Virtual power plant flexible resource dynamic aggregation method and system

Abstract

The invention discloses a flexible resource dynamic aggregation method and system for a virtual power plant. The method comprises the steps that the virtual power plant obtains a power grid frequency adjustment requirement, a flexible resource set and adjustment characteristics of the flexible resource set; the similarity between the resources is calculated, and clustering is carried out by using a Canopy algorithm and a K-medios algorithm; constructing an adjustment demand model and aggregation adjustment characteristics, and evaluating the difference between the two through a performance deviation function; aiming at different adjusting scenes, selecting performance indexes as constraint conditions; establishing a two-stage dynamic aggregation model by adopting a sub-modulo optimization method, and outputting a flexible resource aggregate meeting the power grid demand; according to the method, efficient and accurate dynamic aggregation of flexible resources of the virtual power plant is realized, and the accuracy and response speed of resource aggregation are ensured by optimizing the clustering process.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, and particularly relates to a method and system for dynamically aggregating flexible resources of a virtual power plant. Background Art

[0002] In recent years, with the rapid development of communication technology and the power Internet of Things, the large-scale deployment and application of diverse flexible resources in power systems have become an important trend in the construction of new power systems. Diverse flexible resources include distributed energy, energy storage systems, adjustable loads, etc., and their roles in frequency control, power balance, etc. are becoming increasingly prominent. As a typical representative of flexible resource aggregators, virtual power plants provide efficient regulation capabilities for the power grid by integrating and coordinating dispersed and heterogeneous resources, and have become an important bridge connecting resource individuals and control centers. Currently, virtual power plant technology has been widely applied in many countries and regions, and has demonstrated significant advantages in improving the flexibility, reliability, and economy of power systems.

[0003] In the prior art, the aggregation of flexible resources of virtual power plants usually adopts a method of fixed resource composition. The control center or virtual power plant operator will pre-determine the types, quantities, and connection methods of resources participating in the aggregation, and centrally manage and dispatch the resources based on fixed control strategies.

[0004] However, the fixed resource combination mode adopted by virtual power plants in the prior art lacks the ability to adapt to the dynamic changes of resources, and cannot fully utilize the flexibility and complementarity of resources, resulting in limited aggregation performance. This limitation makes it difficult for power systems to achieve efficient and accurate resource allocation when dealing with complex and changing regulation requirements, affecting the overall control effect. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a method for dynamically aggregating flexible resources of a virtual power plant that can achieve efficient and accurate resource allocation. On the other hand, a system for dynamically aggregating flexible resources of a virtual power plant is provided.

[0006] Technical Solution: A method for dynamically aggregating flexible resources of a virtual power plant according to the present invention includes:

[0007] Obtaining a set of flexible resources available for selection and the frequency regulation characteristics of each flexible resource in the set of flexible resources based on the frequency regulation requirements of the virtual power plant grid;

[0008] Calculating the similarity between each flexible resource, and using the Canopy algorithm to pre-cluster the flexible resources to generate several pre-clusters;

[0009] Extract the number of pre-clusters and the initial cluster centers from the pre-clusters, and run the K-mediods clustering algorithm to refine the clustering of flexible resources to obtain several aggregates;

[0010] Construct a regulation demand model for the virtual power plant to serve the grid and the aggregated regulation characteristics of the aggregates according to the frequency regulation characteristics, and construct a performance deviation function to evaluate the gap between the aggregate regulation characteristics and the regulation demand model and optimize both;

[0011] For different regulation scenarios, select the corresponding performance indicators as the constraints for dynamic aggregation;

[0012] According to the regulation demand model, the aggregated regulation characteristics of the aggregates and the constraints, establish a two-stage dynamic aggregation model under different scenarios by using the submodular optimization method, and obtain the dynamic aggregation results output by the two-stage dynamic aggregation model to form flexible resource aggregates that meet the grid regulation demand.

[0013] Through the above technical solutions, firstly, through the combination of the Canopy algorithm and the K-mediods clustering algorithm, the flexible resources can be classified quickly and accurately, reducing the computational complexity and at the same time improving the rationality of resource aggregation; secondly, based on the construction of the regulation demand model and the performance deviation function, the matching degree between the resource aggregates and the grid demand can be dynamically evaluated to ensure the accuracy and adaptability of resource allocation; in addition, through the two-stage dynamic aggregation model of submodular optimization, the performance indicators can be flexibly selected under different regulation scenarios to optimize the resource allocation, thus significantly improving the response speed and stability of the virtual power plant to grid frequency regulation; overall, this solution not only improves the resource utilization efficiency of the virtual power plant, but also enhances its flexibility and reliability in a complex grid environment, providing strong support for the safe and stable operation of the grid.

[0014] Preferably, the similarity between each flexible resource includes: the degree of similarity shown between different resource individuals in multiple characteristic spaces such as communication delay, dead zone, droop control gain, general response characteristics, and regulation capacity, and the Manhattan distance is used to represent the similarity between flexible resources e u and e v , where m is the number of characteristics considered when calculating the similarity, C j (e u ) is the jth characteristic of flexible resource e u , and C j (e v ) is the jth characteristic of flexible resource e vThe j-th characteristic; by introducing multiple characteristic spaces such as communication delay, dead zone, droop control gain, general response characteristic, and regulation capacity, and using Manhattan distance to quantify the similarity between flexible resources, the present invention can more comprehensively and accurately evaluate the matching degree between resources, thereby improving the accuracy and efficiency of the clustering algorithm. This multi-dimensional similarity calculation method not only enhances the rationality of resource aggregation but also optimizes the dynamic allocation ability of the virtual power plant for flexible resources, enabling it to better adapt to the complex regulation requirements of the power grid, and ultimately improving the resource utilization efficiency and the overall system performance.

[0015] Preferably, the Canopy algorithm includes: determining two distance thresholds T1 and T2 satisfying T1 < T2; randomly selecting a resource from the set M of flexible resources available for selection as the clustering center of the Canopy pre-cluster; dividing according to the distance thresholds, adding all resources with a distance less than T2 from the clustering center to the Canopy pre-cluster, and if the distance between a resource and the clustering center is less than T1, removing the resource from the set M to obtain a pre-cluster; returning to the step of randomly selecting a resource from the set M of flexible resources available for selection as the clustering center of the Canopy pre-cluster until the set M is empty, generating multiple pre-clusters; by adopting the Canopy algorithm and setting two distance thresholds T1 and T2, the present invention can efficiently pre-cluster flexible resources and quickly generate multiple pre-clusters. This algorithm randomly selects resources as clustering centers and dynamically divides resources according to distance thresholds, which not only reduces the computational complexity but also avoids the problem of improper selection of the initial clustering center in traditional clustering methods. At the same time, by removing resources with a distance less than T2, the clustering process is further optimized, improving the efficiency and accuracy of subsequent K-mediods refined clustering, thereby providing a more reliable basis for the dynamic aggregation of resources in the virtual power plant and significantly enhancing the efficiency and accuracy of resource allocation.

[0016] Preferably, the K-mediods clustering algorithm includes: obtaining the number k of pre-cluster bodies divided by Canopy, determining the pre-cluster body centers as the initial cluster centers, calculating the distances between each resource in the selectable flexible resource set M and the initial cluster centers, and assigning them to the clusters where the nearest initial cluster centers are located; for each cluster, replacing the current cluster center with any other resource within the cluster, calculating the total distance from all resources within the cluster to the new current cluster center, such that the total distance from all resources within the cluster to the new current cluster center is less than the total distance from all resources within the cluster to the old current cluster center; repeating the previous step until the cluster centers no longer change or reach the preset maximum number of iterations, obtaining a set of flexible resource clusters, where each cluster in the set represents an aggregate; the K-mediods algorithm ensures the minimization of the total distance of resources within the cluster by continuously optimizing the cluster centers, thereby generating a more reasonable and compact set of flexible resource clusters. This method not only reduces the sensitivity of the clustering result to the selection of the initial center but also improves the accuracy and efficiency of resource aggregation, providing a more reliable basis for the dynamic allocation of flexible resources in a virtual power plant and further enhancing the system's response ability and adaptability to grid regulation requirements.

[0017] Preferably, the construction of the virtual power plant to serve the grid regulation demand model and the aggregate regulation characteristics of the aggregates includes:

[0018] The regulation demand model is a frequency control equivalent model based on second-order expectation, specifically expressed as:

[0019]

[0020] where f g is the response dead zone, τ g is the time delay, K g is the droop control gain, is the regulation time constant, and they are all the desired dynamic frequency regulation characteristics of the grid. ΔP g is the regulation demand, Δf is the frequency regulation deviation, and D res is the grid's demand for regulation capacity;

[0021] The aggregate A is represented by A = {e1, …, e h}, where e h represents the h-th flexible resource in the aggregate. Each type of flexible resource has different regulation characteristics. Based on the basic linear power system load frequency control model, considering the communication delay, regulation mode, and distribution coefficient of each flexible resource, the regulation characteristics of the h-th flexible resource are simulated:

[0022]

[0023] where ΔP his the actual regulation power of the h-th flexible resource after coordination, P h is the unit regulation power of the h-th flexible resource, is the distribution coefficient after coordination of multiple flexible resources in the aggregate A, is the maximum regulation capacity that the h-th flexible resource can provide, K h is the droop control gain coefficient, is the regulation time constant of the h-th flexible resource in the general model, which depends on the type and state of the flexible resource, τ h is the time delay for communication and monitoring, f h is the frequency control threshold of the h-th flexible resource. The aggregate adjusts the aggregate regulation characteristics presented externally by coordinating the distribution coefficients of each resource to adjust the aggregate regulation characteristics presented by the aggregate externally;

[0024] The aggregate regulation characteristics of aggregate A are the Minkowski sum of the regulation powers among all aggregates. Therefore, the aggregate regulation characteristics of the aggregate can be expressed as:

[0025]

[0026] where n = |A| is the total number of flexible resources included in aggregate A.

[0027] By constructing a frequency control equivalent model based on the second-order expectation, and combining the communication delay, regulation mode and distribution coefficient of flexible resources, the regulation characteristics of flexible resources in the virtual power plant can be accurately simulated, and the external regulation characteristics of the aggregate can be dynamically adjusted by coordinating the distribution coefficients. The regulation characteristics of the aggregate are calculated by the Minkowski sum to ensure that it can accurately reflect the comprehensive regulation ability of multiple flexible resources. This modeling method not only improves the response accuracy of the virtual power plant to the grid frequency regulation demand, but also optimizes the dynamic performance of resource aggregation, enabling the virtual power plant to more efficiently and flexibly meet the complex regulation demands of the grid, thereby enhancing the overall stability and reliability of the system.

[0028] Preferably, the construction of the performance deviation function includes:

[0029] Taking the system frequency deviation Δf as the unit step function and inputting it into equations (17) and (20), and discretizing these two formulas, the discretized models of the regulation demand and the aggregate regulation characteristics can be obtained:

[0030]

[0031] where T = lt0 is the steady-state time of the unit step response, t0 and l are the time interval and the discrete sampling length respectively, P t g and P t hIt is the discretized representation of the unit regulation demand P g and the flexible resource unit regulation power P h ;

[0032] According to the discretized model of the regulation demand and the aggregated regulation characteristics, the performance deviation function is defined as:

[0033]

[0034] where and are the performance loss part and the performance surplus part of the aggregate distinguished based on the calculation of the frequency regulation mileage, and are respectively the sets of moments when the aggregate is in performance loss and performance surplus when providing frequency control.

[0035] By constructing the performance deviation function, the gap between the regulation characteristics of the virtual power plant aggregate and the grid regulation demand can be quantitatively evaluated. Specifically, by discretizing the regulation demand and the aggregated regulation characteristics model, and defining the performance loss and performance surplus parts, the deviation in the frequency regulation process can be accurately measured. This deviation function can not only dynamically identify the deficiencies or excesses of the aggregate in frequency regulation, but also provide a clear quantitative basis for optimizing resource allocation, thus significantly improving the response accuracy and adaptability of the virtual power plant to the grid frequency regulation demand, and further enhancing the system stability and resource utilization efficiency.

[0036] Preferably, the different regulation scenarios are represented as S = {S1, S2, S3}, which respectively focus on the resource quantity S1, resource quality S2 and resource cost S3 within the aggregate. Among them, the performance index corresponding to scenario S1 is the total number of flexible resources |A| contained in the resource quantity aggregate A within the aggregate, the performance index corresponding to scenario S2 is the aggregated regulation characteristic ΔP Agg of the aggregate, and the performance index corresponding to scenario S3 is the monotonically increasing aggregate cost function where and are the unit regulation capacity and the cost of the regulation power of the flexible resource e h ; By clearly dividing the different regulation scenarios into resource quantity, resource quality and resource cost, and defining the corresponding performance indexes (such as the total number of resources, aggregated regulation characteristics and cost function) for each scenario, the diverse grid regulation demands can be flexibly adapted. This scenario-based design enables the virtual power plant to dynamically optimize the resource aggregation strategy under different conditions, not only meeting the grid requirements for regulation capacity and quality, but also effectively controlling the regulation cost, thus achieving the multi-objective optimization of resource allocation and significantly improving the economy, flexibility and overall performance of the virtual power plant.

[0037] Preferably, the method of using submodular optimization to establish a two-stage dynamic aggregation model under different scenarios includes:

[0038]

[0039] where i is the scenario flag, formula (28) represents the scenario constraints under different scenarios, L is the resource quantity limit, ω is the allowable performance deviation when the aggregator is formed, and B a is the budget when the aggregator is aggregating; formula (27) is the regulation capacity constraint of the h-th flexible resource in the steady-state response, where the distribution coefficient The maximum value of reflects in the allowable deviation Since Therefore, for resource e h There exists a minimum response power

[0040] During the entire two-stage dynamic aggregation process, the upper layer stage is to select the optimal cluster A←c based on the greedy algorithm opt , to provide the optimal resource composition, and the cluster selection process can be expressed as:

[0041]

[0042] ΔP Agg (c) = ΔP A∪c -ΔP Agg (31)

[0043] ΔT cA (c) = T c (A∪c)-T c (c) (32)

[0044] where, F pd (A) represents the performance deviation between the aggregation regulation characteristics of aggregator A and the grid demand, F pd (A∪c) represents the performance deviation between the aggregation regulation characteristics of the new aggregator formed after cluster c is added to aggregator A and the grid demand, ΔP Agg is the aggregation regulation characteristic of aggregator A, ΔP A∪c is the aggregation regulation characteristic of the new aggregator formed after cluster c is added to aggregator A, T c (A) is the aggregation cost of aggregator A, T c (A∪c) is the aggregation cost of the new aggregator formed after cluster c is added to aggregator A;

[0045] The aggregator will select different clusters to join the aggregator according to different scenarios in the upper layer stage, and then hand it over to the coordination process of the lower layer resources to coordinate the resource distribution coefficient; the resource coordination process of the lower layer is to select A←c each time optAfter that, the allocation factors are coordinated to minimize the performance deviation through formula (26).

[0046] By adopting the submodular optimization, this hierarchical optimization strategy can not only dynamically adjust the resource aggregation strategy according to different scenarios (such as resource quantity, quality, and cost), but also achieve the efficiency and economy of resource allocation through constraint conditions and budget limitations. Finally, this model significantly improves the response ability of the virtual power plant to the grid regulation demand, optimizes the resource utilization efficiency, reduces the regulation cost, and enhances the flexibility and stability of the system.

[0047] A virtual power plant flexible resource dynamic aggregation system according to the present invention includes:

[0048] A data acquisition module, configured to obtain a set of selectable flexible resources and the frequency regulation characteristics of each flexible resource in the set of flexible resources based on the frequency regulation demand of the virtual power plant grid;

[0049] A pre-clustering module, configured to calculate the similarity between each flexible resource and pre-cluster the flexible resources using the Canopy algorithm to generate several pre-clusters;

[0050] A refined clustering module, configured to extract the number of pre-clusters and the initial clustering centers from the pre-clusters, and run the K-mediods clustering algorithm to refine the clustering of the flexible resources to obtain several aggregates;

[0051] A model construction and evaluation module, configured to construct a regulation demand model for the virtual power plant to serve the grid and the aggregate regulation characteristics of the aggregates according to the frequency regulation characteristics, and construct a performance deviation function to evaluate the gap between the aggregate regulation characteristics and the regulation demand model and optimize both;

[0052] A dynamic aggregation constraint module, configured to select corresponding performance indicators as the constraint conditions for dynamic aggregation for different regulation scenarios;

[0053] A dynamic aggregation optimization module, configured to establish a two-stage dynamic aggregation model under different scenarios using the method of submodular optimization according to the regulation demand model, the aggregate regulation characteristics of the aggregates, and the constraint conditions, and obtain the dynamic aggregation result output by the two-stage dynamic aggregation model to form a flexible resource aggregate that meets the grid regulation demand.

[0054] The pre-clustering module is specifically configured to illustrate the similarity between each flexible resource.

[0055] The pre-clustering module is specifically configured to use the Manhattan distance to represent the similarity degree of the flexible resource e u and e v

[0056] ​The refined clustering module is specifically configured to determine two distance thresholds T1 and T2 such that T1 < T2; randomly select a resource from the set M of flexible resources available for selection as the clustering center of the Canopy pre-cluster; perform partitioning according to the distance thresholds, add all resources whose distance from the clustering center is less than T2 to the Canopy pre-cluster, and if the distance of a resource from the clustering center is less than T1, remove the resource from the set M to obtain a pre-cluster; return the step of randomly selecting a resource from the set M of flexible resources available for selection as the clustering center of the Canopy pre-cluster until the set M is empty, generating multiple pre-clusters.

[0057] The refined clustering module is specifically configured to obtain the number k of pre-clusters obtained by Canopy partitioning, determine the pre-cluster centers as the initial clustering centers, calculate the distance between each resource in the set M of flexible resources available for selection and the initial clustering centers, and assign it to the cluster where the nearest initial clustering center is located; for each cluster, replace the current clustering center with any other resource within the cluster, calculate the total distance from all resources within the cluster to the new current clustering center, such that the total distance from all resources within the cluster to the new current clustering center is less than the total distance from all resources within the cluster to the old current clustering center; repeat the previous step until the clustering centers no longer change or reach a preset maximum number of iterations, obtaining a set of flexible resource clusters, where each cluster in the set represents an aggregate.

[0058] The model construction and evaluation module is specifically configured to construct a regulation demand model as a frequency control equivalent model based on second-order expectations. On the basis of the basic linear power system load frequency control model, consider the communication delay, regulation mode, and distribution coefficient of each flexible resource to simulate the regulation characteristics of the hth flexible resource and the aggregate regulation characteristics of the aggregate.

[0059] The model construction and evaluation module is specifically configured to construct a discretized model of the regulation demand and the aggregate regulation characteristics, and define a performance deviation function according to the discretized model of the regulation demand and the aggregate regulation characteristics.

[0060] The dynamic aggregation constraint module is specifically configured to construct different regulation scenarios S = {S1, S2, S3}, and respectively focus on the number of resources S1 within the aggregate, the resource quality S2, and the resource cost S3. Among them, the performance index corresponding to scenario S1 is the total number |A| of flexible resources included in the aggregate A, the performance index corresponding to scenario S2 is the aggregate regulation characteristic ΔP Agg , and the performance index corresponding to scenario S3 is a monotonically increasing aggregate cost function where and are the unit regulation capacity and the cost of the regulation power of the flexible resource e h respectively.

[0061] The dynamic aggregation optimization module is specifically configured to use the submodular optimization method to establish a two-stage dynamic aggregation model under different scenarios. During the entire two-stage dynamic aggregation process, the upper stage is the optimal cluster selection based on the greedy algorithm A←c opt to provide the optimal resource composition. The aggregator will select different clusters to join the aggregator according to different scenarios in the upper stage, and then hand it over to the resource coordination process of the lower layer to coordinate the resource allocation coefficient; the resource coordination process of the lower layer coordinates the allocation factors every time A←c opt is selected.

[0062] A computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the above-mentioned virtual power plant flexible resource dynamic aggregation method is realized.

[0063] A computer device includes a memory and a processor, and a computer program capable of being loaded and executed by the processor to implement the above-mentioned virtual power plant flexible resource dynamic aggregation method is stored on the memory.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: 1. It realizes the efficient and accurate dynamic aggregation of flexible resources of the virtual power plant, and ensures the accuracy and response speed of resource aggregation by optimizing the clustering process; 2. The multi-dimensional similarity evaluation method makes the resource aggregation more in line with the actual operating conditions, improving the overall performance and stability of the aggregator; 3. The two-stage dynamic aggregation model can not only meet the real-time regulation requirements of the power grid, but also find a balance between cost control and resource optimal allocation, enhancing the adaptability and economy of the virtual power plant in different operating environments; 4. The performance deviation function and the optimization adjustment based on it ensure that the aggregator can meet the regulation requirements of the power grid to the greatest extent during actual operation, reduce the performance deviation, and improve the service quality and reliability of the virtual power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is the method flow chart of the present invention;

[0066] Figure 2 is the schematic diagram of the Canopy algorithm process of the present invention;

[0067] Figure 3 is the schematic diagram of the K-mediods clustering algorithm process of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0068] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0069] Such as Figure 1As shown in the figure, a method for dynamically aggregating flexible resources in a virtual power plant according to the present invention includes the following steps:

[0070] (1) The virtual power plant first obtains the current frequency regulation requirements from the power grid dispatching center, including parameters such as the regulation amplitude and regulation speed. At the same time, the virtual power plant obtains the set of flexible resources available within it. These resources may include distributed power generation units, energy storage systems, adjustable loads, etc., as well as the frequency regulation characteristics of each flexible resource (such as the regulation range, response time, regulation accuracy, etc.);

[0071] (2) The virtual power plant calculates the similarity between each flexible resource. The similarity calculation can be based on the regulation characteristics of the resources, such as the regulation range, response time, regulation accuracy, etc. The Canopy algorithm is used to pre-cluster the flexible resources to generate several pre-cluster bodies. The Canopy algorithm quickly divides the resources by setting two thresholds (a loose threshold and a tight threshold) to form a preliminary clustering result;

[0072] (3) The number of clusters and the initial cluster centers are extracted from the pre-cluster bodies generated by the Canopy algorithm, and then the K-mediods clustering algorithm is run to refine the clustering of the flexible resources. The K-mediods algorithm iteratively optimizes and selects the center point (mediod) of each cluster to minimize the distance between the resources within the cluster and the center point, thereby obtaining a more accurate clustering result;

[0073] (4) The virtual power plant constructs a regulation demand model serving the power grid, which describes the specific requirements of the power grid for frequency regulation. At the same time, the virtual power plant constructs an aggregated regulation characteristic model of the aggregator, which describes the overall regulation characteristics of each cluster. By constructing a performance deviation function, the gap between the aggregated regulation characteristics and the regulation demand model is evaluated to ensure that the aggregator can meet the regulation requirements of the power grid;

[0074] (5) For different regulation scenarios, the corresponding performance indicators are selected as the constraints for dynamic aggregation;

[0075] (6) Using the submodular optimization method, a two-stage dynamic aggregation model for different scenarios is established. The first-stage model is used to preliminarily screen the flexible resource aggregators that meet the basic regulation requirements; the second-stage model is used to further optimize the performance of the aggregator to ensure its optimal performance in a specific scenario. Finally, the model outputs the dynamic aggregation result to form a flexible resource aggregator that meets the regulation requirements of the power grid.

[0076] The similarity between flexible resources in Step 2 refers to the degree of similarity shown between different resource individuals in a space containing multiple characteristics. Specifically, considering multiple characteristics such as communication delay, dead zone, droop control gain, general response characteristics, and regulation capacity, the Manhattan distance is used to represent the similarity between flexible resources:

[0077]

[0078] where m is the number of characteristics considered when calculating similarity, C j (e u ) is the j-th characteristic of flexible resource e u , and C j (e v ) is the j-th characteristic of flexible resource e v .

[0079] As Figure 2 shown, the Canopy algorithm in Step 2 is specifically as follows: First step, determine two distance thresholds T1 and T2 and satisfy T1 < T2; Second step, randomly select a resource from the set M of available flexible resources as the clustering center of the Canopy pre-cluster; Third step, perform partitioning according to the distance threshold, add all resources with a distance less than T2 to the current Canopy pre-cluster. If the distance between a resource and the clustering center is less than T1, then remove this resource from the set M. At this time, a pre-cluster can be obtained; Repeat the second step and the third step until the set M is empty, generating multiple pre-clusters.

[0080] As Figure 3 shown, the K-medoids clustering algorithm in Step 3 is specifically as follows: First step, based on the Canopy partitioning, given the number of clusters k and the initial clustering centers, then for each resource in the set M, calculate its distance to all initial clustering centers, and assign it to the cluster where the nearest initial clustering center is located; Second step, for each cluster, try to replace the current clustering center with other resources in the cluster, calculate the total cost after replacement (i.e., the total distance from all resources in the cluster to the new clustering center). If the total cost after replacement decreases, then update the clustering center; Repeat the above steps until the clustering center no longer changes or reaches the maximum number of iterations. Finally, a set N = {c1,…,c l ,…c k} of available flexible resource clusters can be obtained, where cl represents the l-th flexible resource cluster in the set N.

[0081] In cluster analysis, a "cluster" refers to a set of flexible resources with similar characteristics; a "cluster center" is a representative resource individual within the cluster, whose role is to serve as a benchmark for measuring the similarity of other resources within the cluster; a "clustering entity" is the ultimately formed resource aggregation unit that meets the grid regulation requirements, composed of the resources of the entire cluster and its cluster center. The hierarchical relationship among the three is as follows: clusters are formed by dividing through the cluster center, and after performance optimization, the clusters become clustering entities that can directly serve the power grid.

[0082] Throughout the clustering process, the system always maintains real-time monitoring of the clustering quality. Each iteration evaluates the overall compactness of the current clustering scheme, that is, the sum of the distances from all resource points to the center of their respective clusters. The smaller this value, the better the clustering effect. Through this continuously optimized method, a set of representative flexible resource clustering entity sets can ultimately be obtained. Each clustering entity is characterized by its center point, which is the optimal representative point selected from the actual resource points and can reflect the overall characteristics of the clustering entity to the greatest extent.

[0083] The regulation demand model of the virtual power plant serving the power grid in Step 4 is an equivalent model for frequency control based on second-order expectation, specifically expressed as:

[0084]

[0085] where f g is the response dead zone, τ g is the time delay, K g is the droop control gain, is the regulation time constant, and they are all the dynamic frequency regulation characteristics expected by the power grid. ΔP g is the regulation demand, Δf is the frequency regulation deviation, D res is the power grid's demand for regulation capacity. The greater the regulation capacity demand, the higher the requirement for the overall regulation capacity that the aggregation body can provide. This means that the aggregation body may need to aggregate more flexible resources. P g is the unit regulation demand. The greater the unit regulation demand, the higher the power grid's demand for frequency control and the higher the requirement for the fast dynamic regulation ability of the aggregation body.

[0086] Considering that any n flexible resources are aggregated in aggregation body A to provide frequency control services to the power grid, where n = |A| is the total number of flexible resources included in aggregation body A. Aggregation body A can be represented as A = {e1,…,e h ,…,e n}, e hRepresents the h-th flexible resource in the aggregator. Each flexible resource has different regulation characteristics. Based on the basic linear power system load frequency control model, considering the communication delay, regulation mode, and distribution coefficient of each flexible resource, the regulation characteristics of the h-th flexible resource are simulated:

[0087]

[0088] Among them, ΔP h is the actual regulation power of the h-th flexible resource after coordination, and P h is the unit regulation power of the h-th flexible resource (when uncoordinated, it is the regulation power of the flexible resource under the unit frequency deviation of the flexible resource). is the distribution coefficient after coordination of multiple flexible resources in the aggregator A. is the maximum regulation capacity that the h-th flexible resource can provide, and K h is the droop control gain coefficient. is the regulation time constant of the h-th flexible resource in the general model, which depends on the type and state of the flexible resource, and τ h is the time delay for communication and monitoring, and f h is the frequency control threshold of the h-th flexible resource. The aggregator adjusts the aggregated regulation characteristics presented by the aggregator by coordinating the distribution coefficients of each resource.

[0089] Theoretically, the aggregated regulation characteristics of aggregator A are the Minkowski sum of the regulation powers among all aggregators. Therefore, the aggregated regulation characteristics of the aggregator can be expressed as:

[0090]

[0091] The steady-state frequency response under the unit step signal can clearly describe the actual frequency regulation power generated by the aggregated regulation characteristics and the regulation demand model for the power grid when there is a large frequency deviation in the power grid. Therefore, by taking the system frequency deviation Δf as the unit step function and inputting it into equations (34) and (37), and discretizing these two equations, the discretized models of the regulation demand and the aggregated regulation characteristics can be obtained:

[0092]

[0093] Among them, T = lt0 is the steady-state time of the unit step response, t0 and l are the time interval and the discrete sampling length respectively, and P t g and P t h are for the unit regulation demand P g and the flexible resource unit regulation power P hDiscretized representation.

[0094] The performance deviation function in step 4 is defined as:

[0095]

[0096] Where, and are the performance loss part and performance surplus part of the aggregate distinguished based on the calculation of the frequency regulation mileage, specifically:

[0097]

[0098] Where, and are respectively the sets of moments when the aggregate is in performance loss and performance surplus when providing frequency control.

[0099] Three different scenarios in step 5 are represented as S = {S1, S2, S3}. These three scenarios respectively focus on the resource quantity S1, resource quality S2, and resource cost S3 within the aggregate; among them, the performance index corresponding to scenario S1 is the total number of flexible resources |A| contained in the resource quantity aggregate A within the aggregate. This index reflects the aggregation ability of the aggregate in terms of quantity scale. When the system requires a large-scale resource response, the aggregation scheme in this scenario will give priority to ensuring the quantity scale of the participating resources; the performance index corresponding to scenario S2 is the aggregation regulation characteristic ΔP Agg , in this scenario, the system will preferentially select those resources with excellent dynamic characteristics for aggregation, even if this means that the aggregation scale may decrease; the performance index corresponding to scenario S3 is a monotonically increasing aggregation cost function Where and are the unit regulation capacity and the cost of the regulation power of the flexible resource e h , in this scenario, the economic cost of different aggregation schemes will be calculated in real-time during the aggregation process, and it will tend to select the resource allocation scheme with the optimal cost-effectiveness. By introducing this scenario, the system can take into account the economic rationality of the aggregation process while meeting the technical requirements; these scenarios provide additional scenario constraints for the dynamic aggregation process.

[0100] The two-stage dynamic aggregation model based on submodular optimization in step 6 is:

[0101]

[0102] Where, i is the scenario flag, formula (45) represents the scenario constraints under different scenarios, L is the resource quantity limit, ω is the allowable performance deviation when the aggregate is formed, B aIt is the budget of the aggregator during aggregation. Formula (44) is the regulation capacity constraint of the h-th flexible resource transformed from the formula in the steady-state response, where the distribution coefficient The maximum value of Due to For resource e h There exists a minimum response power During the entire two-stage dynamic aggregation process, the upper stage is the optimal cluster selection A←c based on the greedy algorithm opt , to provide the optimal resource composition. This cluster selection process can be expressed as:

[0103]

[0104] ΔP Agg (c)=ΔP A∪c -ΔP Agg (48)

[0105] ΔT cA (c)=T c (A∪c)-T c (c) (49)

[0106] Among them, F pd (A) represents the performance deviation between the aggregation regulation characteristics of aggregator A and the grid demand. F pd (A∪c) represents the performance deviation between the aggregation regulation characteristics of the new aggregator formed after cluster c is added to aggregator A and the grid demand. ΔP Agg is the aggregation regulation characteristic of aggregator A. ΔP A∪c is the aggregation regulation characteristic of the new aggregator formed after cluster c is added to aggregator A. T c (A) is the aggregation cost of aggregator A. T c (A∪c) is the aggregation cost of the new aggregator formed after cluster c is added to aggregator A;

[0107] The aggregator will select different clusters to join the aggregator according to different scenarios in the upper stage, and then hand it over to the coordination process of the lower-layer resources to coordinate the distribution coefficient of the resources. The lower-layer resource coordination process coordinates the distribution factors after each selection of A←c opt to minimize the performance deviation through (43).

[0108] Based on the above method, the embodiments of the present application also disclose a virtual power plant flexible resource dynamic aggregation system, which includes the following modules:

[0109] A data acquisition module, configured to obtain a set of flexible resources available for selection and the frequency regulation characteristics of each flexible resource in the set of flexible resources based on the frequency regulation requirements of the virtual power plant grid;

[0110] A pre-clustering module, configured to calculate the similarity between flexible resources and use the Canopy algorithm to pre-cluster the flexible resources to generate a number of pre-clusters;

[0111] A refinement clustering module, configured to extract the number of pre-clusters and the initial clustering centers from the pre-clusters, and run the K-mediods clustering algorithm to perform refined clustering on the flexible resources to obtain a number of aggregates;

[0112] A model construction and evaluation module, configured to construct a regulation demand model for the virtual power plant to serve the grid and the aggregate regulation characteristics of the aggregates according to the frequency regulation characteristics, and construct a performance deviation function to evaluate the gap between the aggregate regulation characteristics and the regulation demand model and optimize both;

[0113] A dynamic aggregation constraint module, configured to select corresponding performance indicators as the constraint conditions for dynamic aggregation for different regulation scenarios;

[0114] A dynamic aggregation optimization module, configured to establish a two-stage dynamic aggregation model for different scenarios by using the submodular optimization method according to the regulation demand model, the aggregate regulation characteristics of the aggregates, and the constraint conditions, and obtain the dynamic aggregation results output by the two-stage dynamic aggregation model to form an aggregate of flexible resources that meets the grid regulation requirements.

[0115] The pre-clustering module is specifically configured to illustrate the similarity between flexible resources.

[0116] The pre-clustering module is specifically configured to use the Manhattan distance to represent the similarity degree of flexible resources e u and e v .

[0117] The refinement clustering module is specifically configured to determine two distance thresholds T1 and T2 and satisfy T1 < T2; randomly select a resource from the set of flexible resources M available for selection as the clustering center of the Canopy pre-cluster; perform partitioning according to the distance thresholds, add all resources with a distance less than T2 from the clustering center to the Canopy pre-cluster, and if the distance between the resource and the clustering center is less than T1, remove the resource from the set M to obtain a pre-cluster; return the step of randomly selecting a resource from the set of flexible resources M available for selection as the clustering center of the Canopy pre-cluster until the set M is empty to generate multiple pre-clusters.

[0118] The refined clustering module is specifically used to obtain the number k of pre-cluster bodies divided by Canopy, determine the pre-cluster body centers as the initial clustering centers, calculate the distances between each resource in the set M of flexible resources available for selection and the initial clustering centers, and assign them to the clusters where the nearest initial clustering centers are located; for each cluster, replace the current clustering center with any other resource within the cluster, calculate the total distance from all resources within the cluster to the new current clustering center, such that the total distance from all resources within the cluster to the new current clustering center is less than the total distance from all resources within the cluster to the old current clustering center; repeat the previous step until the clustering centers no longer change or reach the preset maximum number of iterations, obtaining a set of flexible resource clustering bodies, and each cluster in the set represents an aggregate body.

[0119] The model construction and evaluation module is specifically used to construct a regulation demand model as an equivalent model of frequency control based on second-order expectation. On the basis of the basic linear power system load frequency control model, considering the communication delay, regulation mode, and distribution coefficient of each flexible resource, it simulates the regulation characteristics of the h-th flexible resource and the aggregate regulation characteristics of the aggregate body.

[0120] The model construction and evaluation module is specifically used to construct a discretized model of regulation demand and aggregate regulation characteristics, and define a performance deviation function according to the discretized model of regulation demand and aggregate regulation characteristics.

[0121] The dynamic aggregation constraint module is specifically used to construct different regulation scenarios S = {S1, S2, S3}, focusing on the number of resources S1, resource quality S2, and resource cost S3 within the aggregate body respectively. Among them, the performance index corresponding to scenario S1 is the total number |A| of flexible resources included in the aggregate body A of the number of resources within the aggregate body, the performance index corresponding to scenario S2 is the aggregate regulation characteristic ΔP of the aggregate body Agg , and the performance index corresponding to scenario S3 is a monotonically increasing aggregate cost function where and are the unit regulation capacity and the cost of regulation power of the flexible resource e h .

[0122] The dynamic aggregation optimization module is specifically used to adopt the method of submodular optimization to establish a two-stage dynamic aggregation model under different scenarios. During the entire two-stage dynamic aggregation process, the upper stage is the selection of the optimal clustering body A←c based on the greedy algorithm opt , to provide the optimal resource composition. The aggregate body will select different clustering bodies to join the aggregate body according to different scenarios in the upper stage, and then hand it over to the resource coordination process at the lower layer to coordinate the resource distribution coefficients; the resource coordination process at the lower layer coordinates the distribution factors after each selection of A←c opt .

[0123] An embodiment of the present invention also discloses a computer-readable storage medium.

[0124] Specifically, the computer-readable storage medium is used to store a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented. Those skilled in the art can understand that all or part of the processes in implementing the methods in the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memories.

[0125] An embodiment of the present invention also discloses a computer device.

[0126] Specifically, the computer device can be a desktop computer, a laptop computer, a palm computer, a cloud server, and other computer devices. The computer device can include, but is not limited to, a processor and a memory. Among them, the processor and the memory can be connected through a bus or other means. Among them, the processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, graphics processing units (GPUs), embedded neural network processors (NPUs), or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or a combination of the above types of chips.

[0127] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above embodiments of the present application. By running the non-transitory software programs, instructions, and modules stored in the memory, the processor executes various functional applications and data processing of the processor, that is, implements the methods in the above method embodiments. The memory may include a program storage area and a data storage area. Among them, the program storage area can store control units and application programs required for at least one function; the data storage area can store data created by the processor and the like. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

Claims

1. A dynamic aggregation method for flexible resources in a virtual power plant, characterized in that, Including: Obtaining a set of alternative flexible resources and the frequency regulation characteristics of each flexible resource in the set according to the frequency regulation requirements of the virtual power plant power grid; Calculating the similarity between each flexible resource, and using the Canopy algorithm to pre-cluster the flexible resources to generate several pre-clusters; Extracting the number of pre-clusters and the initial clustering center from the pre-clusters, and running the K-mediods clustering algorithm to refine the clustering of the flexible resources to obtain several aggregates; Constructing a regulation demand model for the virtual power plant to serve the power grid and the aggregate regulation characteristics of the aggregates according to the frequency regulation characteristics, and constructing a performance deviation function to evaluate the gap between the aggregate regulation characteristics and the regulation demand model and optimize the two; For different regulation scenarios, select the corresponding performance indicators as the constraints for dynamic aggregation; According to the regulation demand model, the aggregate regulation characteristics of the aggregates and the constraints, establish a two-stage dynamic aggregation model under different scenarios by using the submodular optimization method, and obtain the dynamic aggregation result output by the two-stage dynamic aggregation model to form a flexible resource aggregate that meets the power grid regulation requirements.

2. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 1, wherein The similarity between each flexible resource includes: the degree of similarity shown between different resource individuals in multiple characteristic spaces such as communication delay, dead zone, droop control gain, general response characteristics, and regulation capacity.

3. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 1, wherein Using the Manhattan distance to represent the similarity between flexible resources e u and e v , where m is the number of features considered when calculating similarity, C j (e u ) is the j-th feature of flexible resource e u , and C j (e v ) is the j-th feature of flexible resource e v .

4. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 1, wherein The Canopy algorithm includes: Determining two distance thresholds T1 and T2 and satisfying T1 < T2; Randomly selecting a resource from the set M of alternative flexible resources as the clustering center of the Canopy pre-cluster; Dividing according to the distance threshold, adding all resources with a distance less than T2 from the clustering center to the Canopy pre-cluster, and if the distance between the resource and the clustering center is less than T1, removing the resource from the set M to obtain a pre-cluster; Return to the step of randomly selecting a resource from the set M of alternative flexible resources as the clustering center of the Canopy pre-cluster until the set M is empty, and generating multiple pre-clusters.

5. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 4, characterized in that The K-mediods clustering algorithm includes: Obtaining the number k of pre-clusters divided by Canopy, determining the pre-cluster center as the initial clustering center, calculating the distance between each resource in the set M of alternative flexible resources and the initial clustering center, and assigning it to the cluster where the nearest initial clustering center is located; For each cluster, replacing the current clustering center with any other resource in the cluster, and calculating the total distance from all resources in the cluster to the new current clustering center, so that the total distance from all resources in the cluster to the new current clustering center is less than the total distance from all resources in the cluster to the old current clustering center; Repeating the previous step until the clustering center no longer changes or reaches the preset maximum number of iterations, obtaining a set of flexible resource clustering bodies, and each cluster in the set represents an aggregate.

6. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 1, wherein The construction of the regulation demand model for the virtual power plant to serve the power grid and the aggregate regulation characteristics of the aggregates includes: The regulation demand model is a frequency control equivalent model based on second-order expectation, specifically expressed as: Among them, f g is the response dead zone, τ g is the time delay, K g is the droop control gain, is the regulation time constant, and they are all the desired dynamic frequency regulation characteristics of the power grid. ΔP g is the regulation demand, Δf is the frequency regulation deviation, and D res is the power grid's demand for regulation capacity; The aggregate A is represented by A = {e1, …, e h}, where e h represents the h-th flexible resource in the aggregate. Each flexible resource has different regulation characteristics. Based on the basic linear power system load frequency control model, the regulation characteristics of the h-th flexible resource are simulated considering the communication delay, regulation mode, and distribution coefficient of each flexible resource: Among them, ΔP h is the actual regulation power of the h-th flexible resource after coordination, P h is the unit regulation power of the h-th flexible resource, is the allocation coefficient after coordination of multiple flexible resources in the aggregate A, is the maximum regulation capacity that the h-th flexible resource can provide, K h is the droop control gain coefficient, is the regulation time constant of the h-th flexible resource, which depends on the type and state of the flexible resource, τ h is the time delay for communication and monitoring, f h is the frequency control threshold of the h-th flexible resource. The aggregate adjusts the aggregate regulation characteristics presented by the aggregate by coordinating the allocation coefficients of each resource ; The aggregation regulation characteristic of Aggregate A is the Minkowski sum of the regulation powers among all aggregates. The aggregation regulation characteristic of an aggregate is expressed as: where n = |A| is the total number of flexible resources included in Aggregate A.

7. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 6, wherein The constructed performance deviation function includes: Input the system frequency deviation Δf as a unit step function into Formulas (1) and (4), and discretize these two formulas to obtain the discretized models of the regulation demand and the aggregation regulation characteristic: where T = lt0 is the steady-state time of the unit step response, t0 and l are the time interval and the discrete sampling length, respectively, P t g and P t h are the discretized representations of the unit regulation demand P g and the unit regulation power P of flexible resources h ; According to the discretized models of the regulation demand and the aggregation regulation characteristic, the performance deviation function is defined as: Among them, and are the performance loss part and the performance surplus part of the aggregate distinguished based on the calculation of the frequency regulation mileage, and are the sets of moments when the aggregate is in performance loss and performance surplus respectively when providing frequency control.

8. The method for dynamically aggregating flexible resources of a virtual power plant according to claim 7, characterized in that The different adjustment scenarios are represented as S = {S1, S2, S3}, focusing on the resource quantity S1, resource quality S2, and resource cost S3 within the aggregate, respectively. Among them, the performance metric corresponding to scenario S1 is the total number of flexible resources |A| contained in the resource quantity aggregate A within the aggregate, and the performance metric corresponding to scenario S2 is the aggregate adjustment characteristic ΔP of the aggregate Agg , and the performance metric corresponding to scenario S3 is a monotonically increasing aggregate cost function where and are the unit adjustment capacity and the cost of the adjustment power of the flexible resource e h .

9. The virtual power plant flexible resource dynamic aggregation method according to claim 8, characterized in that The method of using submodular optimization to establish a two-stage dynamic aggregation model under different scenarios includes: where i is the scenario flag, formula (12) represents the scenario constraints under different scenarios, L is the resource quantity limit, ω is the allowable performance deviation when the aggregate is formed, and B a is the budget when the aggregate is aggregated; formula (11) is the adjustment capacity constraint of the h-th flexible resource in the steady-state response, where the distribution coefficient The maximum value of reflects in the allowable deviation Since Therefore, for resource e h there exists a minimum response power During the entire two-stage dynamic aggregation process, the upper stage is the selection of the optimal cluster A←c based on the greedy algorithm opt , to provide the optimal resource composition, and the cluster selection process can be expressed as: ΔP Agg (c) = ΔP A∪c -ΔP Agg (15) ΔT cA (c) = T c (A ∪ c) - T c (c) (16) Among them, F pd (A) represents the performance deviation between the aggregation regulation characteristics of aggregate A and the grid demand, F pd (A∪c) represents the performance deviation between the aggregation regulation characteristics of the new aggregate formed after cluster c is added to aggregate A and the grid demand, ΔP Agg is the aggregation regulation characteristic of aggregate A, ΔP A∪c is the aggregation regulation characteristic of the new aggregate formed after cluster c is added to aggregate A, T c (A) is the aggregation cost of aggregate A, T c (A∪c) is the aggregation cost of the new aggregate formed after cluster c is added to aggregate A; In the upper layer stage, the aggregator selects different clusters to join the aggregator according to different scenarios, and then hands over the coordination process of the lower layer resources to coordinate the resource allocation coefficient; the lower layer resource coordination process coordinates the allocation factors each time A←c opt and then coordinates the allocation factors to minimize the performance deviation through formula (10).

10. A flexible resource dynamic aggregation system for a virtual power plant, characterized in that, including: A data acquisition module, which is used to obtain the set of available flexible resources and the frequency regulation characteristics of each flexible resource in the set based on the frequency regulation demand of the virtual power plant grid; A pre-clustering module, which is used to calculate the similarity between flexible resources and use the Canopy algorithm to pre-cluster the flexible resources to generate several pre-clusters; A refined clustering module, which is used to extract the number of pre-clusters and the initial clustering centers from the pre-clusters, and run the K-mediods clustering algorithm to refine the clustering of the flexible resources to obtain several aggregates; A model construction and evaluation module, which is used to construct a regulation demand model for the virtual power plant to serve the grid and the aggregation regulation characteristic of the aggregate according to the frequency regulation characteristics, and construct a performance deviation function to evaluate the gap between the aggregate regulation characteristic and the regulation demand model and optimize both; A dynamic aggregation constraint module, which is used to select the corresponding performance index as the constraint condition for dynamic aggregation for different regulation scenarios; A dynamic aggregation optimization module, which is used to establish a two-stage dynamic aggregation model under different scenarios by using the method of submodular optimization according to the regulation demand model, the aggregation regulation characteristic of the aggregate and the constraint conditions, and obtain the dynamic aggregation result output by the two-stage dynamic aggregation model to form a flexible resource aggregate that meets the grid regulation demand.

Images