An adjustable load aggregation and demand response control system for a virtual power plant

Abstract

This invention belongs to the field of load response and regulation, and particularly relates to a control system for adjustable load aggregation and demand response in a virtual power plant. The system uses a clustering module to aggregate dispersed adjustable load resources into load clusters containing shared resources; a screening module constructs an overlapping cluster hypergraph with shared resources as hyperedges and filters high-value candidate sets; an iterative module calculates the Shapley value revenue share of each load cluster through cooperative game theory, generating a joint demand response strategy with the objectives of maximizing the total alliance revenue and minimizing the Gini coefficient, and dynamically updating the system state; finally, a strategy generation module integrates all strategies to generate executable control commands. This invention solves the problems of low resource integration efficiency, response delay, and large execution deviation in existing technologies, achieving optimized scheduling and collaborative control of load resources, and significantly improving the economy and stability of virtual power plants participating in grid demand response.

Description

Technical Field

[0001] This invention belongs to the field of load response and control, and particularly relates to an adjustable load aggregation and demand response control system for a virtual power plant. Background Technology

[0002] Virtual power plants (VPS) participate in grid demand response by aggregating distributed adjustable loads, representing a key technology for optimizing energy allocation. However, existing systems still have significant shortcomings in load aggregation and response control. The current common model involves the grid issuing control plans several days in advance, with operators manually notifying users to execute them, lacking an efficient real-time coordination mechanism. This approach disconnects command transmission from execution feedback, relying on manual communication, resulting in high system response delays and poor accuracy, failing to meet the grid's urgent need for rapid and precise load regulation. Therefore, designing a closed-loop control system capable of automatically and accurately issuing commands, providing real-time feedback on execution status, and rapidly handling anomalies has become a critical technical challenge for improving the response efficiency and reliability of VPS. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a virtual power plant-based adjustable load aggregation and demand response control system. This system uses a clustering module to aggregate dispersed adjustable load resources into load clusters containing shared resources; a screening module constructs an overlapping cluster hypergraph with shared resources as hyperedges and filters high-value candidate sets; an iterative module calculates the Shapley value revenue share of each load cluster through cooperative game theory, generating a joint demand response strategy with the objectives of maximizing the total alliance revenue and minimizing the Gini coefficient, and dynamically updating the system state; finally, a strategy generation module integrates all strategies to generate executable control commands. This invention solves the problems of low resource integration efficiency, response delay, and large execution deviation in existing technologies, achieving optimized scheduling and collaborative control of load resources, and significantly improving the economy and stability of virtual power plants participating in grid demand response.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A controllable load aggregation and demand response control system for a virtual power plant, comprising:

[0006] The clustering module is used to obtain the attribute information of the power grid demand response command and adjustable load resources, and to perform multi-objective clustering through a clustering algorithm to obtain multiple load clusters, wherein each load cluster includes at least one shared adjustable load resource.

[0007] The filtering module uses the load cluster as a vertex. If there is a shared adjustable load resource between at least two load clusters, the shared adjustable load resource is used as a hyperedge to connect the corresponding load clusters, generating an overlapping cluster hypergraph. The candidate set is obtained by calculating the upper bound of the initial value of the hyperedge.

[0008] The iteration module is used to perform iterative operations, select the current shared adjustable load resource from the candidate set, determine the joint demand response strategy that maximizes the benefits based on the load aggregation cluster connected by the hyperedge corresponding to the current shared adjustable load resource through cost game, and synchronously delete the adjustable load resource in the joint demand response strategy from the load aggregation cluster, update the overlapping cluster hypergraph and the candidate set until the candidate set is an empty set.

[0009] The strategy generation module summarizes the joint demand response strategies from each iteration and the adjustable load resources that have not been deleted from the load aggregation cluster to generate the final demand response control strategy.

[0010] Specifically, the clustering module includes:

[0011] The data acquisition unit is used to collect the attribute information of adjustable load resources in real time, including adjustable capacity, response time, cost coefficient, user compensation willingness, and scheduling requirements and time constraints in the power grid demand response command; the data acquisition unit is also connected to the virtual power plant centralized control platform through a communication interface to obtain power grid demand response commands and real-time electricity price information;

[0012] The feature vector unit constructs a multi-dimensional feature vector for each adjustable load resource based on the attribute information collected by the data acquisition unit. The multi-dimensional feature vector includes a static attribute dimension, a dynamic attribute dimension, and a desired attribute dimension. The static attribute dimension includes rated power and geographical location, the dynamic attribute dimension includes real-time adjustable capacity and response rate, and the desired attribute dimension includes the user-defined cost coefficient and sustainable response time.

[0013] Specifically, the clustering module also includes:

[0014] The clustering algorithm unit performs adjustable load resource clustering based on the multi-dimensional feature vector and a multi-objective clustering algorithm to generate multiple load aggregation clusters. The multi-objective clustering algorithm aims to maximize the similarity of characteristics within clusters and balance the adjustable capacity between clusters, and ensures that each load aggregation cluster contains at least one shared adjustable load resource, which is an adjustable load resource that can be jointly scheduled by multiple load aggregation clusters.

[0015] Specifically, the filtering module includes:

[0016] The hypergraph generation unit is used to generate an overlapping cluster hypergraph by using the load clusters as vertices and, if at least two load clusters have the same shared adjustable load resource, using the shared adjustable load resource as a hyperedge to connect the corresponding load clusters.

[0017] The value calculation unit is used to calculate the initial upper bound of the value of each hyperedge in the hypergraph generation unit; the initial upper bound of the value is calculated based on the adjustable capacity of the shared adjustable load resources, the current grid compensation unit price, the average dispatch cost, and the flexibility weight, wherein the flexibility weight is determined according to the response rate and adjustment direction of the shared adjustable load resources; the adjustment direction is defined as the type of adjustment that can be performed within the current dispatch cycle, including at least upward adjustment only, downward adjustment only, or bidirectional adjustment.

[0018] The candidate set generation unit is used to sort all hyperedges in descending order based on the initial value upper bound calculated by the value calculation unit, and select the top N hyperedges with the highest initial value upper bound to form a candidate set.

[0019] Specifically, the iteration module includes:

[0020] The shared resource selection unit is used to select the shared adjustable load resource with the highest upper bound of the current initial value from the candidate set as the current shared device;

[0021] The cost game unit, based on the load aggregation clusters connected by the superedges corresponding to the currently shared adjustable load resources, determines the joint demand response strategy that maximizes revenue through a cost game model. The cost game unit adopts a cooperative game model to calculate the total alliance revenue under different feasible joint scheduling strategies, and allocates the revenue through the Shapley value to select the joint demand response strategy that maximizes the total alliance revenue. The total alliance revenue is the difference between the grid payment compensation and the total cost of all scheduled adjustable load resources.

[0022] Specifically, the iteration module also includes:

[0023] The resource status update unit is used to synchronously delete the adjustable load resources included in the joint demand response strategy determined by the cost game unit from the corresponding load aggregation cluster, mark the adjustable load resources as scheduled, and synchronously update the remaining adjustable capacity and composition of the load aggregation cluster.

[0024] The update iteration unit, based on the output of the resource status update unit, synchronously updates the structure of the overlapping cluster hypergraph and the candidate set, specifically including: deleting the hyperedge corresponding to the currently scheduled shared adjustable load resource from the overlapping cluster hypergraph;

[0025] All adjustable load resources included in the joint demand response strategy determined in the corresponding iteration round are synchronously deleted from the respective load clusters to which they belong. It is then determined whether any load cluster still contains any shared adjustable load resources. If not, the load cluster is marked as a non-shared cluster. At the same time, the initial value upper bound of all hyperedges whose connection relationship has changed due to the above deletion operation is recalculated, and the candidate set is updated based on the recalculation result.

[0026] Once the candidate set is updated, the updated candidate set is checked. If it is not an empty set, the next round of iteration is triggered; if it is an empty set, the iteration optimization process is terminated.

[0027] Specifically, the strategy generation module includes:

[0028] The strategy aggregation unit is used to aggregate the joint demand response strategies generated by the iterative module in each iteration. The strategy aggregation unit records detailed parameters for each joint demand response strategy, including scheduling time, adjustable load resource list, and expected revenue. The expected revenue is quantified by calculating the total alliance revenue, specifically the difference between the total compensation cost paid by the grid and the total scheduling cost of all participating adjustable load resources after executing the joint demand response strategy. The total compensation cost is determined based on the product of the compensation unit price published by the grid and the total regulating power achieved by the joint demand response strategy. The total scheduling cost is the sum of the costs calculated for each adjustable load resource in the adjustable load resource list based on its unit regulating cost coefficient and the actual scheduling power. The adjustable load resource list includes unique identifiers for all adjustable load resources participating in the joint demand response strategy and their scheduled power values.

[0029] The residual load processing unit generates a final dispatch instruction with a fixed power command for each isolated adjustable load resource based on its real-time adjustable capacity, unit adjustment cost coefficient, and current grid adjustment demand. It then integrates and outputs all generated final dispatch instructions to the final strategy generation unit.

[0030] The final strategy generation unit is used to receive and integrate the joint demand response strategy set output by the strategy aggregation unit and the final scheduling instruction set output by the residual load processing unit, and to send the final scheduling instruction set to the corresponding adjustable load terminal for execution through the virtual power plant centralized control platform; the scheduling parameters of the final scheduling instruction include at least the target load value, the execution time window and the equipment identifier.

[0031] Specifically, the joint demand response strategy that maximizes benefits is determined through a cost-game model, including:

[0032] Based on the load aggregation clusters connected to the superedge corresponding to the current shared adjustable load resources, all load aggregation clusters associated with the superedge and all adjustable load resources contained therein constitute an adjustable load resource alliance.

[0033] Based on the total regulating power demand in the power grid demand response command, and combined with the real-time adjustable capacity, regulation direction and power change rate constraints of each adjustable load resource in the alliance, a set of all feasible joint scheduling strategies that meet the global regulation demand is generated through a linear programming algorithm. Each feasible joint scheduling strategy must fully define the specific scheduling power value and regulation direction of each resource in the alliance.

[0034] The objective function is the total revenue of the adjustable load resource alliance. The objective constraint space is constructed using the boundary values ​​of the dispatch power and real-time adjustable capacity of each adjustable load resource, the sustainable dispatch duration of each adjustable load resource and the minimum duration required by the grid demand response command, and the power change rate and rate penalty factor of each adjustable load resource, as well as the upper and lower limits of the power change rate. The rate penalty factor is constructed by the maximum allowable adjustment time in the grid demand response command and the rated power change rate of the corresponding adjustable load resource, and is used to adjust the contribution of time cost to the objective function.

[0035] Specifically, determining the joint demand response strategy that maximizes returns through a cost-game model also includes:

[0036] Based on the set of feasible joint scheduling strategies, the Shapley value algorithm in cooperative game theory is used to calculate the expected revenue share of each load aggregation cluster under each feasible joint scheduling strategy, specifically:

[0037] Generate all sub-alliance combinations of load aggregation clusters in the adjustable load resource alliance. For each load aggregation cluster, calculate its marginal contribution value in each sub-alliance. The marginal contribution value is defined as the change in the total revenue of the corresponding sub-alliance before and after the load aggregation cluster is added to each sub-alliance.

[0038] Based on the marginal contribution values ​​of all load aggregation clusters, the expected revenue share of each load aggregation cluster in the adjustable load resource consortium is calculated according to the Shapley value calculation formula, which is the weighted average of each marginal contribution value in the ranking of all sub-consortia.

[0039] For each feasible joint scheduling strategy, calculate its corresponding total alliance revenue. At the same time, based on the expected revenue share distribution of each load aggregation cluster, use the Lorenz curve and Gini coefficient formula to calculate the revenue distribution fairness index under each feasible joint scheduling strategy.

[0040] With the goal of maximizing total alliance revenue and minimizing the Gini coefficient as parallel optimization objectives, and mapping the feature vector of each feasible joint scheduling strategy to a two-dimensional evaluation space consisting of total alliance revenue and fairness index, a multi-objective strategy evaluation model is established.

[0041] Specifically, determining the joint demand response strategy that maximizes returns through a cost-game model also includes:

[0042] Based on the multi-objective strategy evaluation model and the non-dominated sorting algorithm in multi-objective optimization, the feasible joint scheduling strategy set is screened to obtain a Pareto optimal strategy subset; each feasible joint scheduling strategy in the Pareto optimal strategy subset satisfies the following: it is a non-dominated solution for both the total alliance revenue and the Gini coefficient, that is, there is no other feasible joint scheduling strategy that can improve the other objective without reducing the performance of either objective;

[0043] In the Pareto optimal strategy subset, the strategy with the highest total alliance revenue is selected as the candidate strategy. If multiple Pareto optimal strategies have the same highest total alliance revenue, the variance of the expected revenue share of each load cluster under the multiple Pareto optimal strategies with the same highest total alliance revenue is calculated, and the Pareto optimal strategy with the smallest variance is selected as the final joint demand response strategy. The final joint demand response strategy includes: unique identifiers of all adjustable load resources participating in the final joint demand response strategy and their scheduling power values, a unified scheduling time window for strategy execution, the expected total alliance revenue value, and a detailed table of expected revenue share calculated by each load cluster based on the Shapley value.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] This invention addresses the shortcomings of existing technologies by constructing an overlapping cluster hypergraph model to achieve accurate modeling and efficient aggregation of distributed resources. It utilizes a multi-objective clustering algorithm to form load aggregation clusters with similar internal characteristics and balanced capacity, and establishes inter-cluster connections through a resource-sharing mechanism. The system employs an iterative optimization method based on cooperative game theory, calculating Shapley values ​​to achieve fair distribution of benefits, and combining a Pareto optimal selection mechanism to simultaneously optimize economic efficiency and allocation fairness. Ultimately, it generates a demand response strategy that comprehensively considers regulation costs, response rates, and sustainability. This application effectively solves the problems of low resource integration efficiency, large response delays, and significant execution deviations in traditional virtual power plant systems, significantly improving the utilization efficiency of load resources and the accuracy of collaborative control, providing the power grid with fast, reliable, and economical optimal regulation capabilities. Attached Figure Description

[0046] Figure 1This is a block diagram of an adjustable load aggregation and demand response control system for a virtual power plant according to Embodiment 1 of the present invention;

[0047] Figure 2 This is a flowchart of the corresponding unit of the adjustable load aggregation and demand response control system of a virtual power plant according to Embodiment 1 of the present invention. Detailed Implementation

[0048] Example 1

[0049] The virtual power plant architecture integrates diverse resources such as photovoltaics, energy storage, controllable loads, and charging piles. Through coordination and optimization modules and the centralized control platform, it achieves energy interaction and information feedback with the electricity market and the power grid. Among them, charging piles and other equipment also have dual identities as source and load. These resources constitute the basic carrier for adjustable load aggregation and scheduling.

[0050] Under this architecture, load resources exhibit highly heterogeneous, multi-constrained, and dynamic characteristics: within the same dispatch domain, there are energy storage loads with millisecond-level responses, flexible production loads with minute-level adjustments, and also residential-side transferable loads that need to consider user experience, such as the adjustment needs of charging piles; resource constraints cover not only the adjustable capacity limit and response rate threshold of the equipment itself, but also the user-set compensation willingness limit and sustainable response duration, while also meeting hard requirements such as the time window of the grid demand response command and the total adjustment power. In practice, virtual power plants need to aggregate these massive heterogeneous adjustable load resources for coordinated response, while grid dispatch demand often changes dynamically with real-time electricity prices and load peak-valley differences, such as emergency load reduction during peak electricity consumption and load increase during periods of high renewable energy generation, and the availability status and adjustment costs of load resources also fluctuate in real time with operating conditions. The aforementioned complexity is amplified dramatically by a common phenomenon: the same adjustable load resource (such as some energy storage and charging piles) will be included in the candidate range by multiple load aggregation clusters under multi-objective clustering (such as clustering by response rate or clustering by adjustment cost), resulting in resource competition and duplicate scheduling risks across clusters and scheduling cycles.

[0051] Existing methods generally adopt two approaches: First, they take static attribute clustering as a guide, formulate response strategies within each load cluster, and then eliminate resource conflicts through cross-cluster coordination; second, they use global optimization algorithms or reinforcement learning models to directly solve for the optimal scheduling scheme for all load resources. The former often results in excessively high coordination costs when the same load resource is absorbed by multiple clusters at the same time, and generally leads to resource allocation struggles, redundant scheduling waste, and response time delays. Although the latter has theoretical optimal potential, its computational complexity increases exponentially when multiple non-convex constraints such as adjustable capacity boundaries, response rate constraints, and user compensation willingness are superimposed. Its interpretability and dynamic replanning stability are difficult to meet the minute-level timeliness requirements of power grid demand response, and it is even more difficult to adapt to the real-time fluctuations of load resource status.

[0052] The common dilemma is that once the adjustable load resources shared across clusters are postponed to the end of the scheduling scheme, the early clustering and intra-cluster strategy have already been solidified. No matter how fine-tuned it is later, it is difficult to avoid resource waste, loss of benefits and response timeouts.

[0053] The core processing logic of this application is based on the aforementioned common dilemma. It does not rely on a single-dimensional load classification or a pre-defined cluster division. Instead, it first recognizes that adjustable load resources naturally exhibit cross-cluster sharing under multi-objective clustering, and pre-emptively establishes this sharing relationship as a triggering factor for scheduling solutions. To this end, after obtaining global load resource attribute information and grid demand response instructions, an overlapping cluster hypergraph is constructed, with load aggregation clusters as vertices and shared adjustable load resource sets (such as energy storage with flexible adjustment capabilities and some charging piles) as hyperedges. This transforms shared load resources from sources of scheduling conflict into collaborative anchor points. A cluster-level alliance game is triggered around these anchor points:

[0054] Each participating load aggregation cluster does not bid for a single shared resource. Instead, it submits a candidate response strategy package that includes the shared anchor point, covers a subset of high-value loads in its cluster, and satisfies the constraint rules. Game theory selection and strategy concatenation are then performed using a combination of alliance technical benefits and execution costs to form a joint demand response strategy that spans multiple clusters and must pass through the anchor point. Unlike the conventional scheduling model of first allocating and then adjusting, this embodiment places alliance game theory, benefit pricing, strategy concatenation, resource clearing (point removal), and dynamic re-clustering within a rolling outer loop. Each time a joint strategy is generated, already scheduled load resources are simultaneously removed, dynamically shrinking the problem scale and avoiding large-scale conflict resolution later.

[0055] It should be noted that, due to the essential requirements of project fairness and scheduling stability, this application adopts a profit distribution and settlement logic based on cooperative game theory:

[0056] Different load clusters naturally differ in resource types, adjustment costs, and user preferences. A purely greedy merging approach can easily lead to low-cost, fast-response load clusters repeatedly occupying high-value scheduling opportunities, causing uneven resource allocation and execution bottlenecks. This embodiment introduces a revenue-sharing mechanism based on Shapley values ​​after the joint response strategy is determined. This mechanism possesses incentive compatibility, ensuring that the optimal response selection of each participating cluster no longer deviates from the global optimal goal, thereby mitigating scheme fluctuations caused by strategic reporting. Simultaneously, the risk penalty for remaining uncovered adjustment needs and the combined potential energy of resource scheduling costs are used as the convergence characterization of the outer loop, ensuring that each round of strategy clearing brings provable overall revenue improvement and guaranteeing stable, engineering-executable scheduling results within minute-level rolling time slots.

[0057] It should be emphasized that the above description is not exhaustive of technical details, but rather clarifies the technical starting point and reproducible implementation boundaries of this embodiment:

[0058] Given known global load resource attributes, multi-source dynamic demand flows, and uniformly coded constraint rules, this paper proposes a collaborative demand response strategy generation and stable convergence across clusters, resource types, and scheduling windows. This is achieved by constructing an overlapping cluster hypergraph and triggering an anchor-based alliance game, coupled with a rolling mechanism of synchronous point deletion and re-clustering. This approach transforms the scheduling conflicts caused by sharing into shared-driven collaborative scheduling, fundamentally resolving the triple challenges of redundant scheduling, replanning overhead, and unfair benefits in existing technologies. It provides a unified solution framework for the large-scale aggregated scheduling and dynamic demand response of virtual power plants. Please refer to [link / reference]. Figure 1 One embodiment of the present invention provides: an adjustable load aggregation and demand response control system for a virtual power plant, comprising:

[0059] The clustering module is used to obtain the attribute information of the power grid demand response command and adjustable load resources, and to perform multi-objective clustering through a clustering algorithm to obtain multiple load clusters, wherein each load cluster includes at least one shared adjustable load resource.

[0060] The filtering module uses the load cluster as a vertex. If there is a shared adjustable load resource between at least two load clusters, the shared adjustable load resource is used as a hyperedge to connect the corresponding load clusters, generating an overlapping cluster hypergraph. The candidate set is obtained by calculating the upper bound of the initial value of the hyperedge.

[0061] The iteration module is used to perform iterative operations, select the current shared adjustable load resource from the candidate set, determine the joint demand response strategy that maximizes the benefits based on the load aggregation cluster connected by the hyperedge corresponding to the current shared adjustable load resource through cost game, and synchronously delete the adjustable load resource in the joint demand response strategy from the load aggregation cluster, update the overlapping cluster hypergraph and the candidate set until the candidate set is an empty set.

[0062] The strategy generation module summarizes the joint demand response strategies from each iteration and the adjustable load resources that have not been deleted from the load aggregation cluster to generate the final demand response control strategy.

[0063] It should be further noted that the clustering module in this embodiment includes:

[0064] A data acquisition unit is used to collect attribute information of adjustable load resources in real time, including adjustable capacity, response time, cost coefficient, user compensation willingness, and scheduling requirements and time constraints in the power grid demand response command. The data acquisition unit is also connected to a virtual power plant centralized control platform via a communication interface to obtain power grid demand response commands and real-time electricity price information. In this embodiment, the real-time electricity price information is used to calculate the power grid compensation unit price. This embodiment obtains the adjustable capacity of adjustable load resources in real time through an equipment monitoring system to characterize the adjustable power range of the load under the current operating state. The response time is obtained through equipment performance test records to characterize the time delay required for the load to complete adjustment from receiving the command. The cost coefficient and user compensation willingness are obtained through user contract agreements to characterize the resource cost per unit of power dispatched and the minimum compensation cost for the user to accept dispatch, respectively. The scheduling requirements and time constraints in the power grid demand response command are obtained through the virtual power plant centralized control platform to characterize the specific power demand, adjustment direction, and execution time window of the power grid for load adjustment. The adjustable load resources include at least electric vehicle charging piles, distributed energy storage systems, flexible industrial and commercial production loads, and building temperature control loads, wherein:

[0065] The adjustable capacity of an electric vehicle charging pile represents the adjustable range of its charging power, while the response time represents the delay characteristics of its power adjustment. The adjustable capacity of a distributed energy storage system represents the adjustable range of its charging and discharging power, while the response time represents the delay characteristics of its charging and discharging state switching. The adjustable capacity of a flexible industrial and commercial production load represents the power range that can be interrupted or transferred in its production process, while the response time represents the delay characteristics of its load adjustment. The adjustable capacity of a building temperature control load represents the adjustable range of its heating / cooling equipment power, while the response time represents the delay characteristics of load changes after the temperature setpoint is adjusted.

[0066] The feature vector unit constructs a multi-dimensional feature vector for each adjustable load resource based on the attribute information collected by the data acquisition unit. The multi-dimensional feature vector includes static attribute dimensions, dynamic attribute dimensions, and desired attribute dimensions. The static attribute dimensions include rated power and geographical location; the dynamic attribute dimensions include real-time adjustable capacity and response rate; and the desired attribute dimensions include user-defined cost coefficients and sustainable response duration. In this embodiment, the response rate is specifically defined as the absolute value of the power change that an adjustable load resource can achieve per unit time. It quantitatively characterizes the resource's dynamic adjustment capability to switch from its current operating state to a target power state. For example, energy storage devices typically have a megawatt-level power conversion capability per minute. Sustainable response duration is explicitly defined as the maximum time span during which a load resource can sustainably maintain its target adjustable power while meeting operational constraints. It characterizes the resource's energy support or state maintenance capability at a specific power level, typically manifested as the continuous power adjustment duration that an industrial process load can withstand due to production process limitations. Real-time adjustable capacity is defined as the immediate available range within which an adjustable load resource can safely and reliably accept scheduling commands for power adjustment under the current operating conditions. It characterizes the resource's instantaneous adjustment potential and available margin at a specific moment. For example, a 100kW energy storage system currently charging at 30kW has a real-time adjustable capacity of 70kW (downward adjustment range for charging power) or 30kW (upward adjustment range for discharging power). The specific value is limited by the battery's state of charge, temperature constraints, and the inverter's instantaneous capability, dynamically reflecting the resource's immediate availability in response to grid regulation demands. It should be further noted that the cost coefficient set by the user in this embodiment is constructed based on the compensation standard explicitly stipulated in the medium-to-long-term cooperation agreement or day-ahead bidding agreement signed between the user and the virtual power plant operator. This standard specifies the basic compensation unit price and possible floating clauses. Secondly, combining the historical bidding data of the adjustable load resource in the day-ahead and real-time markets, weighted average or machine learning prediction methods are used for data cleaning and feature extraction to form the baseline cost curve of the resource. Finally, through the two-way communication interface between the virtual power plant operation platform and the user's mobile APP or dedicated terminal, a confirmation request is initiated before each dispatch instruction is generated. The user can dynamically update and confirm the current cost coefficient by confirming the default settings, manually adjusting, or authorizing the system to automatically optimize. The cost coefficient specifically represents the minimum economic compensation required by the user to accept unit power dispatch. Its numerical representation includes: a fixed value for simple loads, a piecewise linear function based on interruptible duration for industrial process loads, and a dynamic quotation curve related to dispatch depth and duration for flexible resources such as energy storage.

[0067] The clustering algorithm unit performs adjustable load resource clustering based on the multi-dimensional feature vector and a multi-objective clustering algorithm to generate multiple load aggregation clusters. The multi-objective clustering algorithm takes maximizing intra-cluster characteristic similarity and equalizing inter-cluster adjustable capacity as optimization objectives, and ensures that each load aggregation cluster contains at least one shared adjustable load resource, wherein the shared adjustable load resource is an adjustable load resource that can be jointly scheduled by multiple load aggregation clusters.

[0068] In virtual power plants, the dispersed nature of adjustable load resources leads to scheduling conflicts. This is addressed by transforming shared loads, which cause resource allocation conflicts in traditional technologies, into anchor points for collaborative optimization. Specifically, multi-dimensional feature clustering is used to form load aggregation clusters containing shared resources (e.g., clustering resources {R1, R2, R3} into cluster C1, and {R4, R5, R6} into cluster C2, where R1 and R4 are shared resources). Then, cross-cluster connections are established using shared resources as hubs. This process, by pre-processing shared relationships, avoids the response delays and resource waste caused by later coordination difficulties in traditional methods, achieving a paradigm shift from conflict resolution to collaborative driving. It should be further noted that the adjustable load resource clustering combined with multi-objective clustering algorithms in this embodiment includes:

[0069] A1. Receive the multi-dimensional feature vector output from the feature vector unit, and use the Min-Max normalization method to standardize the multi-dimensional feature vector to eliminate the dimensional differences between the features of each dimension and obtain the standardized feature vector.

[0070] A2. Based on preset flexible response judgment conditions, select resources from all adjustable load resources that meet the response rate threshold, have bidirectional adjustment capability, and have a sustainable response duration target, and mark them as the shared adjustable load resource set S. The flexible response judgment conditions include at least: setting a response rate threshold, requiring the adjustment direction to be bidirectionally adjustable, and the sustainable response duration not less than the minimum duration specified by the power grid demand response command. The system compares the actual parameters of each adjustable load resource with the above thresholds through a comparator. Resources that meet all three conditions are marked as shared adjustable load resources.

[0071] A3. Based on the total regulating power demand of the power grid demand response command and the optimal adjustable capacity threshold of a single cluster, and combined with the cluster center determination rules of the K-means++ algorithm, the initial number of cluster centers K is obtained.

[0072] A4. Based on the initial number of cluster centers K, the shared adjustable load resources in the shared adjustable load resource set are evenly distributed to each initial cluster unit. Then, based on the principle of feature similarity, non-shared adjustable load resources are added to obtain K initial load clusters containing the initial members, ensuring that each initial load cluster contains at least one shared adjustable load resource in the shared adjustable load resource set.

[0073] A5. Based on the preset weight allocation rules for each attribute dimension, the distance between each adjustable load resource feature vector and each cluster center vector is calculated using the weighted Euclidean distance formula to obtain a quantitative similarity measurement result. The weight allocation rules for each attribute dimension are preset in the following way: based on the feature importance analysis results and combined with the virtual power plant dispatch demand characteristics, the relative weight of each attribute dimension is determined by the analytic hierarchy process. Among them, the dynamic attribute dimension is given the highest weight because it directly determines the load regulation capacity, the willingness attribute dimension is given a medium weight because it affects the economy, and the static attribute dimension is given the lowest weight because it is relatively stable, thus forming a standardized weight allocation scheme.

[0074] A6. Based on the similarity measurement results, with maximizing the similarity of intra-cluster characteristics as the first optimization objective, obtain the optimal clustering for each resource;

[0075] A7. Based on the sum of adjustable capacity of each cluster, firstly, the mean of the sum of adjustable capacity of all clusters is obtained as the average adjustable capacity. Then, based on the preset inter-cluster balance judgment rule, a reasonable range of adjustable capacity deviation is obtained. For clusters whose adjustable capacity exceeds the reasonable range, the adjustable load resource with the lowest similarity in the cluster is selected based on the similarity measurement result between the resource and the cluster center. At the same time, the resource receiving priority is determined based on the capacity gap of each cluster with insufficient adjustable capacity. The selected resource with the lowest similarity is allocated to the cluster with the highest priority and insufficient adjustable capacity. After the allocation is completed, the sum of adjustable capacity and the average adjustable capacity of all clusters are recalculated. The above resource screening, priority judgment and allocation operations are repeated until the deviation between the adjustable capacity and the average adjustable capacity of all clusters is within the reasonable range, thereby achieving the second optimization goal of inter-cluster adjustable capacity balance.

[0076] A8. Based on the standardized feature vectors of all members in each load cluster, the mean vector of each dimension of features is obtained by means calculation method, and the mean vector is used as the updated cluster center.

[0077] A9. Based on the updated cluster centers, check whether each load cluster still contains at least one shared adjustable load resource from a shared adjustable load resource set. If a load cluster without any shared adjustable load resources is found, a resource allocation operation is performed, including: firstly, selecting a shared adjustable load resource from the load cluster with the best current adjustable capacity balance and allocating it to the load cluster lacking shared resources; simultaneously, to maintain the adjustable capacity balance between clusters, selecting a non-shared adjustable load resource from the load cluster receiving shared resources and allocating it to the original load cluster contributing shared resources; through this bidirectional resource exchange mechanism, an updated set of load clusters that simultaneously satisfies the adjustable capacity balance between clusters and where each load cluster contains at least one shared adjustable load resource is finally obtained; in this embodiment, the adjustable capacity balance is obtained by calculating the coefficient of variation or coefficient of dispersion of the total adjustable capacity of each load cluster, specifically characterizing the uniformity and balance of the distribution of schedulable resource capacity among different load clusters. This indicator is used to measure whether there is a structural imbalance in the clustering results, where some clusters bear excessive regulation pressure while other clusters have insufficient regulation capacity. It is an important optimization goal for achieving fair allocation and coordinated scheduling of load resources.

[0078] A10. Based on the degree of feature difference between the updated cluster centers and the previous cluster centers, the Euclidean distance summation method is used to obtain the center offset.

[0079] A11. Based on the quantitative evaluation results of the inter-cluster adjustable capacity balance (the dispersion of the adjustable capacity of all clusters meets the preset requirements), and the constraint that each cluster contains at least one shared adjustable load resource, determine whether the convergence requirements are met.

[0080] A12. If the center offset is less than the preset convergence criterion, the inter-cluster adjustable capacity balance meets the standard, and the shared resource constraint is satisfied, then the clustering is determined to be converged; if any condition is not met, then return to the multi-objective optimization allocation step and repeat the execution until all convergence conditions are met.

[0081] A13. Based on the converged clustering results, obtain multiple final load aggregation clusters. Each load aggregation cluster contains key parameters such as a list of adjustable load resources for its members, the total adjustable capacity within the cluster, a unique identifier for shared adjustable load resources, the mean value of feature similarity within the cluster, and the evaluation results of adjustable capacity balance deviation between clusters. This ensures that the output results fully meet the multi-objective optimization requirements of maximizing feature similarity within clusters, equalizing adjustable capacity between clusters, and ensuring that each cluster contains at least one shared adjustable load resource.

[0082] It should be further noted that the filtering module in this embodiment includes:

[0083] The hypergraph generation unit is used to generate an overlapping cluster hypergraph by using the load clusters as vertices and, if at least two load clusters have the same shared adjustable load resource, using the shared adjustable load resource as a hyperedge to connect the corresponding load clusters; wherein the vertices of the overlapping cluster hypergraph are load clusters, the hyperedges are shared adjustable load resources, and each hyperedge connects at least two vertices.

[0084] G1. Based on the multiple load clusters output by the clustering module, each load cluster is abstracted into a vertex element in a graph structure to establish a load cluster vertex set. Each vertex in the vertex set is uniquely mapped to the corresponding load cluster, and each vertex carries exclusive load cluster identification information and complete member resource metadata. The load cluster identification information is a unique code used to distinguish different load clusters. The member resource metadata includes the unique identifier, real-time adjustable capacity, response rate, cost coefficient, and shared resource marking status of all adjustable load resources in the load cluster.

[0085] G2. Based on the shared adjustable load resource set, a sequential traversal algorithm is used to traverse all shared adjustable load resources. For each traversed shared adjustable load resource, its corresponding load aggregation cluster set is retrieved through the resource-cluster association mapping table. The resource-cluster association mapping table is a preset association data structure in the output result of the clustering module, which records the correspondence between each adjustable load resource and its corresponding load aggregation cluster. If the retrieved load aggregation cluster set contains at least two different load aggregation cluster identifiers, it is determined that the shared adjustable load resource has cross-cluster connection capability, and it is marked as a valid connection unit. At the same time, the unique resource identifier of the valid connection unit and the list of its corresponding load aggregation cluster identifiers are recorded.

[0086] G3. Construct each valid connection unit as a hyperedge element. The hyperedge element is bound to the valid connection unit according to the "one-to-one correspondence" principle, and the hyperedge simultaneously connects the vertices corresponding to all load aggregation clusters to which the valid connection unit belongs. Each hyperedge element has a built-in data storage structure that records the unique identifier of the shared adjustable load resource it represents, the complete list of connected vertex sets, the adjustable capacity parameters of the resource, and the flexibility weight information. The complete list of connected vertex sets clearly marks the load aggregation cluster identifier corresponding to each vertex.

[0087] G4. Based on the established set of vertices in the load aggregation cluster and the constructed hyperedge connection relationships, a complete overlapping cluster hypergraph data structure is constructed using graph structure combination technology. The data storage method of the overlapping cluster hypergraph selects one or a combination of vertex-hyperedge association matrix or adjacency list structure. The vertex-hyperedge association matrix represents the association relationship between vertices and hyperedges through a two-dimensional array, and the adjacency list structure associates all corresponding hyperedges through a vertex linked list. To ensure query efficiency, a bidirectional vertex-hyperedge index is established in the storage structure to ensure that all hyperedge information associated with any vertex and all vertex information connected by any hyperedge can be quickly located and queried through the index, realizing efficient retrieval and traversal of association relationships.

[0088] G5. A hypergraph structure verification algorithm is used to verify the structure of the generated overlapping cluster hypergraph. The verification includes a first verification item and a second verification item. The first verification item is to check whether the number of vertices connected by each hyperedge is not less than two. The second verification item is to check whether all vertices connected by all hyperedges belong to the load aggregation cluster vertex set established in step G1. If all hyperedges satisfy the first verification item and all vertices satisfy the second verification item, the verification is deemed successful. After the verification is successful, complete overlapping cluster hypergraph structure data is output. The structure data includes details of the load aggregation cluster vertex set, details of the hyperedge set, and a list of vertex-hyperedge association relationships to ensure that the output data can directly support subsequent hyperedge value calculation and candidate set screening operations.

[0089] The value calculation unit is used to calculate the initial upper bound of the value of each hyperedge in the hypergraph generation unit; the initial upper bound of the value is calculated based on the adjustable capacity of the shared adjustable load resources, the current grid compensation unit price, the average dispatch cost, and the flexibility weight, wherein the flexibility weight is determined according to the response rate and adjustment direction of the shared adjustable load resources; the adjustment direction is defined as the type of adjustment that can be performed within the current dispatch cycle, including at least upward adjustment only, downward adjustment only, or bidirectional adjustment.

[0090] It should be further explained that one method for calculating the upper bound of the initial value in this embodiment includes:

[0091] B1. Obtain the adjustable capacity parameters of the shared adjustable load resources. The adjustable capacity parameters are collected and updated in real time by the data acquisition unit, representing the adjustable power range of the shared adjustable load resources within the current scheduling cycle.

[0092] B2. Obtain the current power grid compensation unit price parameter. The current power grid compensation unit price parameter is obtained in real time through the virtual power plant centralized control platform and is determined based on power market transaction data or power grid demand response instructions.

[0093] B3. Calculate the average scheduling cost parameter, that is, based on historical scheduling data or real-time cost information, obtain the average unit power scheduling cost of the shared adjustable load resources by using the arithmetic average method, and obtain the average scheduling cost parameter.

[0094] B4. Determine the flexibility weight parameter. The flexibility weight parameter is calculated based on the response rate parameter and adjustment direction parameter of the shared adjustable load resource. The response rate parameter is acquired by the data acquisition unit and characterizes the power adjustment capability per unit time. The adjustment direction parameter is defined by an enumeration type, including three types: adjustable only upward, adjustable only downward, or adjustable in both directions. The calculation formula for the flexibility weight parameter is: Flexibility weight = Response rate coefficient × Adjustment direction coefficient. The response rate coefficient is determined based on the ratio of the response rate parameter to a preset speed threshold. The adjustment direction coefficient is mapped to a predefined value according to the adjustment direction parameter, and the highest coefficient value is adjustable in both directions. In this embodiment, the preset speed threshold is set based on the technical requirements of the grid demand response command for the adjustment speed. By analyzing the maximum allowable adjustment time and target adjustment power specified in the command, the minimum power change rate required per unit time is calculated. The result is calibrated by combining the response capability distribution of typical load resources in the system, and finally forming a benchmark value for quantitatively evaluating the relative level of resource adjustment speed.

[0095] B5. Calculate the initial value upper bound. Based on the adjustable capacity parameter, current power grid compensation unit price parameter, average dispatch cost parameter, and flexibility weight parameter obtained in steps B1 to B4, calculate the initial value upper bound using a weighted product formula. The weighted product formula is: Initial value upper bound = Adjustable capacity × (Current power grid compensation unit price - Average dispatch cost) × Flexibility weight.

[0096] The candidate set generation unit is used to sort all hyperedges in descending order based on the initial value upper bound calculated by the value calculation unit, and select the top N hyperedges with the highest initial value upper bound to form a candidate set. In this embodiment, the motivation for the candidate set generation unit is to solve the problem of exploding optimization complexity caused by the simultaneous participation of massive shared resources in scheduling. Its core principle is to establish a priority scheduling queue through value sorting and threshold filtering mechanisms. Based on the quantitative evaluation results of the initial value upper bound of the hyperedges, this unit uses a descending sorting algorithm to identify the top N most economical and flexible shared resources, forming a priority scheduling candidate set. The beneficial effect of this approach is that it achieves a reasonable allocation of optimization computing resources, avoiding the computational burden brought by global optimization, and ensuring that high-value resources participate in game decision-making first. This significantly improves system response efficiency while ensuring scheduling quality, providing an efficient search space for subsequent iterative optimization.

[0097] It should be further noted that the iteration module in this embodiment includes:

[0098] The shared resource selection unit is used to select the shared adjustable load resource with the highest upper bound of the current initial value from the candidate set as the current shared device;

[0099] The cost game unit, based on the load aggregation clusters connected by the superedges corresponding to the currently shared adjustable load resources, determines the joint demand response strategy that maximizes revenue through a cost game model. The cost game unit adopts a cooperative game model to calculate the total alliance revenue under different feasible joint scheduling strategies, and allocates the revenue through the Shapley value to select the joint demand response strategy that maximizes the total alliance revenue. The total alliance revenue is the difference between the grid payment compensation and the total cost of all scheduled adjustable load resources.

[0100] It should be further explained that, in this embodiment, the joint demand response strategy that maximizes benefits is determined through a cost-game model, including:

[0101] C1. Based on the load aggregation clusters connected to the superedge corresponding to the current shared adjustable load resources, all load aggregation clusters associated with the superedge and all adjustable load resources contained therein are jointly formed into an adjustable load resource alliance.

[0102] C2. Based on the total regulating power demand in the grid demand response command, and considering the real-time adjustable capacity, regulation direction, and power change rate constraints of each adjustable load resource within the alliance, a set of all feasible joint dispatch strategies that satisfy the global regulating demand is generated using a linear programming algorithm. Each feasible joint dispatch strategy must fully define the specific dispatch power value and regulation direction of each resource within the alliance. The global regulating demand represents the core command issued by the grid side to the virtual power plant alliance as a whole within a specific dispatch cycle, specifying the total amount and direction of power adjustments. Specifically, it is quantified as the absolute value of the total active power regulation to be achieved and a clearly defined regulation type. This demand serves as the basis for the linear programming algorithm. The rigid constraints of the linear programming model ensure that the generated joint scheduling strategy can accurately meet the macro-regulation objectives of the power grid. In this embodiment, to achieve coordinated scheduling of cross-cluster shared adjustable load resources to maximize alliance benefits, the load aggregation clusters and their resources connected by the current shared resources' hyperedges must first be formed into an adjustable load resource alliance. The principle is based on a linear programming algorithm. Under constraints such as the total power regulation demand of the power grid, the real-time adjustable capacity of each resource, the regulation direction (e.g., charging / discharging, load increase / decrease), and the power change rate, a set of feasible joint scheduling strategies that satisfy global requirements is solved, ensuring that resource scheduling conforms to power grid commands without violating equipment operating limitations. For example, a shared energy storage resource connects an industrial load cluster and a commercial building cluster to form an alliance. When the power grid requires an emergency load reduction, the charging / discharging capacity of the energy storage, the load reduction capacity of the industrial equipment, the regulation range of the commercial air conditioners, and their respective power change rates are combined. Linear programming generates a set of all feasible scheduling strategies that satisfy the total load reduction demand, including "energy storage discharge + industrial load reduction + commercial air conditioner temperature adjustment and power reduction," providing a foundation for subsequent benefit optimization and strategy selection.

[0103] C3. Using the total revenue of the adjustable load resource alliance as the objective function, a target constraint space is constructed based on the boundary values ​​of the dispatch power and real-time adjustable capacity of each adjustable load resource, the sustainable dispatch duration of each adjustable load resource and the minimum duration required by the grid demand response command, as well as the power change rate and rate penalty factor of each adjustable load resource and the upper and lower limits of the power change rate constraints. To quantify the impact of adjustable load resource adjustment time on dispatch costs and avoid grid demand response timeouts due to slow adjustment, a rate penalty factor needs to be constructed, where the rate penalty factor is determined by the grid demand response command. The maximum allowable adjustment time and the rated power change rate of the corresponding adjustable load resources are used to determine the contribution of adjustment time cost to the objective function. For example, the rated power change rate of an industrial energy storage is 3 MW / min, the maximum allowable adjustment time of the grid is 12 minutes, and its rate penalty factor is 12 / 3=4; while the rated power change rate of a commercial air conditioner is 1 MW / min, and the corresponding penalty factor is 12 / 1=12. This shows that the time cost contribution of commercial air conditioners is higher, and their adjustment timeliness should be given more attention during scheduling to optimize the total cost and response time.

[0104] C4. Based on the set of feasible joint scheduling strategies, the Shapley value algorithm in cooperative game theory is used to calculate the expected revenue share of each load aggregation cluster under each feasible joint scheduling strategy, specifically:

[0105] C41. Generate all sub-alliance combinations of load aggregation clusters in the adjustable load resource alliance. For each load aggregation cluster, calculate its marginal contribution value in each sub-alliance. The marginal contribution value is defined as the change in the total revenue of the corresponding sub-alliance before and after the load aggregation cluster is added to each sub-alliance. Generating all sub-alliance combinations of load aggregation clusters in the load resource alliance includes:

[0106] C411. Obtain all load aggregation clusters included in the load resource alliance to form a complete alliance member set. Each of them This represents a load cluster; where n represents the total number of load clusters.

[0107] C412. Based on the set of alliance members U, generate all possible subsets of the set using binary encoding or a recursive algorithm, including all combinations from a single member to n-1 members, excluding the empty set and the entire set.

[0108] C413. Transform all subsets generated in C412 into corresponding sub-alliance combinations to establish a complete set of sub-alliance combinations. Each of them Represents a specific sub-alliance, m=2 n -2;

[0109] C414. Establish a mapping relationship between each sub-alliance combination and the original alliance member set, record the specific load aggregation cluster member information contained in each sub-alliance, form a complete combination relationship mapping table, and output a complete data set containing all sub-alliance combinations and their member relationships.

[0110] C42. Based on the marginal contribution values ​​of all load aggregation clusters, calculate the expected revenue share of each load aggregation cluster in the adjustable load resource consortium according to the Shapley value calculation formula, wherein the calculation formula is the weighted average of each marginal contribution value in the ranking of all sub-consolidations; wherein the process of calculating the expected revenue share of each load aggregation cluster in the adjustable load resource consortium includes:

[0111] C421, For the i-th load cluster where the expected revenue share is to be calculated. Iterate through all sub-alliance combinations that do not contain Sub-alliance ;

[0112] C422, for each sub-alliance Obtain the load aggregation cluster from the marginal contribution value storage structure. The change in total alliance revenue before and after joining the sub-alliance is denoted as the marginal contribution value. ;

[0113] C423, for each sub-alliance Calculate the corresponding Shapley value weighting factor, and the formula for calculating the weighting factor is as follows: ,in Sub-alliance The number of load aggregation clusters included, where n represents the total number of load aggregation clusters in the adjustable load resource alliance;

[0114] C424. Multiply each obtained marginal contribution value by the corresponding calculated Shapley value weighting factor to obtain the weighted marginal contribution value for each sub-alliance.

[0115] C425. Using the Shapley value formula, sum all weighted marginal contribution values ​​to obtain the load aggregation cluster. Expected revenue share in the adjustable load resource alliance ;

[0116] C426. Repeat C421 to C425 until the expected revenue share of each load aggregation cluster in the adjustable load resource alliance is calculated, forming a complete revenue distribution scheme.

[0117] C5. For each feasible joint scheduling strategy, calculate its corresponding total alliance revenue. Simultaneously, based on the expected revenue share distribution of each load aggregation cluster, calculate the revenue distribution fairness index under each feasible joint scheduling strategy using the Lorenz curve and Gini coefficient formula. It should be further explained that the process of calculating the revenue distribution fairness index under each feasible joint scheduling strategy using the Lorenz curve and Gini coefficient formula in this embodiment includes:

[0118] C51. Obtain the expected revenue share of each load aggregation cluster under the current feasible joint scheduling strategy, and form a revenue distribution dataset;

[0119] C52. Arrange each load cluster in ascending order of its expected revenue share to form an ordered revenue sequence;

[0120] C53. Calculate the cumulative proportion of the number of load clusters from the first load cluster to the i-th load cluster to the total number of load clusters, and form a sequence on the horizontal axis.

[0121] C54. Calculate the cumulative proportion of the sum of the revenue shares from the first load cluster to the i-th load cluster to the total revenue, forming a vertical axis sequence.

[0122] C55. Using the proportion of cumulative load clusters as the horizontal axis and the proportion of cumulative revenue share as the vertical axis, connect the data points sequentially in the coordinate system to form the Lorenz curve, which represents the actual revenue distribution.

[0123] C56. Draw a diagonal line from the origin to the destination in the same coordinate system. This diagonal line represents the ideal state of perfectly equal distribution of benefits.

[0124] C57. Calculate the area of ​​the region enclosed by the Lorenz curve and the line of absolute fairness. This area value represents the gap between the actual distribution of benefits and the state of perfect fairness.

[0125] C58. Divide the area difference obtained in C57 by the area of ​​the entire triangle below the absolute fairness line to obtain the quantified Gini coefficient value, which is the fairness index of revenue distribution under the current feasible joint scheduling strategy.

[0126] C59. The calculated Gini coefficient value is associated with the corresponding feasible joint scheduling strategy and stored to provide complete fairness quantification data for subsequent multi-objective strategy evaluation.

[0127] C6. A multi-objective strategy evaluation model is established by using the maximization of total alliance revenue and the minimization of the Gini coefficient as parallel optimization objectives, and mapping the feature vector of each feasible joint scheduling strategy to a two-dimensional evaluation space composed of total alliance revenue and fairness indicators. It should be further noted that the specific implementation steps for establishing the multi-objective strategy evaluation model in this embodiment are as follows:

[0128] C61. Obtain the total alliance revenue value and the revenue distribution fairness index value calculated based on the Gini coefficient for each feasible joint scheduling strategy, and form a complete evaluation dataset for each strategy.

[0129] C62. Establish a two-dimensional rectangular coordinate system evaluation space with the total revenue of the alliance as the first dimension and the fairness index of revenue distribution as the second dimension, and complete the mathematical foundation construction of the evaluation framework.

[0130] C63. The total alliance revenue value and the revenue distribution fairness index value of each feasible joint scheduling strategy are used as coordinate values ​​and mapped onto the two-dimensional evaluation space to form the coordinate points corresponding to each strategy.

[0131] C64. In the two-dimensional evaluation space, compare all strategy coordinate points pairwise to identify those strategy points that are not completely surpassed by other strategies in both the total alliance revenue and the fairness of revenue distribution, and mark them as non-dominant strategies.

[0132] C65. Based on the non-dominance relationship determination results, all strategies are divided into different levels: the first level contains all non-dominance strategies, and subsequent levels contain strategies dominated by the strategies of the previous level, forming a complete Pareto level sequence.

[0133] C66. Within the same Pareto level, calculate the combined distance between each policy point and its neighboring policy points on two evaluation dimensions, and quantify the distribution density characteristics of each policy in its level.

[0134] C67. Based on the Pareto rank and congestion distance calculation results, higher-rank strategies are prioritized, and within the same rank, strategies with larger congestion distances are prioritized, forming an optimized strategy set. The output is a complete multi-objective strategy evaluation model containing the distribution information of all feasible joint scheduling strategies in a two-dimensional evaluation space, Pareto rank classification, and congestion calculation results, providing a decision-making basis for the final strategy selection. The motivation for establishing this multi-objective strategy evaluation model in this embodiment is to break through the limitation of traditional scheduling focusing only on economic benefits. Its core principle is to construct a two-dimensional evaluation space of total alliance revenue and Gini coefficient, simultaneously incorporating the inherently conflicting objectives of economic efficiency and allocation fairness into a unified evaluation framework. This model uses non-dominated ranking to identify the Pareto optimal solution set, avoiding the human bias of subjective weighting of multiple objectives. Furthermore, congestion calculation ensures the diversity of the solution set's distribution, thereby systematically generating a strategy set that achieves the best balance between economic benefits and fairness, providing a comprehensive and scientific basis for the final decision.

[0135] C7. Based on the multi-objective strategy evaluation model and combined with the non-dominated sorting algorithm in multi-objective optimization, the feasible joint scheduling strategy set is screened to obtain a Pareto optimal strategy subset; each feasible joint scheduling strategy in the Pareto optimal strategy subset satisfies the following: it is a non-dominated solution for both the total alliance revenue and the Gini coefficient, that is, there is no other feasible joint scheduling strategy that can improve the other objective without reducing the performance of either objective;

[0136] C8. In the Pareto optimal strategy subset, the strategy with the highest total alliance revenue is selected as a candidate strategy. If multiple Pareto optimal strategies have the same highest total alliance revenue, the variance of the expected revenue share of each load aggregation cluster under the multiple Pareto optimal strategies with the same highest total alliance revenue is calculated, and the Pareto optimal strategy with the smallest variance is selected as the final joint demand response strategy. The final joint demand response strategy includes: unique identifiers of all adjustable load resources participating in the final joint demand response strategy and their scheduling power values, a unified scheduling time window for strategy execution, the expected total alliance revenue value, and a detailed table of expected revenue share calculated by each load aggregation cluster based on the Shapley value.

[0137] The resource status update unit is used to synchronously delete the adjustable load resources included in the joint demand response strategy determined by the cost game unit from the corresponding load aggregation clusters, mark the adjustable load resources as scheduled, and synchronously update the remaining adjustable capacity and composition of the load aggregation clusters. The motivation for setting up the resource status update unit in this embodiment is to solve the scheduling conflicts and resource duplication problems caused by asynchronous resource states during dynamic optimization. Its core principle is to establish a closed-loop state synchronization mechanism of "scheduling-marking-updating." After determining the joint strategy in each round of the game, the scheduled resources are immediately removed from the aggregation clusters and their status is marked, and key parameters such as the remaining adjustable capacity of each cluster are synchronously updated. This ensures strict consistency between the optimization model and the real-time resource state, avoiding resource conflicts in subsequent iterations. Furthermore, by dynamically updating the composition structure of the load aggregation clusters, accurate input conditions are provided for the next round of optimization, thus ensuring the coherence and solution quality of multiple rounds of iterative optimization.

[0138] The update iteration unit, based on the output of the resource status update unit, synchronously updates the structure of the overlapping cluster hypergraph and the candidate set, specifically including: deleting the hyperedge corresponding to the currently scheduled shared adjustable load resource from the overlapping cluster hypergraph;

[0139] All adjustable load resources included in the joint demand response strategy determined in the corresponding iteration round are synchronously deleted from the respective load clusters to which they belong. It is then determined whether any load cluster still contains any shared adjustable load resources. If not, the load cluster is marked as a non-shared cluster. At the same time, the initial value upper bound of all hyperedges whose connection relationship has changed due to the above deletion operation is recalculated, and the candidate set is updated based on the recalculation result.

[0140] Once the candidate set is updated, it is checked. If it is not empty, the next iteration is triggered; otherwise, the iteration optimization process is terminated. The motivation of this unit is to avoid redundant allocation of scheduled resources, invalid hyperedges occupying computational resources, and distortion of candidate set value, ensuring that iterative optimization is always efficiently advanced based on real-time resource status and accurate value assessment. The principle is to synchronously update the overlapping cluster hypergraph and the candidate set. First, hyperedges corresponding to scheduled shared adjustable load resources are deleted to eliminate invalid connections. Then, all adjustable load resources included in the joint demand response strategy are removed from each member load cluster. It is determined whether the load cluster still contains shared adjustable load resources and marked as a non-shared cluster. The initial value upper bound of hyperedges with changed connection relationships is recalculated to correct the candidate set. Finally, the process is determined by checking whether the candidate set is empty to decide whether to continue iteration or terminate the process. This ensures that the optimization objects in each iteration are effective resources and high-value hyperedges, achieving precise convergence of the iteration process and maximizing global benefits.

[0141] It should be further noted that the strategy generation module in this embodiment includes:

[0142] The strategy aggregation unit is used to aggregate the joint demand response strategies generated by the iterative module in each iteration. The strategy aggregation unit records detailed parameters for each joint demand response strategy, including scheduling time, adjustable load resource list, and expected revenue. The expected revenue is quantified by calculating the total alliance revenue, specifically the difference between the total compensation paid by the grid and the total scheduling cost of all participating adjustable load resources after executing the joint demand response strategy. The total compensation is determined based on the product of the compensation unit price published by the grid and the total regulating power achieved by the joint demand response strategy. The process of obtaining the total scheduling cost includes: first, extracting the unit regulating cost coefficient and actual scheduling power parameter of each resource from the adjustable load resource list; then, multiplying the unit regulating cost coefficient of each resource by its actual scheduling power to obtain the individual scheduling cost of that resource; finally, summing the individual scheduling costs of all resources using an accumulation algorithm to output the total scheduling cost value of the system. The adjustable load resource list includes unique identifiers of all adjustable load resources participating in the joint demand response strategy and their scheduled power values.

[0143] The residual load processing unit, based on the real-time adjustable capacity, unit adjustment cost coefficient, and current grid adjustment demand of each isolated adjustable load resource, independently generates a final dispatch instruction with a fixed power command for each resource, and integrates and outputs all generated final dispatch instructions to the final strategy generation unit. The motivation of this unit is to fully explore the adjustment potential of isolated adjustable load resources not covered by joint dispatch after iterative optimization, avoid resource idleness, fill the gap in grid adjustment demand, and ensure the full implementation of grid demand response instructions. The principle is based on the real-time adjustable capacity (reflecting actual adjustment capability), unit adjustment cost coefficient (reflecting dispatch economy), and the current unmet residual adjustment demand of the grid, abandoning complex collaborative dispatch mode, independently formulating a fixed power command adapted to its capability and cost for each isolated resource, ensuring both the feasibility and economy of individual resource dispatch, and ensuring the accurate filling of the grid's residual adjustment demand by integrating all instructions. Finally, the integrated dispatch instruction is output to the final strategy generation unit to achieve efficient utilization of all adjustable load resources and complete response to grid demand.

[0144] It should be further explained that the process of generating a final scheduling instruction with a fixed power command independently for each resource in this embodiment includes:

[0145] Based on the status of adjustable load resources after the termination of iterative optimization, adjustable load resources that are not covered by any joint demand response strategy and are not marked as scheduled are selected to form an isolated adjustable load resource set. The real-time adjustable capacity and unit adjustment cost coefficient of each isolated adjustable load resource in the isolated adjustable load resource set are obtained through the data synchronization interface. At the same time, the unmet remaining grid adjustment demand is obtained from the grid dispatching platform, including the total remaining adjustment power, adjustment direction and execution time window. A parameter verification mechanism is used to verify the completeness and accuracy of the collected resource parameters and grid demand parameters to ensure that the parameters meet the requirements for generating dispatching instructions.

[0146] The demand difference calculation method is used to subtract the regulation power covered by all joint demand response strategies from the total regulation demand of the power grid to obtain the current remaining regulation power of the power grid; the type of the remaining regulation demand of the power grid is determined by the regulation direction determination logic, and the effective duration of the dispatching instruction is determined by the execution time window to form a standardized power grid remaining demand parameter package.

[0147] For each isolated adjustable load resource, a matching verification is performed based on the adjustment direction corresponding to its real-time adjustable capacity and the adjustment direction in the power grid remaining demand parameter package to determine whether the resource adjustment direction is consistent with the power grid remaining adjustment demand; at the same time, it is verified whether the real-time adjustable capacity of the isolated adjustable load resource meets the minimum adjustment power threshold of the power grid remaining adjustment demand, and whether its unit adjustment cost coefficient does not exceed the power grid's preset scheduling cost upper limit, and a subset of effective isolated adjustable load resources that simultaneously meet the requirements of direction matching, capacity compliance and cost compliance are selected;

[0148] A cost-priority allocation algorithm is adopted to sort the effective isolated adjustable load resource subset in ascending order according to the unit adjustment cost coefficient, and prioritize the scheduling of the isolated adjustable load resource with the lowest unit adjustment cost coefficient; based on the difference between the real-time adjustable capacity limit of each isolated adjustable load resource and the remaining adjustment power of the power grid, a fixed adjustment power not exceeding the real-time adjustable capacity limit is allocated to each resource, and the execution time window, adjustment direction and fixed power value corresponding to the fixed power instruction are defined, generating an independent standardized scheduling instruction for each isolated adjustable load resource;

[0149] All generated independent standardized dispatch instructions are categorized and integrated according to adjustable load resource type and adjustment direction. The deviation between the integrated total adjustable power and the remaining adjustable power of the power grid is calculated using the total power summation verification method. If the deviation exceeds the preset allowable range, the deviation is compensated by fine-tuning the fixed power value of the isolated adjustable load resource with the second lowest unit adjustment cost coefficient until the integrated total adjustable power matches the remaining adjustable power of the power grid. At the same time, the parameter format of each standardized dispatch instruction is standardized, clearly defining the core fields including resource unique identifier, fixed power value, execution time period, and adjustment direction to ensure that the instruction can be directly parsed and executed by the corresponding adjustable load terminal.

[0150] All final scheduling instructions after integration and verification are transmitted to the final strategy generation unit through the communication interface. At the same time, the generation log of each final scheduling instruction is recorded. The generation log contains parameter information, power allocation logic and cost accounting results of the corresponding isolated adjustable load resources, providing data support for subsequent scheduling effect tracking and strategy optimization.

[0151] The final strategy generation unit is used to receive and integrate the joint demand response strategy set output by the strategy aggregation unit and the final scheduling instruction set output by the residual load processing unit, and to send the final scheduling instruction set to the corresponding adjustable load terminal for execution through the virtual power plant centralized control platform; the scheduling parameters of the final scheduling instruction include at least the target load value, the execution time window and the equipment identifier.

[0152] This application achieves efficient aggregation and precise response of adjustable load resources in a virtual power plant by constructing a collaborative control architecture of clustering-screening-iteration-generation. Specifically, in the clustering stage, the system employs a multi-objective clustering algorithm, aggregating heterogeneous load resources into load clusters with similar internal characteristics, balanced capacity, and shared resources based on multi-dimensional feature vectors. This establishes a fundamental infrastructure for resource collaboration from the source, solving the problem of scheduling fragmentation caused by resource dispersion. In the screening stage, by constructing an overlapping cluster hypergraph with load clusters as vertices and shared resources as hyperedges, the conflict sources in traditional technologies are transformed into collaborative anchor points. Furthermore, based on adjustable capacity, grid compensation unit price, scheduling cost, and flexibility... The active weight calculation of the upper bound of the hyperedge value effectively reduces the complexity of the optimization problem and ensures that high-value resources participate in scheduling first. In the iterative optimization phase, the system employs a coalition game mechanism. For each candidate shared resource's corresponding load coalition, a set of feasible strategies that meet global adjustment requirements is generated through linear programming. The Shapley value algorithm is used for revenue allocation, establishing a multi-objective evaluation model aimed at maximizing the total coalition revenue and minimizing the Gini coefficient. Non-dominated sorting is used to select Pareto optimal strategies. This organic combination of techniques ensures the optimal balance between economy and fairness in the scheduling scheme. In the dynamic update phase, a closed-loop mechanism of "scheduling-update-re-optimization" is established through resource state synchronization updates and overlapping cluster hypergraph topology reconstruction, ensuring the coherence of multiple iterations and the quality of the solution. Finally, the system generates a complete executable scheme by integrating the joint strategy and residual load instructions. These technological innovations work together to transform the conflicts caused by sharing into advantages driven by collaboration, fundamentally solving the technical dilemmas of traditional methods such as low resource integration efficiency, large response delays, and significant execution deviations, and achieving simultaneous improvement in scheduling accuracy, response rate, and economic benefits.

[0153] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A virtual power plant adjustable load aggregation and demand response control system, characterized in that, The method comprises the following steps: A clustering module is used to obtain attribute information of power grid demand response instructions and adjustable load resources, and to perform multi-objective clustering through a clustering algorithm to obtain a plurality of load aggregation clusters, wherein at least one shared adjustable load resource is included in the load aggregation cluster; A screening module takes the load aggregation cluster as a vertex, and if there is a shared adjustable load resource between at least two load aggregation clusters, takes the shared adjustable load resource as a superedge connecting the corresponding load aggregation cluster, generates an overlapping cluster hypergraph, and screens a candidate set through calculation of an initial upper bound of the value of the superedge; An iteration module is used to perform an iteration operation, to select a current shared adjustable load resource from the candidate set, to determine a joint demand response strategy maximizing the revenue based on the load aggregation clusters connected by the superedge corresponding to the current shared adjustable load resource through a cost game, and to synchronously delete the adjustable load resource in the joint demand response strategy from the load aggregation cluster, to update the overlapping cluster hypergraph and the candidate set until the candidate set is empty; A strategy generation module is used to generate a final demand response control strategy by aggregating the joint demand response strategy of each iteration and the adjustable load resource not deleted in the load aggregation cluster. The screening module comprises: A hypergraph generation unit is used to take the load aggregation cluster as a vertex, and if there is a same shared adjustable load resource between at least two load aggregation clusters, to take the shared adjustable load resource as a superedge connecting the corresponding load aggregation cluster, and to generate an overlapping cluster hypergraph; A value calculation unit is used to calculate an initial upper bound of the value of each superedge in the hypergraph generation unit; the initial upper bound is calculated by weighting the adjustable capacity of the shared adjustable load resource, the current power grid compensation unit price, the average dispatching cost and the flexibility weight, wherein the flexibility weight is determined according to the response rate and the adjustment direction of the shared adjustable load resource; the adjustment direction is defined as the adjustment type that can be performed in the current dispatching period, and at least includes only up-regulation, only down-regulation or bidirectional regulation; A candidate set generation unit is used to sort all superedges in descending order based on the initial upper bound of the value calculated by the value calculation unit, and to screen the first N superedges with the highest initial upper bound of the value to form a candidate set.

2. The controllable load aggregation and demand response control system of a virtual power plant of claim 1, wherein, The clustering module comprises: A data acquisition unit is used to acquire attribute information of adjustable load resources in real time, including adjustable capacity, response time, cost coefficient, user compensation willingness, and dispatching requirements and time constraints in the power grid demand response instruction; the data acquisition unit is also connected with a virtual power plant centralized control platform through a communication interface to obtain power grid demand response instructions and real-time electricity price information; A feature vector unit is used to construct a multi-dimensional feature vector of each adjustable load resource based on the attribute information acquired by the data acquisition unit; the multi-dimensional feature vector comprises a static attribute dimension, a dynamic attribute dimension and a willingness attribute dimension, wherein the static attribute dimension comprises rated power and geographical position, the dynamic attribute dimension comprises real-time adjustable capacity and response rate, and the willingness attribute dimension comprises a user-set cost coefficient and a sustainable response duration.

3. The controllable load aggregation and demand response control system of a virtual power plant of claim 2, wherein, The clustering module further comprises: The clustering algorithm unit performs adjustable load resource clustering based on the multi-dimensional feature vector and a multi-objective clustering algorithm to generate a plurality of load aggregation clusters; the multi-objective clustering algorithm maximizes the similarity of the characteristics within a cluster and balances the adjustable capacity between clusters, and ensures that each load aggregation cluster contains at least one shared adjustable load resource, which is an adjustable load resource that can be scheduled by multiple load aggregation clusters.

4. The controllable load aggregation and demand response control system of a virtual power plant of claim 3, wherein, The iteration module includes: The shared resource selection unit selects a shared adjustable load resource with the highest current initial value upper bound from the candidate set as a current shared device; The cost game unit determines a joint demand response strategy that maximizes revenue based on the load aggregation clusters connected by the hyperedge corresponding to the current shared adjustable load resource through a cost game model; the cost game unit uses a cooperative game model to calculate the total alliance revenue under different feasible joint scheduling strategies and allocates the revenue through Shapley value to select a joint demand response strategy that maximizes the total alliance revenue, which is the difference between the compensation paid by the power grid and the total cost of all scheduled adjustable load resources.

5. A virtual power plant adjustable load aggregation and demand response control system as claimed in claim 4, wherein, The iteration module further includes: The resource state update unit synchronously deletes the adjustable load resources contained in the joint demand response strategy determined by the cost game unit from the corresponding load aggregation clusters, marks the adjustable load resources as scheduled, and synchronously updates the remaining adjustable capacity and composition of the load aggregation clusters; The update iteration unit synchronously updates the structure of the overlapping cluster hypergraph and the candidate set based on the output of the resource state update unit, specifically including: deleting the hyperedge corresponding to the current shared adjustable load resource that has been scheduled from the overlapping cluster hypergraph; Synchronously delete all adjustable load resources contained in the joint demand response strategy determined in the corresponding iteration round from each of the load aggregation clusters to which they belong, and determine whether any of the load aggregation clusters still contains any shared adjustable load resource, if not, mark the load aggregation cluster as a non-shared cluster, and recalculate the initial value upper bound of the hyperedge whose connection relationship has changed due to the deletion operation, and update the candidate set based on the recalculated result; When the candidate set is updated, check the updated candidate set, if it is not an empty set, trigger the next iteration operation; if it is an empty set, terminate the iteration optimization process.

6. A virtual power plant adjustable load aggregation and demand response control system as claimed in claim 5, wherein, The strategy generation module includes: a strategy aggregation unit configured to aggregate the joint demand response strategies generated by the iteration module in each iteration round, and record detailed parameters of each joint demand response strategy, including a dispatch time, a list of adjustable load resources, and an expected revenue, wherein the expected revenue is quantitatively represented by calculating a total revenue of the alliance, and a specific value is a difference between a total compensation fee paid by the power grid and a total dispatch cost of all participating adjustable load resources after the joint demand response strategy is executed, the total compensation fee is determined based on a compensation unit price issued by the power grid and a total regulation power achieved by the joint demand response strategy, and the total dispatch cost is a sum of costs of the adjustable load resources in the list of adjustable load resources calculated according to unit regulation cost coefficients and actual dispatch powers of the adjustable load resources; a residual load processing unit configured to independently generate a final dispatch instruction with a fixed power instruction for each isolated adjustable load resource based on real-time adjustable capacities, unit regulation cost coefficients, and current power grid regulation demands of the adjustable load resources, and integrate and output all generated final dispatch instructions to a final strategy generation unit; a final strategy generation unit configured to receive and integrate the joint demand response strategy set output by the strategy aggregation unit and the final dispatch instruction set output by the residual load processing unit, and send the final dispatch instruction set to corresponding adjustable load terminals for execution through the virtual power plant centralized control platform, and collect real-time execution result information of the final dispatch instruction set by using the data acquisition unit, so as to cyclically adjust an execution process of the power grid demand response instruction, wherein a dispatch parameter of the final dispatch instruction at least includes a target load value, an execution time window, and a device identifier.

7. A virtual power plant adjustable load aggregation and demand response control system as claimed in claim 6, wherein, determining a joint demand response strategy maximizing the revenue through a cost game model, including: based on the load aggregation clusters connected by the hyperedge corresponding to the current shared adjustable load resource, all the load aggregation clusters associated with the hyperedge and all the adjustable load resources contained therein are collectively formed into an adjustable load resource alliance; according to a total regulation power demand in the power grid demand response instruction, in combination with real-time adjustable capacities, regulation directions, and power change rate constraints of the adjustable load resources in the alliance, a linear programming algorithm is used to generate a set of all feasible joint dispatch strategies satisfying the global regulation demand, wherein each feasible joint dispatch strategy needs to completely define a specific dispatch power value and a regulation direction of each resource in the alliance; The total income of the adjustable load resource alliance is taken as a target function, and a target constraint space is constructed by using the boundary value of the scheduling power and the real-time adjustable capacity of each adjustable load resource, the minimum sustained period required by the power grid demand response instruction and the sustained scheduling time length of each adjustable load resource, and the upper limit and lower limit constraint of the power change rate and the rate penalty factor of each adjustable load resource. The rate penalty factor is constructed by the maximum adjustment time allowed in the power grid demand response instruction and the rated power change rate of the corresponding adjustable load resource, and is used to adjust the contribution degree of the time cost in the target function.

8. A virtual power plant adjustable load aggregation and demand response control system as claimed in claim 7, wherein, The joint demand response strategy maximizing the income is determined through the cost game model, and the joint demand response strategy maximizing the income further includes: Based on the set of feasible joint scheduling strategies, the Shapley value algorithm in cooperative game is used to calculate the expected income share of each load aggregation cluster under each feasible joint scheduling strategy, specifically: All sub-alliance combinations of the load aggregation clusters in the adjustable load resource alliance are generated, and for each load aggregation cluster, the marginal contribution value of the load aggregation cluster in each sub-alliance is calculated, and the marginal contribution value is defined as the change amount of the total income of the corresponding sub-alliance before and after the load aggregation cluster joins each sub-alliance; Based on the marginal contribution values of all load aggregation clusters, the expected income share of each load aggregation cluster in the adjustable load resource alliance is calculated according to the Shapley value calculation formula, and the calculation formula is the weighted average of the marginal contribution values in the order of all sub-alliances; For each feasible joint scheduling strategy, the total income of the corresponding alliance is calculated, and based on the distribution of the expected income share of each load aggregation cluster, the fairness index of the income distribution under each feasible joint scheduling strategy is calculated by using the Lorenz curve and Gini coefficient formula; The maximization of the total income of the alliance and the minimization of the Gini coefficient are taken as parallel optimization targets, and the feature vector of each feasible joint scheduling strategy is mapped into a two-dimensional evaluation space composed of the total income of the alliance and the fairness index, and a multi-objective strategy evaluation model is established.

9. A virtual power plant adjustable load aggregation and demand response control system as claimed in claim 8, wherein, The joint demand response strategy maximizing the income is determined through the cost game model, and the joint demand response strategy maximizing the income further includes: Based on the multi-objective strategy evaluation model, the set of feasible joint scheduling strategies is screened by using the non-dominated sorting algorithm in multi-objective optimization, and a Pareto optimal strategy subset is obtained; each feasible joint scheduling strategy in the Pareto optimal strategy subset satisfies: it is a non-dominated solution in the optimization targets of the total income of the alliance and the Gini coefficient, that is, there is no other feasible joint scheduling strategy that can improve one target without reducing the performance of the other target. In the Pareto optimal strategy subset, the strategy with the highest total alliance revenue is selected as the candidate strategy. If multiple Pareto optimal strategies have the same highest total alliance revenue, the variance of the expected revenue share of each load cluster under the multiple Pareto optimal strategies with the same highest total alliance revenue is calculated, and the Pareto optimal strategy with the smallest variance is selected as the final joint demand response strategy. The final joint demand response strategy includes: unique identifiers of all adjustable load resources participating in the final joint demand response strategy and their scheduling power values, a unified scheduling time window for strategy execution, the expected total alliance revenue value, and a detailed table of expected revenue share calculated by each load cluster based on the Shapley value.

Images