A virtual power plant resource dynamic aggregation method for a power spot market

Abstract

This invention belongs to the field of virtual power plant technology and relates to a dynamic resource aggregation method for virtual power plants oriented towards the electricity spot market. The invention extracts distributed resource feature vectors and maps them to response type labels; determines the price fluctuation cycle based on the spot electricity price sequence, calculates the temporal matching degree between each resource and the market rhythm based on the feature vectors, and integrates clustering weights and response type labels to construct a comprehensive distance; clusters resources into dynamic aggregation units with the objective of minimizing the variance of the comprehensive distance; generates resource aggregation strategies based on the aggregation units and issues control commands; iteratively updates the clustering weights and aggregation parameters based on the clearing deviation rate fed back by market clearing feedback. This invention solves the technical problems of existing static partitioning ignoring market dynamics, clustering models easily merging heterogeneous resources, and lacking closed-loop optimization, achieving improved market responsiveness of virtual power plants, consistency of aggregation unit control, and adaptive optimization effects for continuous transactions.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant technology and relates to a method for dynamic aggregation of virtual power plant resources for the electricity spot market. Background Technology

[0002] With the deepening of power market reforms, the spot market is playing an increasingly prominent role in resource allocation. Virtual power plants, as integrated management platforms for distributed resources, can participate in market transactions through unified coordination and control, thereby improving resource utilization efficiency.

[0003] Currently, the core issue for virtual power plants participating in the spot market lies in how to effectively aggregate numerous distributed resources with varying response characteristics to form competitive bidding entities. Existing methods mostly employ static grouping or simple threshold classification, which fails to meet the dynamic demands of the market. Specifically, traditional methods often rely on static division based on resource rating parameters or spatial location, neglecting the matching relationship between the dynamic response characteristics of resources and the timing of market electricity price changes. This results in delayed responses from aggregated groups during critical periods, impacting the trading revenue of virtual power plants.

[0004] Existing clustering algorithms mostly use a general distance model, which only focuses on the degree of numerical proximity. This can easily lead to resources with significant differences in key response characteristics such as adjustment ability, action sensitivity and response category being incorrectly merged, reducing the control consistency of the clustered units.

[0005] Existing strategies often employ open-loop control after generation, failing to dynamically adjust the grouping criteria and characteristics based on actual market clearing results. This leads to the accumulation of historical deviations, causing resource organization and strategy formulation in subsequent cycles to deviate from actual operational needs, thus restricting the economy and reliability of virtual power plants in continuous trading. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a method for dynamic aggregation of virtual power plant resources for the electricity spot market is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: a method for dynamic aggregation of virtual power plant resources for the electricity spot market, comprising: acquiring the spot electricity price sequence of the electricity spot market and telemetry operation data of distributed resources; analyzing the physical operation characteristics of each resource based on the telemetry operation data; extracting feature vectors including regulation rate, response delay and regulation dead zone; and mapping each resource to a response type label according to its numerical distribution density.

[0008] The price fluctuation cycle is determined based on the spot electricity price sequence, and the temporal matching degree of each resource relative to the price fluctuation cycle is determined by combining the feature vector. A comprehensive distance is constructed based on the initialized cluster weights, temporal matching degree and response type labels.

[0009] With the goal of minimizing the variance of the comprehensive distance within each dynamic aggregation unit, each resource cluster is divided into dynamic aggregation units, and the central features within each unit are recorded as aggregation parameters.

[0010] Based on the dynamic aggregation unit, a resource aggregation strategy is generated by solving linear programming and control instructions are issued. According to the clearing deviation rate of market clearing feedback, the clustering weights and aggregation parameters used for resource allocation and strategy generation in the next control cycle are iteratively updated.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention extracts and utilizes the adjustment rate, response delay and adjustment dead zone of resources to form a feature vector, and calculates the time sequence matching degree in combination with the price fluctuation cycle. This can characterize the dynamic response capability of each resource and its degree of fit with the market rhythm, thereby overcoming the defect of the traditional static division method ignoring the dynamic changes of the market, ensuring that the virtual power plant unit formed by aggregation can respond accurately during the critical electricity price fluctuation period, and effectively improving the profitability of market transactions.

[0012] (2) This invention constructs a comprehensive distance that integrates clustering weights, temporal matching degree and response type labels, and performs clustering division with the goal of minimizing the variance of the comprehensive distance within each dynamic aggregation unit. This solves the problem that existing clustering algorithms use a general distance model that only focuses on the degree of numerical closeness and is prone to causing resources with significant differences in response characteristics to be incorrectly merged. This ensures the homogeneity of resources within the aggregation unit and improves the consistency of virtual power plant control over aggregation groups.

[0013] (3) By obtaining the clearing deviation rate of market clearing feedback, this invention iteratively updates the clustering weights and aggregation parameters used for resource allocation and strategy generation in the next control cycle. This solves the problem that existing strategies often use open-loop control after generation and do not dynamically adjust the grouping basis and group characteristics according to the actual market clearing results. This avoids the cumulative effect of historical deviations and realizes the adaptive optimization of resource organization and strategy formulation of virtual power plants in continuous trading. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the steps of a method for dynamically aggregating virtual power plant resources for the electricity spot market, as described in this invention.

[0016] Figure 2This is a flowchart of the method for obtaining response type tags in this invention.

[0017] Figure 3 This is a flowchart illustrating the specific method for constructing the comprehensive distance in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details a specific scheme for a method of dynamically aggregating virtual power plant resources for the electricity spot market provided by this invention.

[0021] Please see Figure 1 As shown, the present invention provides a method for dynamic aggregation of virtual power plant resources for the electricity spot market, the method comprising S1 to S4.

[0022] S1. Obtain the spot electricity price sequence of the electricity spot market and the telemetry operation data of distributed resources. Based on the telemetry operation data, analyze the physical operation characteristics of each resource, extract feature vectors containing regulation rate, response delay and regulation dead zone, and map each resource to a response type label according to its numerical distribution density.

[0023] Considering that virtual power plant resource aggregation needs to be based on price signals from the electricity spot market and the actual operating status of distributed resources, the physical operating characteristics of distributed resources, such as adjustment rate, response delay, and adjustment dead zone, directly determine their responsiveness to market price fluctuations. Resources with different responsiveness need to be classified with differentiated labels in order to provide a basis for subsequent clustering and aggregation.

[0024] Therefore, by collecting the spot electricity price sequence of the electricity spot market and simultaneously accessing the telemetry operation data of distributed resources, and analyzing the power change curves, command timestamps and noise intervals in the data, key feature vectors are extracted, and each resource is labeled with a response type based on the statistical distribution law of the feature vectors.

[0025] In one specific embodiment, firstly, the spot electricity price sequence of a certain region's electricity spot market for multiple consecutive control cycles, as well as telemetry operation data such as real-time power and operating status of distributed resources like photovoltaics, energy storage, and controllable loads, are collected. Then, through time synchronization and interpolation processing, the two are aligned to the same control time granularity. The control cycle is typically set to 15 minutes, consistent with the clearing time granularity of the electricity spot market.

[0026] Furthermore, in order to extract physical characteristic parameters that can quantitatively describe the response capabilities of each distributed resource, feature analysis is performed on the telemetry operation data.

[0027] Specifically, based on the power change curve in the telemetry operation data, the slope of the power change per unit time is calculated as the adjustment rate, which reflects how fast the resource is adjusted. The timestamps of the historical scheduling instructions associated with the telemetry operation data are extracted as instruction issuance timestamps. The time difference between the instruction issuance timestamp and the actual start of the resource power change is calculated as the response delay, which reflects the degree of lag of the resource in response to the instructions.

[0028] In telemetry operation data, intervals where power changes are below the noise threshold are identified along the time series. The corresponding scheduling command change amplitudes within these intervals are extracted, and the minimum value among all extracted amplitudes is statistically defined as the adjustment dead zone. This value reflects the resource's resolution or non-responsiveness to minute commands. The noise threshold is set to 0.5% of the rated power.

[0029] Finally, the above-mentioned adjustment rate, response delay, and adjustment dead zone are combined to form a feature vector.

[0030] Further, please refer to Figure 2 As shown, after obtaining the feature vector of each resource, in order to further simplify the complexity of subsequent classification processing, the feature vector needs to be normalized and type mapped to eliminate the influence of different dimensions and discretize the continuous features into response type labels that are easy to cluster and distinguish.

[0031] Specifically, the feature vectors corresponding to all distributed resources within the current control cycle are used to form a dataset, and the feature vectors within the dataset are normalized.

[0032] Then, based on the normalized feature vector numerical distribution density, the high-density interval in the first preset proportion, representing resources with the best adjustment performance, is mapped to a fast response type label; the low-density interval in the second preset proportion, representing resources with the slowest adjustment performance, is mapped to a slow response type label; and resources in the middle interval (60%) are mapped to a regular response type label, thus forming response type labels. The preset proportion can be adjusted by the implementer according to actual circumstances.

[0033] S2. Determine the price fluctuation cycle based on the spot electricity price sequence, determine the temporal matching degree of each resource relative to the price fluctuation cycle by combining the feature vector, and construct the comprehensive distance based on the initialized cluster weights, temporal matching degree and response type labels.

[0034] Given the significant cyclical fluctuations in electricity spot market prices, influenced by factors such as day-ahead and intraday market clearing frequencies and load change patterns, a mismatch between resource response delays and market price cycles can prevent virtual power plants from promptly capitalizing on market signals and even lead to deviations.

[0035] Therefore, when aggregating resources, it is necessary to consider not only the physical characteristics of the resources but also the temporal alignment between their response speed and the rate of market price changes. Furthermore, since response type tags are physical attributes of resources, resources with similar tags should be aggregated during aggregation to maintain homogeneity within the unit.

[0036] In one specific embodiment, the price fluctuation cycle is first determined based on the spot electricity price sequence. Specifically, the sliding window length and step size are set, such as 24 hours and 1 hour. The variance of the electricity price data within the window is calculated by sliding the sliding window over the spot electricity price sequence to obtain the electricity price fluctuation variance sequence, which is used to characterize the degree of change in market uncertainty over time.

[0037] Next, the autocorrelation coefficient sequence of the electricity price fluctuation variance sequence is calculated. The autocorrelation coefficient is used to measure the degree of self-similarity of the electricity price fluctuation variance sequence under different lag times.

[0038] The specific calculation method is as follows: Set the lag time *k* to the offset of the data points in the sequence. For example, *k=1* indicates that the sequence is shifted backward by one data point. For each lag time *k*, calculate the covariance between the electricity price fluctuation variance sequence and the sequence after shifting backward by *k* data points. Divide this covariance by the product of the standard deviations of the two sequences to obtain the autocorrelation coefficient for the corresponding lag time *k*. Iterate through all integers from 1 to the maximum lag time (e.g., taking half the total sequence length) as the lag time to form the autocorrelation coefficient sequence.

[0039] Then, autocorrelation coefficients with a lag order of 0 are excluded from the autocorrelation coefficient sequence, and the lag time corresponding to the first local maximum is selected as the price fluctuation cycle. This cycle reflects the time span required for the market electricity price to complete one typical fluctuation. If no local maximum exists in the sequence, the lag time corresponding to the first drop of the autocorrelation coefficient to a preset threshold, such as below 0.5, can be selected as a substitute.

[0040] Furthermore, after obtaining the price fluctuation cycle, it is necessary to quantitatively compare the response speed of resources with the rhythm of market fluctuations in order to assess the degree of timing fit of each resource in market regulation. Therefore, the timing matching degree of each resource is calculated by combining the response delay in the resource feature vector.

[0041] Specifically, the absolute value of the difference between the response delay of each resource and the price fluctuation cycle is calculated, and the normalized error is obtained by dividing the absolute value of the difference by the price fluctuation cycle. If the error is less than 1, the time series matching degree is obtained by subtracting the error from 1. The closer the value is to 1, the better the resource response speed matches the market change rhythm; if the error is greater than or equal to 1, the matching degree is zero, indicating that the resource response is too lagging and does not match the market cycle at all.

[0042] Further, please refer to Figure 3 As shown, after obtaining the temporal matching degree, it needs to be integrated with the resource type attribute and cluster weight to construct a metric that can comprehensively reflect the similarity between resources. Therefore, a comprehensive distance is constructed based on the initialized cluster weight, temporal matching degree and response type label.

[0043] Specifically, when constructing the comprehensive distance, the clustering weights of each resource in the current control period are initialized to the default value of 1, and the clustering weights are multiplied by the temporal matching degree to obtain the weighted temporal matching degree.

[0044] Simultaneously, based on the response type labels of each resource, a label consistency coefficient is set between each pair of resources. The label consistency coefficient is set as follows: first, the response type labels of two resources are compared. If the labels are exactly the same, the label consistency coefficient is set to 1, indicating that no additional distance penalty is generated when the types are the same; if the labels are different, the label consistency coefficient is a penalty value greater than 1.

[0045] The penalty value can be set according to the degree of difference between the tags: the tags are divided into three levels: fast response, normal response, and slow response. When two resource tags differ by one level, the penalty value is set to 2. When they differ by two levels, the penalty value is set to 3.

[0046] It is understood that the magnitude of the penalty value is positively correlated with the degree of difference in the labels. Those skilled in the art can set other penalty values ​​that conform to this principle according to actual needs, such as penalty values ​​based on continuous function mapping.

[0047] Finally, the weighted temporal matching degree is multiplied by the label consistency coefficient to obtain the comprehensive distance used for subsequent clustering. The smaller this comprehensive distance, the higher the similarity between the two resources in terms of temporal matching degree and response type attribute, and the more suitable they are to be assigned to the same dynamic aggregation unit.

[0048] S3. With the goal of minimizing the variance of the comprehensive distance within each dynamic aggregation unit, each resource cluster is divided into dynamic aggregation units, and the central features within each unit are recorded as aggregation parameters.

[0049] Considering that virtual power plants need to aggregate distributed resources with similar characteristics into units for unified scheduling, the rationality of clustering directly determines the operating efficiency of the aggregated units. The goal is to minimize the variance of the comprehensive distance to ensure the homogeneity of resources within the unit. At the same time, the central features of each unit need to be determined as aggregation parameters to provide unit feature basis for the subsequent generation of aggregation strategies.

[0050] Therefore, by using an iterative clustering algorithm to divide resources into multiple dynamic aggregation units, calculating the central features within each unit, and using these features as equivalent parameters for that unit in subsequent policy generation, a dimensionality-reduced representation of the resource group is achieved.

[0051] In one specific embodiment, a clustering algorithm is used to divide the resources with the goal of minimizing the sum of variances of the comprehensive distances between resources within each dynamic aggregation unit.

[0052] Specifically, the execution process of the clustering algorithm includes: first, K-Means++ is used to initialize the cluster centers, that is, the first center point is randomly selected from the dataset, and then the resource with the greatest comprehensive distance from the selected center point is selected as the next center point each time, until M initial center points are selected; then, the comprehensive distance between each resource and the cluster center is calculated, and the resources are assigned to the cluster with the closest distance to form a preliminary dynamic aggregation unit.

[0053] The clustering algorithm can adopt K-Means and introduce the K-Means++ initialization strategy. The number of cluster centers can be set according to the total adjustable resources of the virtual power plant. For example, when the number of resources is 50 to 200, M can be set to 5 to 10 units.

[0054] Next, the cluster centers of each cluster are iteratively updated, and resources are reallocated. In each iteration, the mean of the feature vectors of all resources within each cluster is used as the new cluster center. The process of calculating the combined distance between resources and the updated cluster centers and reallocating resources is repeated until the convergence condition is met.

[0055] The convergence condition is as follows: iteration stops when the change in cluster centers between two iterations is less than a preset iteration threshold of 0.001, i.e., the sum of the squares of the changes in each dimension of the feature vectors of all cluster centers is less than 0.001. The converged clusters are then determined as dynamic aggregation units.

[0056] After clustering is completed and the resources contained in each dynamic aggregation unit are determined, aggregation parameters that represent the overall adjustment capability of the unit need to be extracted. Specifically, the weighted average of the feature vectors of each resource within each dynamic aggregation unit is calculated, and this weighted average is used as the central feature within each unit. The weights of the weighted average can be set as the proportion of the adjustable capacity of each resource; implementers can set the weight calculation method according to specific circumstances.

[0057] S4. Based on the dynamic aggregation unit, a resource aggregation strategy is generated by solving linear programming and control instructions are issued. According to the clearing deviation rate of market clearing feedback, the clustering weights and aggregation parameters used for resource allocation and strategy generation in the next control cycle are iteratively updated.

[0058] Considering that virtual power plants need to declare electricity volume and price strategies to participate in the electricity spot market, and that the accuracy of the strategies directly affects economic benefits and assessment fees, and that the randomness of the market environment requires the aggregation algorithm to have self-learning capabilities, it is necessary to solve for the optimal declaration strategy through linear programming, and use the market clearing feedback loop to correct the clustering weights and parameters to achieve continuous optimization of the resource aggregation strategy.

[0059] In one specific embodiment, linear programming is first performed based on the divided dynamic aggregation units to generate a resource aggregation strategy and issue control commands.

[0060] Specifically, by traversing all resources within the dynamic aggregation unit, the maximum and minimum adjustable power of each resource are summed to obtain the upper and lower limits of the output of the dynamic aggregation unit.

[0061] Simultaneously, a linear fit is performed on the most recent N data points of the spot electricity price sequence to obtain the fitting slope. The fitting slope is multiplied by the control period duration and then added to the electricity price of the last data point in the spot electricity price sequence to obtain the market clearing electricity price for the next control period, which is used as the strategy declaration price. Here, N is taken as the data from the most recent 4 time periods, and the linear fit uses the least squares method.

[0062] Then, an objective function is constructed with the goal of maximizing the sum of the product of the strategic declared electricity price and the strategic declared electricity volume minus the operating cost. Within the output boundary, the optimal strategic declared electricity volume and strategic declared electricity price are obtained through linear programming, forming a resource aggregation strategy. Specifically, the linear programming involves calling an optimization solver to solve the objective function using the simplex method, with the upper and lower output boundaries as constraints.

[0063] Finally, the power setting value of each resource is calculated based on the power consumption reported in the strategy, and then sent to each resource as a control command.

[0064] Furthermore, after the instructions are issued and executed, in order to correct the algorithm parameters to adapt to the actual market clearing situation and eliminate the impact of historical bias on subsequent strategies, it is necessary to iteratively update the clustering weights and aggregation parameters based on market clearing feedback.

[0065] Specifically, the actual winning bid volume and settlement price from the market clearing feedback are obtained, the volume deviation rate and price deviation rate are calculated, and the weighted sum of the two is used as the clearing deviation rate. The volume deviation rate is the absolute value of the difference between the actual winning bid volume and the strategic bid volume divided by the strategic bid volume; the price deviation rate is the absolute value of the difference between the settlement price and the strategic bid price divided by the strategic bid price; the weighting parameters of the weighted sum include the volume deviation weight and the price deviation weight, both ranging from 0 to 1 and summing to 1. The implementer can configure these values ​​according to specific circumstances.

[0066] Then, this deviation rate is used to iteratively update the clustering weights for the next control cycle. Specifically, the clustering weights for the current control cycle are adjusted accordingly. Perform an update to obtain the updated cluster weights. The update rules are as follows: ,in, This represents the clearing deviation rate of the dynamic aggregation unit containing the resource. The default penalty coefficient is , and .

[0067] This formula ensures that the greater the deviation, the greater the weight decay, thereby weakening the impact of the resource on the overall distance during the next clustering cycle.

[0068] Finally, the updated clustering weights and the central features of the dynamic aggregated units generated in this control cycle are stored as aggregation parameters for resource allocation and strategy generation in the next control cycle.

[0069] In summary, this invention constructs a closed-loop dynamic aggregation and strategy optimization framework by comprehensively considering the resource's feature vector, the temporal matching degree with the price fluctuation cycle, and the clearing deviation rate of market clearing feedback. Its core technical means are: first, extracting the adjustment rate, response delay, and adjustment dead zone to form a feature vector and mapping it to a response type label; then, constructing a comprehensive distance based on the price fluctuation cycle and clustering weights, and performing dynamic clustering with the objective of minimizing the variance of the comprehensive distance within each dynamic aggregation unit; finally, iteratively updating the clustering weights and aggregation parameters based on the clearing deviation rate.

[0070] This scheme improves the flexibility and scientific nature of resource aggregation within the virtual power plant, enabling the aggregation unit to accurately respond to market signals and continuously optimize based on historical performance. This enhances the overall efficiency and profitability of the virtual power plant in the electricity spot market and is widely applicable to electricity market environments with a high proportion of distributed resources, demonstrating good engineering feasibility.

[0071] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0072] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0073] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0074] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0075] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic aggregation of virtual power plant resources for power spot market, characterized in that, include: Acquire the spot electricity price sequence of the electricity spot market and the telemetry operation data of distributed resources. Based on the telemetry operation data, analyze the physical operation characteristics of each resource, extract feature vectors including regulation rate, response delay and regulation dead zone, and map each resource to response type label according to its numerical distribution density. The method for obtaining the response type tag is as follows: The feature vectors corresponding to all distributed resources within the current control cycle are used to form a dataset, and the feature vectors within the dataset are normalized. Based on the normalized feature vector numerical distribution density, resources located in the first preset ratio range are mapped to fast response type labels, resources located in the second preset ratio range are mapped to slow response type labels, and resources located in the middle range are mapped to regular response type labels, thus forming response type labels. The price fluctuation cycle is determined based on the spot electricity price sequence, and the temporal matching degree of each resource relative to the price fluctuation cycle is determined by combining the feature vector. A comprehensive distance is constructed based on the initialized cluster weights, temporal matching degree and response type labels. The method for obtaining the temporal matching degree is as follows: Calculate the absolute value of the difference between the response delay of each resource and the price fluctuation period, and divide the absolute value of the difference by the price fluctuation period to obtain the normalized error. If the normalization error is less than 1, then subtract the normalization error from 1 to obtain the time series matching degree; Conversely, the timing match degree is zero. With the goal of minimizing the variance of the comprehensive distance within each dynamic aggregation unit, each resource cluster is divided into dynamic aggregation units, and the central features within each unit are recorded as aggregation parameters. Based on the dynamic aggregation unit, a resource aggregation strategy is generated by solving linear programming, and control instructions are issued. According to the clearing deviation rate fed back by the market clearing, the clustering weights and aggregation parameters used for resource allocation and strategy generation in the next control cycle are iteratively updated.

2. The method of claim 1, wherein, The extraction of feature vectors including adjustment rate, response delay, and adjustment dead zone is specifically as follows: Based on the power change curve in the telemetry operation data, the slope of the power change per unit time is calculated as the adjustment rate. Extract the timestamps of historical scheduling instructions associated with telemetry operation data as instruction issuance timestamps, and calculate the time difference from the instruction issuance timestamp to the moment when resource power actually begins to change as response delay. In the telemetry operation data, find the interval where the power change is lower than the noise threshold, and count the minimum command change amplitude within the interval as the adjustment dead zone. The regulation rate, response delay, and regulation dead zone are combined to form a feature vector.

3. The method of claim 1, wherein, The method for obtaining the price fluctuation cycle is as follows: Set the sliding window length and step size, and calculate the variance of the electricity price data within the window by sliding it over the spot electricity price series to obtain the electricity price fluctuation variance series. Calculate the autocorrelation coefficient sequence of the electricity price fluctuation variance sequence, select the lag time corresponding to the first local maximum value other than zero lag, and determine it as the price fluctuation cycle.

4. The method for dynamic aggregation of virtual power plant resources for the electricity spot market as described in claim 1, characterized in that, The comprehensive distance is constructed based on the initialized clustering weights, temporal matching degree, and response type labels, specifically as follows: Initialize the clustering weights of each resource within the current control cycle to the default values, and multiply the clustering weights by the temporal matching degree to obtain the weighted temporal matching degree; Based on the response type labels of each resource, a label consistency coefficient is set between resources. If two resources have the same label, the label consistency coefficient is 1. If they are different, the label consistency coefficient is a penalty value greater than 1. The penalty value is positively correlated with the degree of label difference. The weighted temporal matching degree is multiplied by the label consistency coefficient to obtain the comprehensive distance.

5. The method of claim 1, wherein, The process of dividing each resource cluster into dynamic aggregation units specifically involves: With the goal of minimizing the sum of variances of the comprehensive distances between resources within each dynamic aggregation unit, a clustering algorithm is used to divide the resources. The execution process of the clustering algorithm includes: initializing cluster centers, calculating the comprehensive distance between each resource and the cluster center and assigning the resource to the nearest cluster, iteratively updating the cluster centers until convergence, and using the converged clusters as dynamic aggregation units.

6. The method of claim 1, wherein, The method for obtaining the central features within each unit is as follows: Calculate the weighted average of the feature vectors of each resource within the dynamic aggregation unit, and use the weighted average as the central feature within each unit.

7. The method of claim 1, wherein, The process of generating a resource aggregation strategy based on a dynamic aggregation unit through linear programming and issuing control commands specifically involves: By iterating through all resources within the dynamic aggregation unit, the maximum and minimum adjustable power of each resource are summed to obtain the upper and lower limits of the output of the dynamic aggregation unit. The slope of the fit is obtained by performing a linear fit based on the N most recent data points of the spot electricity price series. The fitted slope is multiplied by the control period duration and then added to the last data point price of the spot electricity price series to obtain the market clearing price for the next control period, which is used as the strategy declaration price. Construct an objective function that maximizes the sum of the product of the strategic bid price and the strategic bid electricity volume minus the operating cost. Within the output boundary, calculate the optimal strategic bid electricity volume and strategic bid price through linear programming to form a resource aggregation strategy. The power setting value of each resource is calculated based on the declared power consumption according to the strategy, and then sent to each resource as a control command.

8. The method of claim 7, wherein, The process of iteratively updating the clustering weights and aggregation parameters used for resource allocation and strategy generation in the next control cycle based on the clearing deviation rate feedback from the market clearing is as follows: Obtain the actual winning bid volume and settlement price from the market clearing feedback, calculate the volume deviation rate and price deviation rate, and use the weighted sum of the two as the clearing deviation rate; wherein, the volume deviation rate is the absolute value of the difference between the actual winning bid volume and the strategic bid volume divided by the strategic bid volume, and the price deviation rate is the absolute value of the difference between the settlement price and the strategic bid price divided by the strategic bid price. The cluster weights of each resource within the current control period are penalized and decayed according to the clearing deviation rate corresponding to that resource to obtain the updated cluster weights of each resource; wherein, the magnitude of the penalized decay is positively correlated with the clearing deviation rate. The updated clustering weights and the central features of the dynamically aggregated units generated in this control cycle are stored as aggregation parameters for resource allocation and strategy generation in the next control cycle.

Images