establishing the system-dominant operating feature sequence, the operating mode data set is divided using the hierarchical clustering method to form an operating mode classification set represented by the cluster center. Then, based on differences in new energy penetration rate, load level, etc., operating points with boundary operating characteristics in each category are screened. After verification, it can provide strong support for the mode department to formulate power grid operation rules to ensure the safety of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The present invention is further described below with reference to the accompanying drawings and specific embodiments:
[0021] Figure 1 Select the cut-set section partition diagram formed by the main interconnection lines of a provincial power grid;
[0022] Figure 2 This is the active power distribution characteristic diagram of the operation mode under direct mean clustering;
[0023] Figure 3 Active power distribution characteristics of the operating mode under hierarchical mean clustering. DETAILED DESCRIPTION
[0024] In order to facilitate the understanding of the present application, the specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thoroughly understood.
[0025] Principle of hierarchical clustering method:
[0026] The hierarchical mean clustering method can effectively improve the defects of the traditional mean clustering method. It aims to reduce the diversity of samples within a class and enhance the interpretability of clustering results through a systematic, layer-by-layer screening process. The method first adjusts the weights of the n labels in the data set. In the initial stage, a high weight is assigned to the first key label and a low weight is assigned to other labels. Then, the first layer of mean clustering is completed based on this configuration. After clustering, the original data set is divided into k subsets, each corresponding to a cluster center. For each subset, the system resets the weight of the previous layer of key labels to low, sets a high weight for the next key label, and then performs the next layer of mean clustering. This process is iterated until all labels are clustered in turn. Finally, the clustering results extracted in the last layer form a hierarchical clustering structure of the entire data set, providing an explanation of the data distribution and internal relationships. The pseudo code of the hierarchical clustering method algorithm is as follows:
[0027]
[0028] The hierarchical mean clustering algorithm includes the following processes:
[0029] (1) For a data set formed by n types of labeled data, select one type of labeled data, reset its data weight to 1, and reset the weights of the remaining labeled data to 0, and perform the first-level mean clustering;
[0030] (2) Based on the clustering results of (1), the data set is divided into k first-level sub-data sets. For each sub-data set, the labels with a first-level clustering weight of 1 are set to zero, and the second-level data labels are selected for second-level mean clustering.
[0031] (3) Based on the clustering results of (2), continue to split the data set. For each second-level sub-data set, set the labels with a second-level clustering weight of 1 to zero, select the third data label for third-level mean clustering, and repeat this process until all n labels are exhausted;
[0032] (4) Extract the result of the nth layer of clustering as the final result of hierarchical clustering of the data set.
[0033] The present invention provides a hierarchical clustering method for operation data of a high-proportion new energy power system, and uses the hierarchical clustering method in the extraction of power system operation mode data, comprising the following steps:
[0034] Step 1. Topological partitioning: Before applying the hierarchical mean clustering method to extract data on typical operation modes of the power system, the topological network of the power system is first partitioned to form a flow section. The power direction and size of the inter-regional tie lines represent the scale of power exchange between the cut-set sections of the power grid area. The clustering result using the active direction data of the tie lines as the clustering feature realizes the characterization of the spatial distribution characteristics of the power system operation mode. The rule is set as follows: the initial direction of the active flow of the tie line in the normal operation mode is set to the positive direction as 1, the opposite direction as -1, and no flow on the line as 0.
[0035] Step 2. Feature selection: The original data collected from the power grid company needs to be cleaned and preprocessed, and the key characteristics that can characterize the operation of the power grid are selected: load-side active power, grid-side power flow, source-side synchronous machine output, source-side new energy unit output, etc. to construct the feature quantity under the operation mode, and the initial dimensionality reduction of the operation mode is achieved in the new feature space:
[0036]
[0037] is the active value of the load at time t, representing the load side characteristics; It is the active power flow direction or active power value of the line at time t, representing the grid-side characteristics; is the active output of the synchronous generator at time t, representing the source side characteristics; is the active power output of the new energy unit at time t, representing the source side characteristics;
[0038] Step 3. Data equivalence dimensionality reduction: According to the results of the power system topology partitioning, the features with strong correlation are merged to achieve dimensionality reduction of the features: Since the load characteristics of the power plants in the power system area are relatively small, the load characteristics of the power plants in each partition are combined. The data is linearly added, and the overall characteristics of the regional load are used to replace the independent plant load characteristics, which improves the clustering algorithm's extraction effect on key features and completes further dimensionality reduction. The operation mode data is expressed as p' t
[0039]
[0040] By partitioning the power system based on network topology characteristics and reducing the dimensionality of the data, merging features with strong regularity can not only effectively simplify the data set structure and reduce computational complexity, but also retain key data features, thereby improving data processing efficiency and the accuracy of model analysis.
[0041] Step 4. Use the hierarchical mean clustering algorithm to hierarchically cluster the power system operation data;
[0042] (1) For the dataset p' formed by 4 types of labeled data t , select load data as labels, reset its data weight to 1, and reset the weights of other labels to 0, and perform first-level mean clustering;
[0043] (2) Based on the first-level load data clustering results, the data set is divided into k sub-data sets. For each sub-data set, the labels of the load data with a weight of 1 are set to zero, and the second-level data line flow data are selected and its label is set to 1 for the second-level mean clustering;
[0044] (3) Based on the clustering results of the second line flow data, continue to split the data set. For each second-layer sub-data set, set the labels with a second-layer line flow clustering weight of 1 to zero, select the third type of data label to perform the third-layer thermal power output mean clustering, and repeat this process until all four types of labels are exhausted;
[0045] (4) The cluster center formed by the final cluster after the fourth layer of clustering is extracted and used as the final result of hierarchical clustering of the data set to form the sample set to be verified.
[0046] Step 5. Select typical operating modes: Perform N-1 verification on each sample in the operating mode to be verified sample set under the system static safety analysis framework, evaluate the risk level of system operation at the operating point, extract generalized typical operating modes, and use visualization methods to identify and analyze the operating characteristics of risk areas.
[0047] The simulation results of the present invention applied to the provincial power grid operation mode extraction example are as follows:
[0048] Among the historical flow data collected by a provincial power grid dispatching system: the period from March to June in 2023 with relatively large wind power output was selected, with a sampling interval of 30 minutes, including more than 300 stations of 220kV and above, and about 520 AC lines, forming a total of 3640 sets of active power data. Figure 1 The cut-section partitioning results formed by the main tie lines for the power system are divided into 11 areas. Table 1 shows the detailed information of the partition tie line table.
[0049] Table 1. District contact line table
[0050]
[0051] After partitioning and reducing the dimension of the data, according to The basic data set of the operating mode is formed, and the data set is hierarchical mean clustered in the order of load side, grid side, and source side characteristics. After completing the last layer of mean clustering, the class centers of all classes are extracted to form the sample set of the operating mode to be verified. The load and unit active output are divided into three operating sections: small, medium, and large according to the output range of less than 30%, 30%-70%, and more than 70%. The representative sample set of the operating mode formed after hierarchical clustering is shown in Table 2:
[0052] Table 2 Hierarchical clustering results of the operation modes of some provincial networks in 2023 (only the samples accounting for the top 35% of the clustering results are listed)
[0053]
[0054] Hierarchical clustering effect analysis:
[0055] 1. Diversity of operation modes: As can be seen from Table 2, the clustering results cover the typical operation modes of the province's power system under different loads (small, medium, and large), different types of energy combinations (wind power, photovoltaic, thermal power, and hydropower), and different flow direction configurations. The results can reflect the effective distinction of the hierarchical clustering method for the diverse operation modes of the high-proportion new energy power system;
[0056] 2. Representativeness of categories: The "class proportion" indicator of each category shows the representativeness of the cluster center relative to the total samples. A higher percentage indicates that these cluster centers can better represent a certain proportion of operating modes. For example, the class with sequence number 1 accounts for 8.10%, which indicates that the operating mode under this source-load characteristic accounts for a large proportion of the total samples and is a common operating mode;
[0057] 3. Comparison of the effect of source-side output ratio division (compared with the direct mean clustering method): Set the same final number of clusters, select two classes with similar characteristics after direct mean clustering and hierarchical mean clustering, and compare the active value distribution range of source-side output and load in each class. The direct mean clustering method only considers the numerical similarity (Euclidean distance) and ignores the physical scenario of the operation mode. The same class still contains complex operation modes. For example, the class contains thermal power active output ranging from 3258MW to 9207MW, and the class center sample value is 7500MW. The typicality of thermal power active output and load distribution between operation modes in the class is not obvious. Figure 2 As shown in Figure 2, after the mean clustering method is used to divide the operation mode samples within the class according to the characteristic sequence, the volatility between classes is significantly reduced, as shown in Figure 2. Figure 3 As shown; Therefore, this application proposes a knowledge data hybrid driven operation mode extraction method, which extracts generalized typical operation modes in high-proportion renewable energy power systems through a hierarchical clustering algorithm. By selecting key operation features, the sample operation scenarios are further refined, and the operation modes characterized by the features are systematically clustered. Simulation of historical data of provincial power grids shows that the hierarchical clustering method is superior to the traditional mean clustering method in capturing the diversity of operation modes and selecting representative methods.
[0058] The hierarchical clustering method of generalized typical operation mode data driven by knowledge data hybrid proposed in the present invention is used to extract typical operation modes of new power systems. The network topology, unit output, load and line flow information obtained from the power grid operation flow data are partitioned, encoded and dimensionally reduced, and the key operation feature sequence of the system is selected in combination with artificial experience. The basic data set of the operation mode is divided using the hierarchical partitioning clustering method to form an operation mode classification sample set represented by various centers. The operation data type set set by artificial experience is combined to carry out partitioned hierarchical mean clustering representation, which is more suitable for the clustering algorithm of power system operation mode feature division. It can reduce the difference between samples within the mean clustering result class, and enhance the physical interpretability of the clustering result.
[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.
Claims
1. A hierarchical clustering method for operation data of a high-proportion renewable energy power system, characterized in that: The following steps are involved: Step 1. Topological partitioning: partition the power system topological network to form a power flow section; Step 2. Feature selection: Select key characteristics that can characterize the operation of the power grid from the original data collected from the power grid company to construct the feature quantity under the operation mode; Step 3. Data equivalence dimensionality reduction: Based on the results of the power system topology partitioning, the features with strong correlation are merged to achieve dimensionality reduction of the features; Step 4. Use the hierarchical mean clustering algorithm to hierarchically cluster the power system operation data; Step 5. Select typical operating modes: Perform N-1 verification on each sample in the set of samples to be verified under the system static safety analysis framework, evaluate the risk level of system operation at the operating point, extract generalized typical operating modes, and use visualization methods to identify and analyze the operating characteristics of risk areas.
2. According to claim 1, a hierarchical clustering method for operation data of a high-proportion new energy power system is characterized in that: In step 1, the rule is set as follows: the initial direction of the active power flow of the tie line in the normal operation mode is set to be 1 in the positive direction, and -1 in the opposite direction, and 0 when there is no power flow on the line.
3. The hierarchical clustering method for operating data of a high-proportion new energy power system according to claim 1 is characterized in that: The key characteristics selected in step 2 include: load-side active power, grid-side power flow, source-side synchronous machine output, and source-side new energy unit output.
4. The hierarchical clustering method for operating data of a high-proportion new energy power system according to claim 3 is characterized in that: The raw data collected from the power grid company in step 2 needs to be cleaned and preprocessed to achieve preliminary dimensionality reduction of the operation mode in the new feature space.
5. The hierarchical clustering method for operating data of a high-proportion new energy power system according to claim 1 is characterized in that: In step 3, the plant load data in each partition are linearly added, and the overall regional load characteristics are used to replace the independent plant load characteristics, so as to improve the clustering algorithm's extraction effect on key features and complete further dimensionality reduction of the operation mode data.
6. The hierarchical clustering method for operating data of a high-proportion new energy power system according to claim 1 is characterized in that: The hierarchical mean clustering algorithm includes the following processes: (1) For a data set formed by n types of labeled data, select one type of labeled data, reset its data weight to 1, and reset the weights of the remaining labeled data to 0, and perform the first-level mean clustering; (2) Based on the clustering results of (1), the data set is divided into k first-level sub-data sets. For each sub-data set, the labels with a first-level clustering weight of 1 are set to zero, and the second-level data labels are selected for second-level mean clustering. (3) Based on the clustering results of (2), continue to split the data set. For each second-level sub-data set, set the labels with a second-level clustering weight of 1 to zero, select the third data label for third-level mean clustering, and repeat this process until all n labels are exhausted; (4) Extract the result of the nth layer of clustering as the final result of hierarchical clustering of the data set.
Citation Information
Cited By
Multi-dimensional typical operation mode library generation method for high-proportion new energy and related device
CN121478778A
