Virtual power plant distributed resource aggregation data processing method and system and storage medium

Abstract

The invention provides a virtual power plant distributed resource aggregation data processing method and system and a storage medium, and the method comprises the steps: processing the real-time and historical data of each distributed resource based on a time index decay weight, constructing a time weighted feature vector containing the rated capacity, the power fluctuation rate and other indexes, dividing the resources into a plurality of resource clusters through a clustering algorithm, and carrying out the clustering of the resources, cluster center features considering historical scheduling compliance and data confidence are calculated, the adjustment potential energy deviation degree of each resource and the cluster center to which the resource belongs on a multi-dimensional index is evaluated, an in-cluster deviation degree dispersion coefficient is calculated according to evaluation, an adjustment threshold value is determined, the resources with the deviation degree exceeding the threshold value are removed, and the resources with the deviation degree exceeding the threshold value are eliminated. And allocating aggregation weights to other resources according to the deviation degree, and aggregating weighted data of the reserved resources to obtain a total aggregation state of the virtual power plant.

Description

Technical Field

[0001] This application belongs to the field of data processing, and in particular relates to a method, system and storage medium for processing distributed resource aggregation data of a virtual power plant. Background Technology

[0002] Virtual power plants are an advanced form of energy management that aggregates geographically dispersed distributed energy resources, such as wind power, photovoltaics, energy storage systems, and controllable loads, through information and communication technologies, forming a unified whole to participate in the electricity market and grid dispatch. Assessing and processing aggregated data of distributed resources is a prerequisite for ensuring the safe, stable, and economical operation of virtual power plants. However, the operating states of distributed resources vary greatly. Existing aggregation methods typically classify or cluster resources first, but these methods often rely on static physical attributes such as type and rated capacity, neglecting real-time characteristics such as power fluctuations, communication link quality, and response latency. This results in clustering results that fail to accurately reflect the actual operating state of resource clusters at a specific time point. Furthermore, when a resource's state parameters, such as power deviation, exceed a preset constant limit, the resource is removed from the aggregation calculation, failing to adapt to variations in the dispersion of states within different resource clusters. For example, a resource cluster composed of wind turbines is inherently more volatile, while a cluster composed of energy storage is relatively stable; using the same exclusion criteria is clearly unreasonable. Furthermore, for the resources that pass the screening, most of them are aggregated using equal weight or simple weighting based only on capacity. They fail to allocate weights according to the real-time deviation of resources from the cluster center features, resulting in more stable and ideal resources having little impact on the final result, which reduces the reliability and usability of the aggregation state. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes a distributed resource aggregation data processing method for virtual power plants, comprising the following steps:

[0004] Real-time and historical data of each distributed resource are acquired, and a time-weighted feature vector is constructed for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality. Based on the time-weighted feature vector, a preset clustering algorithm is used to divide the distributed resources into at least two resource clusters, and the cluster center features of each resource cluster are calculated.

[0005] The multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster is calculated based on the active power deviation, adjustment response time mismatch, and unexpected state switching cost. Based on the multidimensional adjustment potential deviation of all distributed resources in each resource cluster, the intra-cluster deviation dispersion coefficient of the resource cluster is calculated, and the adjustment threshold is obtained using the deviation dispersion coefficient.

[0006] If the deviation of the multidimensional adjustment potential energy of a distributed resource exceeds the adjustment threshold of its resource cluster, the resource is removed in this aggregation; otherwise, the resource is retained and an aggregation weight is assigned to it. The aggregation weight is negatively correlated with the deviation of the multidimensional adjustment potential energy. The weighted data of all retained resources are aggregated to obtain the total aggregation state of the virtual power plant.

[0007] Optionally, the step of constructing a time-weighted feature vector for each distributed resource based on rated capacity, historical power volatility, response latency, and communication link quality includes:

[0008] Rated capacity is used as a static characteristic dimension, with the value being the nominal rated capacity; historical power fluctuation rate, response latency, and communication link quality are used as dynamic characteristic dimensions. The calculation formula is:

[0009]

[0010] in, Let j be the value of the j-th dynamic feature dimension at time t. This represents the latest observation value of the j-th dynamic feature at time t. The feature value of the previous time step. Let be the smoothing factor for the j-th dynamic feature dimension, and The smoothing factor The value is preset based on the time-varying characteristics of the corresponding feature, and the smoothing factor for response delay and communication link quality is greater than the smoothing factor for historical power fluctuation.

[0011] The static feature dimension value and the three dynamic feature dimension values ​​are combined in a preset order of rated capacity, historical power fluctuation rate, response delay and communication link quality to form a four-dimensional vector, which is the time-weighted feature vector.

[0012] Optionally, the step of dividing the distributed resources into at least two resource clusters using a preset clustering algorithm includes:

[0013] The K-Means++ clustering algorithm is used for partitioning, with a preset number of resource clusters K of 5;

[0014] In the K-Means++ clustering algorithm, the distance between resources is calculated as a weighted Euclidean distance, where the weights of the four dimensions—rated capacity, historical power fluctuation rate, response latency, and communication link quality—are 0.2, 0.4, 0.3, and 0.1, respectively.

[0015] Optionally, the calculation of the cluster center characteristics of each resource cluster includes:

[0016] The product of historical scheduling compliance and data confidence is used as the weight of the weighted average.

[0017] The historical scheduling compliance rate is the ratio of the number of times the distributed resource successfully responded to scheduling commands in the past 30 days to the total number of scheduling commands.

[0018] Data confidence is quantified based on the completeness and timeliness of data uploads, and its value is a floating-point number between 0 and 1;

[0019] Cluster center characteristics The calculation formula is: ,in Let i be the feature vector of the i-th resource in cluster k. Let be the weight of the i-th resource within cluster k.

[0020] Optionally, the calculation of the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on active power deviation, adjustment response time mismatch, and unexpected state switching costs includes:

[0021] The formula for calculating the deviation of the multidimensional adjustment potential energy is as follows: ,in It is the absolute value of the difference between the current actual active power value and the scheduling target value. The rated capacity of the resource;

[0022] The absolute value of the difference between the actual response time and the standard response time defined by the cluster center characteristics. Used as a reference time constant;

[0023] This represents the cost of unexpected state transitions. It is 1 when an unscheduled start-up, shutdown, or mode switch occurs, and 0 otherwise.

[0024] , and These are the preset weighting coefficients.

[0025] Optionally, the calculation of the intra-cluster deviation dispersion coefficient of the resource cluster, and the use of the deviation dispersion coefficient to obtain the adjustment threshold, includes:

[0026] Calculate the mean deviation of the multidimensional adjustment potential energy of all distributed resources within the cluster. and standard deviation With coefficient of variation As the dispersion coefficient of the intra-cluster deviation;

[0027] The formula for calculating the adjustment threshold T is: ,in Based on the threshold, To adjust the sensitivity coefficient.

[0028] Optionally, the resource allocation aggregation weight, wherein the aggregation weight is negatively correlated with the multidimensional adjustment potential deviation, includes:

[0029] The aggregation weights are calculated using a Gaussian function, and the specific calculation formula is as follows: ,in The aggregation weight is D, the multidimensional adjustment potential deviation of the resource is s, and the preset parameter for controlling the width of the weight distribution is s, which is 0.5 times the adjustment threshold of the resource cluster.

[0030] On the other hand, the present invention also proposes a distributed resource aggregation data processing system for virtual power plants, comprising the following units:

[0031] The feature extraction unit is used to acquire real-time and historical data of each distributed resource, and construct a time-weighted feature vector for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality; based on the time-weighted feature vector, the distributed resources are divided into at least two resource clusters using a preset clustering algorithm, and the cluster center features of each resource cluster are calculated.

[0032] The calculation unit is used to calculate the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on the active power deviation, adjustment response time mismatch, and unexpected state switching cost; and to calculate the intra-cluster deviation dispersion coefficient of the resource cluster based on the multidimensional adjustment potential deviation of all distributed resources in each resource cluster, and to obtain the adjustment threshold using the deviation dispersion coefficient.

[0033] An aggregation unit is used to remove a resource in the current aggregation if the deviation of the multidimensional adjustment potential energy of the distributed resource exceeds the adjustment threshold of the resource cluster to which it belongs; otherwise, the resource is retained and an aggregation weight is assigned to the resource, wherein the aggregation weight is negatively correlated with the deviation of the multidimensional adjustment potential energy; the weighted data of all retained resources are aggregated to obtain the total aggregation state of the virtual power plant.

[0034] Optionally, the step of constructing a time-weighted feature vector for each distributed resource based on rated capacity, historical power volatility, response latency, and communication link quality includes:

[0035] Rated capacity is used as a static characteristic dimension, with the value being the nominal rated capacity; historical power fluctuation rate, response latency, and communication link quality are used as dynamic characteristic dimensions. The calculation formula is:

[0036]

[0037] in, Let j be the value of the j-th dynamic feature dimension at time t. This represents the latest observation value of the j-th dynamic feature at time t. The feature value of the previous time step. Let be the smoothing factor for the j-th dynamic feature dimension, and The smoothing factor The value is preset based on the time-varying characteristics of the corresponding feature, and the smoothing factor for response delay and communication link quality is greater than the smoothing factor for historical power fluctuation.

[0038] The static feature dimension value and the three dynamic feature dimension values ​​are combined in a preset order of rated capacity, historical power fluctuation rate, response delay and communication link quality to form a four-dimensional vector, which is the time-weighted feature vector.

[0039] Optionally, the step of dividing the distributed resources into at least two resource clusters using a preset clustering algorithm includes:

[0040] The K-Means++ clustering algorithm is used for partitioning, with a preset number of resource clusters K of 5;

[0041] In the K-Means++ clustering algorithm, the distance between resources is calculated as a weighted Euclidean distance, where the weights of the four dimensions—rated capacity, historical power fluctuation rate, response latency, and communication link quality—are 0.2, 0.4, 0.3, and 0.1, respectively.

[0042] Optionally, the calculation of the cluster center characteristics of each resource cluster includes:

[0043] The product of historical scheduling compliance and data confidence is used as the weight of the weighted average.

[0044] The historical scheduling compliance rate is the ratio of the number of times the distributed resource successfully responded to scheduling commands in the past 30 days to the total number of scheduling commands.

[0045] Data confidence is quantified based on the completeness and timeliness of data uploads, and its value is a floating-point number between 0 and 1;

[0046] Cluster center characteristics The calculation formula is: ,in Let i be the feature vector of the i-th resource in cluster k. Let be the weight of the i-th resource within cluster k.

[0047] Optionally, the calculation of the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on active power deviation, adjustment response time mismatch, and unexpected state switching costs includes:

[0048] The formula for calculating the deviation of the multidimensional adjustment potential energy is as follows: ,in It is the absolute value of the difference between the current actual active power value and the scheduling target value. The rated capacity of the resource;

[0049] The absolute value of the difference between the actual response time and the standard response time defined by the cluster center characteristics. Used as a reference time constant;

[0050] This represents the cost of unexpected state transitions. It is 1 when an unscheduled start-up, shutdown, or mode switch occurs, and 0 otherwise.

[0051] , and These are the preset weighting coefficients.

[0052] Optionally, the calculation of the intra-cluster deviation dispersion coefficient of the resource cluster, and the use of the deviation dispersion coefficient to obtain the adjustment threshold, includes:

[0053] Calculate the mean deviation of the multidimensional adjustment potential energy of all distributed resources within the cluster. and standard deviation With coefficient of variation As the dispersion coefficient of the intra-cluster deviation;

[0054] The formula for calculating the adjustment threshold T is: ,in Based on the threshold, To adjust the sensitivity coefficient.

[0055] Optionally, the resource allocation aggregation weight, wherein the aggregation weight is negatively correlated with the multidimensional adjustment potential deviation, includes:

[0056] The aggregation weights are calculated using a Gaussian function, and the specific calculation formula is as follows: ,in The aggregation weight is D, the multidimensional adjustment potential deviation of the resource is s, and the preset parameter for controlling the width of the weight distribution is s, which is 0.5 times the adjustment threshold of the resource cluster.

[0057] Furthermore, the present invention also proposes a computer-readable storage medium on which a computer program is stored, the computer program implementing the method as described in the first aspect when executed by a processor.

[0058] This invention integrates multiple dimensions, including power, time, and state stability, when calculating resource deviation, enabling the determination of resource operating status. Furthermore, it establishes a rejection threshold that matches the degree of state dispersion within each resource cluster, allowing for reasonable screening criteria for resource clusters with different fluctuation characteristics. For resources that pass the screening, different weights are assigned to them in the aggregation calculation based on their deviation degree, ensuring that resources with more stable states and closer to the ideal state dominate the results, thus improving the accuracy and reliability of the overall aggregation state of the virtual power plant. Attached Figure Description

[0059] Figure 1 A flowchart of a specific embodiment;

[0060] Figure 2 A schematic diagram illustrating the formation of feature vectors for distributed resources;

[0061] Figure 3 This is a schematic diagram of resource clustering and cluster center calculation;

[0062] Figure 4 A schematic diagram for determining the adjustment threshold. Detailed Implementation

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0064] In the first embodiment, the present invention proposes a method for processing distributed resource aggregation data in a virtual power plant, such as... Figure 1 As shown, it includes the following steps:

[0065] S1. Obtain real-time and historical data of each distributed resource, and construct a time-weighted feature vector for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality; based on the time-weighted feature vector, use a preset clustering algorithm to divide the distributed resources into at least two resource clusters, and calculate the cluster center features of each resource cluster.

[0066] The SCADA system collects real-time operational data from each distributed resource, retrieving data from a historical database over a specific period, such as the past 24 hours. For each distributed resource, the feature vector consists of four dimensions: rated capacity, the nameplate parameter of the resource, is a static value; in one embodiment, historical power volatility is obtained by calculating the standard deviation of the historical power sequence; response latency is the average time from receiving a command to the power response reaching 90% of the target value across multiple historical scheduling events; and communication link quality is scored based on a combination of packet loss rate and network latency. When calculating historical power volatility and response latency, a time-exponential decay function is used to assign weights to historical data points, with higher weights for data points closer to the current time (e.g., data from 5 minutes ago has a higher weight than data from 12 hours ago). This ensures that the constructed feature vector better reflects the recent operational characteristics of the resource, rather than its long-term average characteristics. Figure 2 As shown.

[0067] In yet another optional embodiment, the construction of a time-weighted feature vector for each distributed resource based on rated capacity, historical power volatility, response latency, and communication link quality includes:

[0068] Rated capacity is used as a static characteristic dimension, with the value being the nominal rated capacity; historical power fluctuation rate, response latency, and communication link quality are used as dynamic characteristic dimensions. The calculation formula is:

[0069]

[0070] in, Let j be the value of the j-th dynamic feature dimension at time t. This represents the latest observation value of the j-th dynamic feature at time t. The feature value of the previous time step. Let be the smoothing factor for the j-th dynamic feature dimension, and The smoothing factor The value is preset based on the time-varying characteristics of the corresponding feature, and the smoothing factor for response delay and communication link quality is greater than the smoothing factor for historical power fluctuation.

[0071] The static feature dimension value and the three dynamic feature dimension values ​​are combined in a preset order of rated capacity, historical power fluctuation rate, response delay and communication link quality to form a four-dimensional vector, which is the time-weighted feature vector.

[0072] For example, for a 100kW rooftop photovoltaic system, the rated capacity is a static 100kW. Historical power fluctuations, response delay, and communication link quality are dynamically changing. When calculating dynamic indicators, a weighted smoothing method is used. For instance, if the latest delay is 30 milliseconds, the smoothing factor λ corresponding to the response delay is, for example, 0.7. Therefore, the updated response delay characteristic value is not the instantaneous 30 milliseconds, but 36 milliseconds, which more smoothly represents the current response capability.

[0073] Clustering algorithms such as K-means++ or DBSCAN are used to divide all distributed resources into multiple resource clusters based on the distance of their time-weighted feature vectors in a four-dimensional feature space. For example, one cluster may consist of energy storage devices with fast response and low volatility, while another cluster may consist of wind turbines with slow response and high volatility. In an optional embodiment, the division of distributed resources into at least two resource clusters using a preset clustering algorithm includes:

[0074] The K-Means++ clustering algorithm is used for partitioning, with a preset number of resource clusters K of 5;

[0075] In the K-Means++ clustering algorithm, the distance between resources is calculated as a weighted Euclidean distance, where the weights of the four dimensions—rated capacity, historical power fluctuation rate, response latency, and communication link quality—are 0.2, 0.4, 0.3, and 0.1, respectively.

[0076] In this embodiment, the partitioning process uses the K-Means++ algorithm, setting the number of target clusters K to 5, and assigning all distributed resources to five categories with different operational characteristics. During clustering, the similarity or distance between resources is calculated using weighted Euclidean distance, with different characteristic dimensions contributing differently to the distance. For example, when comparing the feature vectors of two resources, their difference in the historical power volatility dimension is multiplied by a weight of 0.4, while their difference in the communication link quality dimension is multiplied by a weight of 0.1. This highlights the importance of historical power volatility and response latency as key features. In one embodiment, before clustering, each dimension is standardized, for example, by Z-score, or normalized, for example, by Min-Max.

[0077] When calculating the cluster center characteristics of each cluster, a comprehensive weight is calculated for each resource. In an optional embodiment, the comprehensive weight consists of two parts: historical scheduling compliance, i.e., the degree of matching between the actual response power and the command power of the resource in past scheduling; and data confidence, i.e., the integrity and real-time score of the data uploaded by the resource. There are various ways to calculate data confidence, and this invention does not specifically limit this one. The weight is obtained by normalizing these two indicators and then weighting them together. The cluster center characteristics are obtained by multiplying the feature vector of each resource by the comprehensive weight corresponding to the resource, summing all the results, and then dividing by the sum of the weights. Resources with good scheduling performance and high data quality contribute more to the definition of the cluster center, such as... Figure 3 As shown.

[0078] More specifically, the calculation of the cluster center characteristics of each resource cluster includes:

[0079] The product of historical scheduling compliance and data confidence is used as the weight of the weighted average.

[0080] The historical scheduling compliance rate is the ratio of the number of times the distributed resource successfully responded to scheduling commands in the past 30 days to the total number of scheduling commands.

[0081] Data confidence is quantified based on the completeness and timeliness of data uploads, and its value is a floating-point number between 0 and 1;

[0082] Cluster center characteristics The calculation formula is: ,in Let i be the feature vector of the i-th resource in cluster k. Let be the weight of the i-th resource within cluster k.

[0083] A comprehensive weight is calculated for each distributed resource within the cluster. This comprehensive weight comprises two dimensions: historical scheduling compliance (e.g., if a resource received 100 scheduling commands in the past 30 days and successfully executed 95, its compliance is 0.95) and data confidence (e.g., if a resource has a 98% data upload integrity rate and good timeliness, its confidence can be rated as 0.98). The resource's weight is the product of these two dimensions, which is 0.931. After obtaining the weights of all resources within the cluster, a weighted average method is used to calculate the cluster center feature. Assuming a cluster contains two resources, and resource A's feature vector is... With a weight of 0.931, the feature vector of resource B is... The weight is 0.850. Then the central feature of the cluster... This is calculated using the formula above. Resources with good scheduling records and high data quality will have a greater say in defining the typical behavioral characteristics of a cluster, making the cluster center more representative.

[0084] S2, calculate the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on the active power deviation, adjustment response time mismatch and unexpected state switching cost; calculate the intra-cluster deviation dispersion coefficient of the resource cluster based on the multidimensional adjustment potential deviation of all distributed resources in each resource cluster, and use the deviation dispersion coefficient to obtain the adjustment threshold.

[0085] For any distributed resource within the cluster, the current operating state of the resource is obtained. The deviation calculation includes three aspects: first, active power deviation, which in one embodiment is the absolute value of the difference between the current actual output power and the ideal output power predicted based on the cluster center characteristics; second, adjustment response time mismatch, which is the difference between the actual response time of the resource under the most recent scheduling instruction and the average response time represented by the cluster center characteristics; and third, unexpected state switching cost, which is assigned a large penalty value if the resource experiences unplanned start-up, shutdown, or disconnection, otherwise it is zero. After dimensionless processing of the deviations or costs of the three dimensions, they are fused into a single quantitative multidimensional adjustment potential deviation index by linear weighted summation, representing the degree of deviation between the current state of the resource and the average quality level within the cluster. In an optional embodiment, the calculation of the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on active power deviation, adjustment response time mismatch, and unexpected state switching cost includes:

[0086] The formula for calculating the deviation of the multidimensional adjustment potential energy is as follows: ,in It is the absolute value of the difference between the current actual active power value and the scheduling target value. The rated capacity of the resource;

[0087] The absolute value of the difference between the actual response time and the standard response time defined by the cluster center characteristics. Used as a reference time constant;

[0088] This represents the cost of unexpected state transitions. It is 1 when an unscheduled start-up, shutdown, or mode switch occurs, and 0 otherwise.

[0089] , and These are the preset weighting coefficients.

[0090] Calculate the active power deviation term. Assuming a resource with a rated capacity of 10 MW, a scheduling target of 8 MW, and an actual output of 8.2 MW, the absolute deviation is 0.2 MW, and the value of the active power deviation term is 0.02. Calculate the response time mismatch term. If the standard response time of the cluster center is 3 seconds, the resource actually used 4 seconds, and the reference time constant is 10 seconds, then the absolute time difference is 1 second, and the value of the response time mismatch term is 0.1.

[0091] Assess whether there is any unexpected state transition. If the resource shuts down on its own without scheduling instructions, The value is 1 if the item is empty, and 0 otherwise. Assume the resources are functioning correctly. =0. Use preset weights such as It is 0.5. It is 0.4. The value is 0.1. The weighted sum of the three items above yields a deviation D of 0.05. The value D represents the overall difference between the current state of the resource and its ideal state.

[0092] After calculating the multidimensional adjustment potential deviation of each resource within the cluster, the deviation values ​​are further treated as a dataset, and the coefficient of variation (COP), i.e., the ratio of the standard deviation to the mean, is calculated. The COP is the deviation dispersion coefficient within the cluster. The deviation dispersion coefficient reflects the consistency of the resource states within the cluster; a larger coefficient indicates greater differences in the states of the resources within the cluster and greater overall instability. A preset nonlinear inverse proportional function, such as an inverse proportional function or an S-shaped decay function, is used to determine the adjustment threshold for removing abnormal resources. For example, the threshold is calculated as: threshold = base constant / (1 + dispersion coefficient × adjustment factor). For resource clusters with large internal state differences, a high dispersion coefficient results in a low adjustment threshold and a more stringent screening standard; conversely, for stable clusters with very consistent internal states, a low dispersion coefficient results in a correspondingly higher adjustment threshold and a more lenient standard. Figure 4 As shown. In an optional embodiment, the intra-cluster deviation dispersion coefficient of the computing resource cluster, and the adjustment threshold obtained using the deviation dispersion coefficient, include:

[0093] Calculate the mean deviation of the multidimensional adjustment potential energy of all distributed resources within the cluster. and standard deviation With coefficient of variation As the dispersion coefficient of the intra-cluster deviation;

[0094] The formula for calculating the adjustment threshold T is: ,in Based on the threshold, To adjust the sensitivity coefficient.

[0095] Specifically, the multidimensional adjustment potential deviation values ​​calculated for each resource within the cluster are collected, and the mean μ and standard deviation σ of these deviation values ​​are calculated. For example, if the mean deviation of a cluster is 0.1 and the standard deviation is 0.03, then the coefficient of variation, or coefficient of variation, is 0.3. The coefficient of variation represents the relative dispersion level of the deviation degree of each resource state within the cluster. The adjustment threshold is dynamically calculated based on the coefficient of variation. A base threshold is assumed. The coefficient of variation is 0.5, and the adjustment sensitivity coefficient α is 1.5. Using the formula, the adjustment threshold T is approximately 0.319. If the other cluster has better internal consistency, with a coefficient of variation of 0.1, the adjustment threshold will be higher, approximately 0.430. This results in unstable clusters with large internal state differences having lower adjustment thresholds and being more likely to trigger adjustment commands, while stable clusters have higher tolerance.

[0096] S3. If the deviation of the multidimensional adjustment potential energy of the distributed resource exceeds the adjustment threshold of the resource cluster to which it belongs, the resource is removed in this aggregation; otherwise, the resource is retained and an aggregation weight is assigned to the resource. The aggregation weight is negatively correlated with the deviation of the multidimensional adjustment potential energy. The weighted data of all retained resources are aggregated to obtain the total aggregation state of the virtual power plant.

[0097] The multidimensional adjustment potential deviation calculated for each resource is compared with the adjustment threshold of its resource cluster. If the deviation is greater than the threshold, the resource is considered to be in an abnormal state at the current moment and will not participate in subsequent aggregation calculations. If the deviation is less than or equal to the threshold, the resource is retained, and an aggregation weight is calculated for it. The aggregation weight is preferably determined using a monotonically decreasing function, such as a Gaussian function or an exponentially decaying function, with the resource's deviation as the independent variable. The smaller the deviation, the closer the resource's operating state is to the ideal center of the cluster, and the closer the calculated aggregation weight is to 1. Conversely, the larger the deviation, the smaller the weight, but it will not be zero. This ensures that resources in a better state have a greater say in the aggregation results.

[0098] Multiply the actual active power value of each of the distributed resources retained in the previous step by its corresponding aggregation weight, and sum all the products. The weighted sum is the total aggregated active power status presented by the virtual power plant to the outside world at the current moment.

[0099] In an optional embodiment, the resource allocation aggregation weight, which is negatively correlated with the multidimensional adjustment potential deviation, includes:

[0100] The aggregation weights are calculated using a Gaussian function, and the specific calculation formula is as follows: ,in The aggregation weight is D, the multidimensional adjustment potential deviation of the resource is s, and the preset parameter for controlling the width of the weight distribution is s, which is 0.5 times the adjustment threshold of the resource cluster.

[0101] The contribution of resources to the aggregation response is determined by their real-time operational status. The width parameter *s* in the Gaussian function is linked to the cluster's adjustment threshold *T*. For example, if a cluster's adjustment threshold *T* is 0.4, then the *s* value for all resources within the cluster is 0.2. The *s* value determines the rate at which the weight decreases with increasing deviation. For each resource within the cluster, its respective multidimensional adjustment potential deviation *D* is substituted into the formula to calculate the aggregation weight. Assuming resource A has a deviation *D* of 0.1, its aggregation weight is approximately 0.882. Another poorly performing resource, B, has a deviation *D* of 0.3, and its aggregation weight is approximately 0.325. This shows that resources with smaller deviations receive higher aggregation weights, indicating they are more suitable for participating in aggregation tasks, while resources with large deviations are suppressed to ensure the overall quality and reliability of the aggregated resources.

[0102] In the second embodiment, the present invention also proposes a distributed resource aggregation data processing system for virtual power plants, comprising the following units:

[0103] The feature extraction unit is used to acquire real-time and historical data of each distributed resource, and construct a time-weighted feature vector for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality; based on the time-weighted feature vector, the distributed resources are divided into at least two resource clusters using a preset clustering algorithm, and the cluster center features of each resource cluster are calculated.

[0104] The calculation unit is used to calculate the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster based on the active power deviation, adjustment response time mismatch, and unexpected state switching cost; and to calculate the intra-cluster deviation dispersion coefficient of the resource cluster based on the multidimensional adjustment potential deviation of all distributed resources in each resource cluster, and to obtain the adjustment threshold using the deviation dispersion coefficient.

[0105] An aggregation unit is used to remove a resource in the current aggregation if the deviation of the multidimensional adjustment potential energy of the distributed resource exceeds the adjustment threshold of the resource cluster to which it belongs; otherwise, the resource is retained and an aggregation weight is assigned to the resource, wherein the aggregation weight is negatively correlated with the deviation of the multidimensional adjustment potential energy; the weighted data of all retained resources are aggregated to obtain the total aggregation state of the virtual power plant.

[0106] In a third embodiment, the present invention provides a computer-readable storage medium on which a computer program is stored, the computer program implementing the method described in the first embodiment when executed by a processor.

[0107] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0108] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0109] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for processing distributed resource aggregation data in a virtual power plant, characterized in that, Includes the following steps: Real-time and historical data of each distributed resource are acquired, and a time-weighted feature vector is constructed for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality. Based on the time-weighted feature vector, a preset clustering algorithm is used to divide the distributed resources into at least two resource clusters, and the cluster center features of each resource cluster are calculated. The multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster is calculated based on the active power deviation, adjustment response time mismatch, and unexpected state switching cost. Based on the multidimensional adjustment potential deviation of all distributed resources within each resource cluster, the intra-cluster deviation dispersion coefficient of the resource cluster is calculated, and the adjustment threshold is obtained using the deviation dispersion coefficient. If the deviation of the multidimensional adjustment potential of a distributed resource exceeds the adjustment threshold of its resource cluster, the resource will be removed in this aggregation. Otherwise, resources are retained and aggregated weights are assigned to them, the aggregated weights being negatively correlated with the deviation of the multidimensional adjustment potential energy; the weighted data of all retained resources are aggregated to obtain the overall aggregated state of the virtual power plant.

2. The method according to claim 1, characterized in that, The construction of a time-weighted feature vector for each distributed resource based on rated capacity, historical power fluctuation rate, response latency, and communication link quality includes: Rated capacity is used as a static characteristic dimension, with the value being the nominal rated capacity; historical power fluctuation rate, response latency, and communication link quality are used as dynamic characteristic dimensions. The calculation formula is: in, Let j be the value of the j-th dynamic feature dimension at time t. This represents the latest observation value of the j-th dynamic feature at time t. The feature value of the previous time step. Let be the smoothing factor for the j-th dynamic feature dimension, and The smoothing factor The value is preset based on the time-varying characteristics of the corresponding feature, and the smoothing factor for response delay and communication link quality is greater than the smoothing factor for historical power fluctuation. The static feature dimension value and the three dynamic feature dimension values ​​are combined in a preset order of rated capacity, historical power fluctuation rate, response delay and communication link quality to form a four-dimensional vector, which is the time-weighted feature vector.

3. The method according to claim 1, characterized in that, The method of dividing distributed resources into at least two resource clusters using a pre-defined clustering algorithm includes: The K-Means++ clustering algorithm is used for partitioning, with a preset number of resource clusters K of 5; In the K-Means++ clustering algorithm, the distance between resources is calculated as a weighted Euclidean distance, where the weights of the four dimensions—rated capacity, historical power fluctuation rate, response latency, and communication link quality—are 0.2, 0.4, 0.3, and 0.1, respectively.

4. The method according to claim 1, characterized in that, The calculation of the cluster center characteristics of each resource cluster includes: The product of historical scheduling compliance and data confidence is used as the weight of the weighted average. The historical scheduling compliance rate is the ratio of the number of times the distributed resource successfully responded to scheduling commands in the past 30 days to the total number of scheduling commands. Data confidence is quantified based on the completeness and timeliness of data uploads, and its value is a floating-point number between 0 and 1; Cluster center characteristics The calculation formula is: ,in Let i be the feature vector of the i-th resource in cluster k. Let be the weight of the i-th resource within cluster k.

5. The method according to claim 1, characterized in that, The calculation of the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster, based on active power deviation, adjustment response time mismatch, and unexpected state switching costs, includes: The formula for calculating the deviation of the multidimensional adjustment potential energy is as follows: , in It is the absolute value of the difference between the current actual active power value and the scheduling target value. The rated capacity of the resource; The absolute value of the difference between the actual response time and the standard response time defined by the cluster center characteristics. Used as a reference time constant; This represents the cost of unexpected state transitions. It is 1 when an unscheduled start-up, shutdown, or mode switch occurs, and 0 otherwise. , and These are the preset weighting coefficients.

6. The method according to claim 1, characterized in that, The intra-cluster deviation dispersion coefficient of the computational resource cluster is used to obtain the adjustment threshold, including: Calculate the mean deviation of the multidimensional adjustment potential energy of all distributed resources within the cluster. and standard deviation With coefficient of variation As the dispersion coefficient of the intra-cluster deviation; The formula for calculating the adjustment threshold T is: ,in Based on the threshold, To adjust the sensitivity coefficient.

7. The method according to claim 1, characterized in that, The aforementioned is the resource allocation aggregation weight, which is negatively correlated with the multidimensional adjustment potential deviation, including: The aggregation weights are calculated using a Gaussian function, and the specific calculation formula is as follows: ,in The aggregation weight is D, the multidimensional adjustment potential deviation of the resource is s, and the preset parameter for controlling the width of the weight distribution is s, which is 0.5 times the adjustment threshold of the resource cluster.

8. A distributed resource aggregation data processing system for a virtual power plant, characterized in that, Includes the following units: The feature extraction unit is used to acquire real-time and historical data of each distributed resource, and construct a time-weighted feature vector for each distributed resource based on rated capacity, historical power fluctuation rate, response delay and communication link quality; based on the time-weighted feature vector, the distributed resources are divided into at least two resource clusters using a preset clustering algorithm, and the cluster center features of each resource cluster are calculated. The calculation unit is used to calculate the multidimensional adjustment potential deviation between the current operating state of each distributed resource and the cluster center characteristics of its resource cluster, based on the active power deviation, adjustment response time mismatch, and unexpected state switching cost. Based on the multidimensional adjustment potential deviation of all distributed resources within each resource cluster, the intra-cluster deviation dispersion coefficient of the resource cluster is calculated, and the adjustment threshold is obtained using the deviation dispersion coefficient. The aggregation unit is used to remove resources in this aggregation if the deviation of the multidimensional adjustment potential of a distributed resource exceeds the adjustment threshold of its resource cluster. Otherwise, resources are retained and aggregated weights are assigned to them, the aggregated weights being negatively correlated with the deviation of the multidimensional adjustment potential energy; the weighted data of all retained resources are aggregated to obtain the overall aggregated state of the virtual power plant.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-7.

Images