Virtual power plant edge cloud end double-layer structure collaborative optimization configuration system

Abstract

The application provides a virtual power plant edge cloud double-layer structure collaborative optimization configuration system, relates to the virtual power plant resource regulation and control technical field, and has the improved module degree of communication cost fusion as the target for partition aggregation. After the external characteristics of each aggregate are deduced and the power feasible region boundary is projected, the cloud is reported. The optimal alliance structure is formed by the market fit degree index driven optimization of the coalition game model. Finally, the market transaction instruction is optimally decomposed and issued to the terminal execution layer by layer from the cloud to the edge. In view of the problem that the static distance model cannot capture the difference of heterogeneous medium channel attenuation and the risk of dynamic routing congestion, the channel nonlinear attenuation penalty quantization and relay node dynamic queuing delay evaluation are introduced, and the multi-dimensional communication feature comprehensive cost fusion is carried out through the nonlinear fusion operator, so that the partition result truly reflects the channel quality difference and dynamic load risk in the heterogeneous communication environment, and the communication robustness and regulation and control response reliability in the high-frequency regulation and control scene are guaranteed.

Description

Technical Field

[0001] This application relates to the field of virtual power plant resource regulation technology, and more specifically, to a collaborative optimization configuration system based on a virtual power plant edge-cloud dual-layer structure. Background Technology

[0002] With the large-scale integration of distributed energy resources into the power system, virtual power plants, as key platforms for aggregating and regulating flexible resources, need to efficiently coordinate and optimize the allocation of massive heterogeneous resources. A single centralized architecture is difficult to meet the needs of real-time resource acquisition, rapid response, and global optimization. Building a collaborative optimization and allocation system with a two-layer structure of edge and cloud has become an important path to improve the regulation efficiency of virtual power plants.

[0003] In the existing hierarchical control system, the partition quality of the edge control layer directly determines the overall control performance. Accurate quantification of inter-node communication costs is the core of partition optimization. Existing solutions use the Euclidean distance between nodes multiplied by a unit cost coefficient and a fixed configuration cost as a measure of communication cost, implicitly assuming that communication cost is linearly proportional to physical distance and that link quality is homogeneous and static. However, virtual power plants are deployed in a complex electromagnetic environment with flexible resources, and the access communication media include heterogeneous types such as fiber optics, wireless cellular, and power line carriers. The channel attenuation patterns of different media are fundamentally different, and Euclidean distance alone cannot characterize the actual channel transmission quality. Furthermore, in actual communication networks, data needs to be forwarded hop-by-hop through multiple relay nodes. In high-frequency control scenarios, core relay nodes are prone to severe queuing delays due to traffic surges. When channel attenuation and routing congestion worsen simultaneously, a catastrophic retransmission storm effect can be triggered. Static distance models are insensitive to these multi-dimensional dynamic characteristics, leading to severe distortion of the communication cost matrix. This causes nodes with weak communication or congested paths to be incorrectly aggregated, threatening the stringent requirements of virtual power plants for millisecond-level control response time.

[0004] Therefore, an optimized collaborative configuration system based on a dual-layer structure of virtual power plant edge and cloud is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a collaborative optimization configuration system based on a virtual power plant edge-cloud dual-layer structure, comprising: The resource status acquisition module is used to collect the injected power, load power, and adjustable capability parameters including power capacity and ramp rate of each flexible resource terminal at each operating moment based on the topological connection relationship represented by the set of nodes and the set of branches in the power network to obtain a resource status dataset. The partition aggregation optimization module is used to perform partition aggregation optimization on the flexible resources in the resource status dataset to obtain an edge partitioning scheme while satisfying the constraint of the highest aggregation communication cost. The external characteristic identification module is used to identify the external characteristics and map the feasible region of flexible resources in the resource status dataset based on the edge partitioning scheme to obtain the adjustable feature set of the aggregate. The alliance game solving module is used to solve the dynamic alliance game of each aggregator by using the market fit index of adjustable power range, response time and duration of the adjustable features of the aggregator as the alliance utility function to obtain the cloud alliance structure. The instruction decomposition and distribution module is used to perform multi-level optimal decomposition and layer-by-layer distribution of global trading instructions based on the collaborative matching relationship between the cloud alliance structure and the received market trading instructions, so as to obtain the terminal control instruction set.

[0006] Compared with existing technologies, the collaborative optimization configuration system based on a dual-layer edge-cloud structure for virtual power plants provided in this application first collects the topology information and adjustable capability parameters of each flexible resource terminal and encapsulates them into a unified dataset. Then, at the edge control layer, it performs partitioning and aggregation with the improved modularity of communication cost fusion as the optimization objective. Next, it performs external characteristic equivalent deduction and power feasible domain boundary projection on each aggregate, and reports the aggregate characteristics to the cloud platform layer. The alliance game model drives the optimization to form the optimal alliance structure using the market fit index. Finally, the market transaction instructions are optimally decomposed layer by layer from the cloud to the edge and sent to the terminal devices for execution. To address the problem that the differences in channel attenuation of heterogeneous media and the risk of dynamic routing congestion in node communication cost quantification are difficult to capture by static distance models, the system introduces channel nonlinear attenuation penalty quantification and relay node dynamic queuing delay congestion cost assessment. It also uses a nonlinear fusion operator based on logarithmic smoothing to perform comprehensive cost fusion of multi-dimensional communication characteristics, so that the partitioning results can truly reflect the differences in channel quality and dynamic load risks in heterogeneous communication environments, ensuring the communication robustness and control response reliability of the virtual power plant in high-frequency control scenarios. Attached Figure Description

[0007] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1 This is a system block diagram of a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application; Figure 3This is a block diagram of the partition aggregation optimization module in the virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application; Figure 4 This is a block diagram of the communication cost quantification unit in the virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application; Figure 5 This is a block diagram of the alliance game solving module in the virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application; Figure 6 This is a block diagram of the instruction decomposition and distribution module in the virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application. Detailed Implementation

[0009] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0010] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0011] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0012] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0013] With the large-scale integration of distributed energy resources into the power system, virtual power plants (VPS), as key platforms for aggregating and regulating decentralized flexible resources, face severe challenges in their architecture design and resource allocation. Existing VPS mostly adopt a flat, centralized management and control model, which suffers from prominent problems when dealing with massive heterogeneous terminals, such as limited data collection dimensions, lack of communication economic constraints in partitioned aggregation, and a disconnect between cloud alliance formation and market demand. This results in the efficiency of decomposition and execution of control commands failing to meet the stringent millisecond-level response time requirements of high-frequency interaction scenarios such as peak shaving and frequency regulation. Simultaneously, the adjustable capability parameters of flexible resource terminals, such as power capacity and ramp rate, are not effectively bound to network topology information, resulting in a lack of accurate underlying data support for upper-level optimization decisions. The accuracy of external characteristic identification and feasible domain mapping after resource aggregation is insufficient, thus restricting the overall competitiveness and reliability of VPS in the power market. Therefore, this application proposes a collaborative optimization configuration system based on a dual-layer edge-cloud structure for VPS. At the terminal layer, based on the topological connectivity represented by the set of power network nodes and branches, the system collects the injected power, load power, and adjustable capability parameters including power capacity and ramp rate for each flexibility resource. After flexibility margin assessment and spatial attribute binding, a resource status dataset is generated. At the edge control layer, communication cost is incorporated as an economic penalty into the modularity evaluation index. Under the constraint of the highest aggregation communication cost, the system performs partitioned aggregation optimization of flexibility resources to obtain an edge partitioning scheme. Based on this scheme, external characteristic identification and feasible region mapping are performed on each aggregate to generate an adjustable feature set for the aggregate. At the cloud layer... At the platform layer, market fit indicators such as adjustable power range, response time, and duration are used as alliance utility functions to solve dynamic alliance game among various aggregates to form a cloud alliance structure. Finally, based on the collaborative matching relationship between the cloud alliance structure and market transaction instructions, global transaction instructions are decomposed and distributed layer by layer from the cloud platform layer to the edge control layer through multi-level optimal decomposition, generating a set of terminal control instructions that can directly drive terminal devices to execute physical responses. This enables the virtual power plant to achieve efficient collaborative optimization and precise control of distributed flexible resources while ensuring the economic efficiency of partitioned communication and the market competitiveness of aggregates.

[0014] Figure 1 This is a system block diagram of a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in a collaborative optimization configuration system based on a virtual power plant edge-cloud dual-layer structure, according to an embodiment of this application. Figure 1 and Figure 2As shown, the edge-cloud dual-layer collaborative optimization configuration system 100 based on an embodiment of this application includes: a resource status acquisition module 110, used to acquire the injected power, load power, and adjustable capability parameters including power capacity and ramp rate of each flexible resource terminal at each operating time based on the topological connection relationship represented by the node set and branch set in the power network to obtain a resource status dataset; a partition aggregation optimization module 120, used to perform partition aggregation optimization on the flexible resources in the resource status dataset under the constraint of satisfying the highest aggregation communication cost to obtain an edge partitioning scheme; and an external characteristic identification module 130, used for... The edge partitioning scheme identifies the external characteristics and maps the feasible domains of flexible resources in the resource status dataset to obtain the adjustable feature set of the aggregate; the alliance game solving module 140 is used to solve the dynamic alliance game for each aggregate based on the market fit index of adjustable power range, response time and duration in the adjustable feature set of the aggregate as the alliance utility function to obtain the cloud alliance structure; the instruction decomposition and distribution module 150 is used to perform multi-level optimal decomposition and layer-by-layer distribution of global transaction instructions from the cloud platform layer to the edge control layer based on the collaborative matching relationship between the cloud alliance structure and the received market transaction instructions to obtain the terminal control instruction set.

[0015] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the resource status acquisition module 110 is used to collect the injected power, load power, and adjustable capability parameters including power capacity and ramp rate of each flexible resource terminal at each operating moment, based on the topological connection relationship represented by the node set and branch set in the power network, to obtain a resource status dataset. It should be noted that since the upper-layer optimization decision of the virtual power plant relies on the accurate status perception of the underlying flexible resources, and the operating data of various distributed resource terminals exhibit heterogeneity in acquisition dimension, time granularity, and spatial affiliation, without unified topological association and feature binding, the edge control layer will be unable to obtain complete and structured information on resource adjustable capabilities. Therefore, the technical solution of this application first collects the injected power, load power, and adjustable capability parameters including power capacity and ramp rate of each flexible resource terminal at each operating moment, based on the topological connection relationship represented by the node set and branch set in the power network, to obtain a resource status dataset. Through the above processing, a structured data foundation with both topological spatial attributes and multidimensional adjustable features can be provided for the subsequent partition aggregation optimization of the edge control layer.

[0016] More specifically, in a concrete example of this application, the generation process of the resource status dataset includes four progressive steps: topology instantiation, net power calculation, margin assessment binding, and serialization encapsulation. In the topology instantiation step, the node set and branch set in the power network are first read. The node set includes the physical connection locations of various flexible resource terminals such as distributed photovoltaic inverter access points, energy storage converter grid connection points, and electric vehicle charging pile metering points. The branch set records the electrical connection paths between the aforementioned nodes, such as 10kV feeder segments and 0.4kV low-voltage lines. Based on the connection status between each node and branch, the spatial connectivity relationship of the node set and branch set is instantiated, transforming the discrete node numbers and branch numbers into network topology physical information with adjacency relationship description capabilities. This information is stored in the form of an adjacency matrix or adjacency list data structure, fully representing the spatial topological location of each flexible resource terminal in the distribution network and their electrical connection paths.

[0017] In the net power calculation stage, based on the difference between injected power and load power at each operating moment, the net power difference is calculated and time-series spliced ​​for each node. For each node, within each discrete control time step, the total active power injected by the node is subtracted from the total active load power to obtain the real-time net power value of the node at that moment. Then, the net power calculation results of each time step are spliced ​​in chronological order to form a real-time net power sequence of the node covering the complete scheduling cycle. Taking a photovoltaic inverter access node as an example, if the photovoltaic injected power of the node is 50kW and the local load power is 20kW at noon, then the net power at that moment is 30kW, indicating that the node is in the state of feeding power back to the grid at that moment.

[0018] In the margin assessment binding phase, for each flexibility resource terminal node, its upward flexibility adjustment margin needs to consider two physical constraints simultaneously: first, the static capacity difference between the upper limit of the equipment's rated power capacity and the current net power; and second, the maximum power increment allowed within a single control time step due to ramp rate limitations. This margin takes the smaller value of the two constraints, and its calculation expression is as follows:

[0019] in, For the first Each terminal node The upward flexibility adjustment margin at any time, For the first The upper limit of the rated active power capacity of each terminal node. For the first Each node Real-time net power at any given moment For the first The upper limit of the ramp rate for each terminal device. This represents the discrete control time step of the system. Taking an energy storage converter with a rated power capacity of 100kW and an upward ramp rate of 20kW / min as an example, if the net power at the current moment is 60kW and the control time step is 1min, then the capacity difference is 40kW, the ramp allowable increment is 20kW, and the upward flexibility adjustment margin of this terminal at that moment is 20kW. The calculated flexibility margin characteristics of each terminal at each moment are mapped and bound to the inherent parameters of the equipment one-to-one according to the node number index and the spatial location in the network topology physical information, generating multi-dimensional feature binding data for the terminals.

[0020] In the serialization and encapsulation stage, based on the preset lightweight communication protocol data dictionary standard, the terminal's multi-dimensional feature-bound data undergoes unified structure transformation and serialization compression. The topology location field, net power timing field, adjustable capability parameter field, and flexibility margin field in the aforementioned bound data are uniformly transformed according to the predefined data dictionary key-value pair format. Then, binary serialization compression processing is performed on the transformed structured data to generate a resource status dataset, which is released to the communication channel for use by the edge control layer.

[0021] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the partition aggregation optimization module 120 is used to perform partition aggregation optimization on the flexible resources in the resource status dataset under the constraint of the highest aggregation communication cost to obtain an edge partitioning scheme. It should be noted that, given that the flexible resources of the virtual power plant aggregation are spatially dispersed across different distribution network feeders and transformer power supply ranges, if partitioning is performed solely based on electrical topology density while ignoring the communication cost between the edge controller and each terminal, the communication links between flexible resources within the same aggregation will be too long or the communication cost too high. In high-frequency interaction scenarios such as peak shaving and frequency regulation, this can lead to transmission delays or even communication paralysis when the edge controller issues dense control commands, directly weakening the aggregation's ability to guarantee millisecond-level control response timeliness. Furthermore, excessively high communication infrastructure configuration costs will erode the economic benefits of the virtual power plant participating in the electricity market. Therefore, the technical solution of this application further optimizes the flexible resources in the resource status dataset by partition aggregation under the constraint of the highest aggregation communication cost to obtain an edge partitioning scheme. Through the above processing, while ensuring the tightness of the electrical topology within each aggregate, communication economy can be incorporated into the partitioning decision. This allows the generated edge partitioning scheme to balance topology connectivity quality and controllable communication costs, providing a reasonably structured edge control layer infrastructure for subsequent external characteristic identification of each aggregate and cloud alliance game.

[0022] Figure 3This is a block diagram of the partition aggregation optimization module in a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application. Figure 3 As shown, the partitioning aggregation optimization module 120 includes: a topology graph mapping unit 121, used to perform topology graph mapping and parameter analysis on the resource state dataset based on the node set and branch set in the resource state dataset to obtain a physical topology undirected graph; a communication cost quantification unit 122, used to perform node communication cost quantification on any two nodes in the physical topology undirected graph based on the spatial geographic coordinates of each node and a preset unit distance communication cost coefficient to obtain a weighted graph model and a node communication cost matrix; a modularity evaluation construction unit 123, used to construct a modularity evaluation by fusing the communication cost of the adjacency matrix and the edge degree of the nodes in the weighted graph model with the node communication cost matrix as an economic penalty term to obtain an improved modularity objective function and an aggregation constraint condition matrix; and a partitioning optimization solution unit 124, used to perform heuristic graph theory partitioning optimization solution in the constraint space based on the maximum aggregation communication cost limit set by the aggregation constraint condition matrix to obtain an edge partitioning scheme.

[0023] In the aforementioned collaborative optimization configuration system 100 based on a dual-layer edge-cloud structure of a virtual power plant, the topology graph mapping unit 121 is used to perform topology graph mapping and parameter parsing on the resource state dataset based on the node set and branch set in the resource state dataset to obtain a physical topological undirected graph. It should be noted that since the resource state dataset is stored in a serialized compressed binary format, the node set and branch set information is still in a flattened data field state, which cannot directly support the spatial adjacency query and node distance traversal operations required for subsequent communication cost quantification calculations. Based on this, the technical solution of this application further performs topology graph mapping and parameter parsing on the resource state dataset based on the node set and branch set in the resource state dataset to obtain a physical topological undirected graph. Through the above processing, the flattened data fields can be transformed into a graph theory data structure with spatial connectivity expression capabilities, providing a traversable topological carrier for the accurate quantification of communication costs between nodes.

[0024] More specifically, in a concrete example of this application, the received resource status dataset is first deserialized and decompressed to restore the original fields of the encapsulated node set and branch set. The node set contains the ID index of each flexible resource terminal and its corresponding spatial geographic coordinates, while the branch set records the electrical connection path numbers between each node and their first and last node identifiers. Based on this, using the terminal numbers in the node set as vertices in the graph theory and the electrical connection paths in the branch set as edges, the above correspondence between vertices and edges is written into an adjacency matrix or adjacency list data structure, completing the mapping transformation from linear data fields to graph theory spatial connectivity relationships. During the mapping process, since the electrical connection paths between flexible resource terminals in the distribution network are not directional (i.e., the 10kV feeder segment between node A and node B allows power to flow from A to B and from B to A), the constructed graph structure is an undirected graph. Simultaneously, the power capacity, ramp rate, net power timing, and flexibility margin parameters carried by each node in the resource status dataset are parsed and extracted, and then attached as vertex attributes to the corresponding graph nodes, ultimately generating a physical topological undirected graph that combines spatial topological connectivity and multidimensional parameter attributes of nodes.

[0025] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the communication cost quantification unit 122 is used to quantify the node communication cost of any two nodes in the physical topology undirected graph based on the spatial geographic coordinates of each node and a preset unit distance communication cost coefficient to obtain a weighted graph model and a node communication cost matrix. It should be noted that, given that the physical topology undirected graph only represents the electrical connectivity between flexible resource terminals and has not yet quantified the communication economic cost incurred when deploying edge controllers for data interaction between nodes, in the hierarchical and partitioned control architecture of the virtual power plant, the communication cost between terminals and edge controllers within the same aggregate directly affects the timeliness of control command transmission and the economic feasibility of aggregated operation. Without accurate measurement of the inter-node communication cost, subsequent partitioned aggregation optimization will be unable to effectively balance topology compactness and communication economy. Therefore, the technical solution of this application further quantifies the node communication cost of any two nodes in the physical topology undirected graph based on the spatial geographic coordinates of each node and a preset unit distance communication cost coefficient to obtain a weighted graph model and a node communication cost matrix. Through the above processing, the economic cost of communication between each pair of nodes can be bound to the topology in the form of numerical edge weights. This provides a weighted graph carrier that combines topological attributes and communication cost attributes, as well as a cost measurement basis that can directly participate in matrix operations, for the subsequent construction of improved modularity indicators.

[0026] Figure 4This is a block diagram of the communication cost quantification unit in a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application. Figure 4 As shown, the communication cost quantization unit 122 includes: a channel attenuation quantization subunit 1221, used to perform channel nonlinear attenuation penalty quantization on any two nodes based on the spatial geographic coordinates of each node in the physical topology undirected graph and the media reference attenuation rate and environmental noise coefficient in the heterogeneous media attribute set to obtain a channel attenuation penalty matrix; a routing congestion assessment subunit 1222, used to accumulate the hop count overhead and dynamic queuing delay congestion cost of each relay node on the routing path based on the shortest path search result of the physical topology undirected graph and the real-time traffic throughput of each relay node to obtain a dynamic routing congestion matrix; and a comprehensive cost fusion subunit 1223, used to perform comprehensive cost fusion calculation and edge weight binding on multi-dimensional communication features based on the channel attenuation penalty matrix, the dynamic routing congestion matrix, and the fixed configuration cost of the edge controller to obtain a weighted graph model and a communication cost matrix between nodes.

[0027] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the channel attenuation quantization subunit 1221 is used to perform channel nonlinear attenuation penalty quantization on any two nodes based on the spatial geographic coordinates of each node in the physical topology undirected graph and the media reference attenuation rate and environmental noise coefficient in the heterogeneous media attribute set to obtain the channel attenuation penalty matrix. It should be noted that, given that the aforementioned node communication cost quantification only uses the Euclidean distance between any two nodes multiplied by the unit distance communication cost coefficient and superimposed with the fixed configuration cost of the edge controller as the measurement basis for communication cost, this method implicitly assumes that the communication cost is strictly linearly proportional to the physical straight-line distance. However, the communication media on which the flexible resource terminals of the virtual power plant aggregate access the edge controller are highly heterogeneous, covering various types such as fiber optic leased lines, 5G cellular wireless networks, and power line carriers. The channel attenuation laws of different media are fundamentally different. Euclidean distance alone cannot distinguish the channel quality differences caused by media heterogeneity and environmental interference, resulting in a deviation between the generated communication cost matrix and the actual channel transmission quality. Based on this, the technical solution of this application further quantifies the channel attenuation penalty for any two nodes by using the spatial geographic coordinates of each node in the physical topology undirected graph and the media reference attenuation rate and environmental noise coefficient in the heterogeneous media attribute set, in order to obtain the channel attenuation penalty matrix. Through the above processing, the nonlinear attenuation characteristics of heterogeneous communication media under different transmission distances and electromagnetic environments can be accurately characterized from the physical channel layer, avoiding the erroneous aggregation of vulnerable nodes in the same partition.

[0028] More specifically, in a particular example of this application, since the communication media types of the flexible resource terminals accessing the edge controller in the virtual power plant scenario are different, and the signal attenuation patterns of different media exhibit drastically different nonlinear characteristics under different transmission distances and electromagnetic environments, relying solely on Euclidean distance cannot characterize the real channel transmission quality. The signal attenuation of fiber optic links has a weak, approximately linear correlation with transmission distance, while wireless cellular signals exhibit exponential nonlinear attenuation characteristics when penetrating building walls and metal obstructions. Power line carrier communication is strongly interfered with by the harmonic noise of the distribution network itself and load fluctuations. Therefore, it is necessary to extract the spatial geographic coordinates of any two nodes from the undirected graph of the physical topology to calculate the Euclidean distance, and simultaneously obtain the media reference attenuation rate corresponding to the communication link between the two nodes and the electromagnetic noise coefficient of the environment from a pre-stored set of heterogeneous media attributes. Based on this, a heterogeneous channel generalized attenuation function is constructed to quantify the real physical channel penalty value between any pair of nodes, and its expression is:

[0029] in, For nodes With nodes Channel attenuation penalty value between The reference attenuation rate is determined by the medium type of the communication link between the two nodes. For nodes With nodes The physical Euclidean distance between them The spatial propagation loss coefficient for a specific medium. This is the path loss exponent, which takes a value of 2 in line-of-sight transmission scenarios and a value of 3 to 4 in complex urban areas or industrial plant obstruction scenarios. The ambient electromagnetic noise figure between the two nodes. Let be the spatial angle between the electromagnetic noise interference source signal and the direct path of communication between the two nodes. In the above formula, It partially and accurately fits the nonlinear exponential attenuation characteristics of wireless signals and power line carriers when penetrating obstacles such as buildings and metal shields. As the transmission distance increases or the path loss exponential increases, the attenuation penalty value will climb at an exponential rate, rather than just increasing linearly. The system also incorporates the interference effect of the spatial angle between electromagnetic noise and communication signals. This allows for adaptively applying high attenuation penalties even when two flexible resource terminals are physically close in harsh electromagnetic environments, provided their communication paths are within areas of strong electromagnetic interference. This fundamentally prevents the erroneous aggregation of vulnerable nodes within the same partition. After the above calculations, the channel attenuation penalty values ​​for all node pairs are assembled into a channel attenuation penalty matrix, which serves as the first dimension input for subsequent fusion calculations.

[0030] Taking a flexible resource deployment scenario within an industrial park as an example, the physical Euclidean distance between a distributed photovoltaic inverter access node A and a grid-connected energy storage converter node B within the park is only 30m. However, a large metal factory building obstructs the communication between the two nodes. Their communication link uses a 5G cellular wireless network, resulting in a path loss index of 3.5 due to the industrial building obstruction. The ambient electromagnetic noise figure is also relatively high, leading to a channel attenuation penalty of approximately 1.292 calculated using the heterogeneous channel generalized attenuation function. Conversely, the physical Euclidean distance between an electric vehicle charging station metering node C and node A outside the park is 200m. However, the communication link between the two nodes uses a dedicated fiber optic line with no electromagnetic obstruction. The corresponding dielectric reference attenuation rate and ambient noise figure are both low, resulting in a calculated channel attenuation penalty of approximately 0.152, significantly lower than the attenuation penalty for nodes with a shorter physical distance between A and B. The results show that under heterogeneous communication media and complex electromagnetic environment conditions, nodes that are physically closer may actually suffer higher channel attenuation penalties, verifying the necessity of introducing a nonlinear attenuation quantization mechanism to realistically characterize the differences in communication channel quality in virtual power plant scenarios.

[0031] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the routing congestion assessment subunit 1222 is used to accumulate the congestion cost of hop count and dynamic queuing delay for each relay node on the routing path based on the shortest path search result of the physical topology undirected graph and the real-time traffic throughput of each relay node, thereby obtaining a dynamic routing congestion matrix. It should be noted that, given that the aforementioned channel attenuation penalty matrix only characterizes the nonlinear attenuation characteristics of heterogeneous media from the physical channel layer, and completely ignores the dynamic characteristics of the communication network topology layer, in actual edge communication networks, data transmission between two terminal nodes is not a direct point-to-point connection, but requires hop-by-hop forwarding through multiple relay routing nodes. Furthermore, when the virtual power plant is in high-frequency interaction scenarios such as peak shaving or frequency modulation, the traffic throughput of the core relay node will increase sharply, resulting in severe queuing delay or even packet loss. Static distance metrics are completely insensitive to this dynamic congestion phenomenon. Based on this, the technical solution of this application further calculates a dynamic routing congestion matrix by accumulating the hop count overhead and dynamic queuing delay congestion costs for each relay node on the routing path, based on the shortest path search results of the physical topology undirected graph and the real-time traffic throughput of each relay node. Through the above processing, dynamic load risks on communication paths can be captured from the network topology layer, preventing nodes with severe bottlenecks on communication paths from being assigned to the same aggregate, and ensuring that the partitioning results still have sufficient communication margin under dynamic load conditions.

[0032] More specifically, in a concrete example of this application, considering that in high-frequency interaction scenarios such as peak shaving and frequency regulation of virtual power plants, data transmission in the edge communication network is not a direct point-to-point connection, but requires hop-by-hop forwarding through multiple relay routing nodes. Furthermore, core relay nodes experience severe queuing delays due to traffic surges during periods of intensive control command issuance. Static distance metrics cannot capture this dynamic congestion risk. Therefore, it is necessary to determine the optimal communication route between any two nodes and all relay nodes along the path using a shortest path search algorithm based on the physical topology undirected graph, while simultaneously collecting the real-time traffic throughput of each relay node on the path at the current moment. Combining the classic model in queuing theory, the comprehensive congestion cost, including fixed hop count overhead and dynamic queuing delay, is calculated, and its expression is:

[0033] in, For nodes With nodes Dynamic routing congestion costs between them For nodes With nodes The set of all relay nodes included in the optimal routing path between them. For a specific relay node on the path. The fixed base hop count cost for a single route forwarding. The hardware baseline processing latency for relay nodes, For relay nodes In the present Actual throughput at any given moment For relay nodes The maximum communication bandwidth capacity of the hardware. In the above formula, An asymptotic congestion barrier structure was constructed. When the real-time throughput of a relay node approaches its hardware bandwidth limit, the denominator approaches zero, and the congestion cost approaches infinity, thus forming an insurmountable cost barrier at the mathematical level. This characteristic effectively prevents the risk of communication paralysis caused by a large influx of control commands into the communication path of the same edge controller under extreme control scenarios in virtual power plants, ensuring that the partitioning results still have sufficient communication margin under dynamic load conditions. After the above calculations, the routing congestion costs of all node pairs are assembled into a dynamic routing congestion matrix, which serves as the second dimension input for subsequent fusion calculations.

[0034] Taking the flexible resource deployment scenario in the aforementioned industrial park as an example, although the physical distance between distributed photovoltaic inverter access node A and energy storage converter grid-connected node B is only 30m, the two nodes belong to different subnet segments in terms of communication topology. Their data packets need to cross 3 routing switching nodes to reach the other end. During the peak photovoltaic output period at noon, the edge controller needs to send power adjustment commands to a large number of terminals and collect status feedback data in a very short time window. At this time, the real-time traffic throughput of the second relay node on the path has reached 92% of its maximum hardware communication bandwidth capacity. According to the above formula, the congestion cost contributed by this relay node increases sharply, making the dynamic routing congestion cost between node A and node B much higher than the congestion cost between node A, which is 200m away but whose communication path only passes through one low-load relay node, and electric vehicle charging pile metering node C. The results indicate that two physically close nodes may face severe routing bottlenecks in the communication topology. If they are classified into the same aggregate, the aggregate will lose its responsiveness due to communication paralysis during actual control execution, directly threatening the stringent requirements of the virtual power plant for millisecond-level control response timeliness.

[0035] In the aforementioned collaborative optimization configuration system 100 based on a dual-layer edge-cloud structure of a virtual power plant, the comprehensive cost fusion subunit 1223 is used to perform comprehensive cost fusion calculation and edge weight binding on multi-dimensional communication characteristics based on the channel attenuation penalty matrix, the dynamic routing congestion matrix, and the fixed configuration cost of the edge controller to obtain a weighted graph model and an inter-node communication cost matrix. It should be noted that, given that the channel attenuation penalty matrix and the dynamic routing congestion matrix characterize communication costs from two heterogeneous dimensions—the physical channel layer and the network topology layer—their dimensions, numerical ranges, and distribution characteristics differ, a simple linear weighted summation method for fusion can easily lead to a bias where extreme values ​​in one dimension dominate the overall cost calculation. Furthermore, when a communication link simultaneously faces severe channel attenuation and increased routing congestion, the two are not independently superimposed but can trigger a catastrophic retransmission storm effect. Therefore, the technical solution of this application further performs comprehensive cost fusion calculation and edge weight binding on multi-dimensional communication characteristics based on the channel attenuation penalty matrix, the dynamic routing congestion matrix, and the fixed configuration cost of the edge controller to obtain a weighted graph model and an inter-node communication cost matrix. Through the above processing, the avalanche effect when channel attenuation and routing congestion worsen simultaneously can be keenly captured by the nonlinear fusion operator, and the mispairing of nodes on high-risk communication links can be blocked during the partition aggregation stage.

[0036] More specifically, in a concrete example of this application, since the channel attenuation penalty matrix and the dynamic routing congestion matrix characterize communication costs from two heterogeneous dimensions—the physical channel layer and the network topology layer, respectively—their dimensions, numerical ranges, and distribution characteristics differ. If a simple linear weighted summation method is used for fusion, it is highly likely that extreme values ​​in one dimension will dominate the overall cost calculation, leading to a bias. In the actual operation of a virtual power plant, when a communication link simultaneously faces severe channel attenuation and increased routing congestion, these two conditions do not independently accumulate but trigger a catastrophic retransmission storm effect. That is, channel attenuation leads to an increased packet transmission failure rate, failed packets trigger an automatic retransmission mechanism, and retransmitted packets further exacerbate the congestion of routing nodes. Congestion, in turn, increases the probability of transmission timeout, forming a positive feedback vicious cycle. Therefore, it is necessary to extract the attenuation penalty value of each node pair in the channel attenuation penalty matrix and the congestion cost value of each node pair in the dynamic routing congestion matrix, and combine this with the fixed configuration cost of the edge controller in the communication cost parameters to be processed. A nonlinear fusion operator based on logarithmic smoothing is then constructed for comprehensive cost calculation, and its expression is:

[0037] in, For the final generated nodes With nodes The overall communication cost between them Fixed configuration cost for edge controllers, This is the global communication cost benchmark adjustment coefficient. It is the natural logarithm function. The sensitivity weight parameter for channel attenuation penalty, For nodes in the channel attenuation penalty matrix With nodes The corresponding decay penalty value, For the sensitivity weight parameter of dynamic routing congestion cost, For nodes in the dynamic routing congestion matrix With nodes The corresponding congestion generation value, This represents the amplification factor for catastrophic retransmission storms. In the above formula, the introduction of the natural logarithm function effectively suppresses the excessive impact of outliers on the overall cost while preserving the discriminative power of features across different dimensions, achieving a smooth fusion of multi-dimensional features. Cross-product term. This is the core difference between this fusion operator and the traditional linear weighting method. When channel attenuation and routing congestion are both high, the value of the cross term will be amplified sharply at a product rate. After being mapped by a logarithmic function, it can still maintain the excess penalty effect, thus keenly capturing and quantifying the avalanche effect of the retransmission storm, and blocking the mispairing of nodes on high-risk communication links during the partitioning aggregation stage. The comprehensive communication cost of all node pairs calculated is assembled into an inter-node communication cost matrix, and this matrix is ​​used as an edge weight attribute to be inversely bound to the physical topology undirected graph, generating a weighted graph model that combines topological structure and multi-dimensional communication cost attributes.

[0038] Taking the flexible resource deployment scenario in the aforementioned industrial park as an example, between distributed photovoltaic inverter access node A and energy storage converter grid-connected node B, the channel attenuation penalty value has been obtained from the aforementioned calculation. The congestion cost for dynamic routing is approximately 1.292. The second relay node on the path experiences high throughput, reaching 92% of its hardware bandwidth capacity during the peak midday solar power output period. In the overall cost fusion calculation, due to... and Simultaneously at a high level, cross-product term This results in an excess penalty far exceeding the simple linear superposition of the two, leading to the combined communication cost between node A and node B after logarithmic function mapping. It was assigned an extremely high value. The channel attenuation penalty value between electric vehicle charging pile metering node C and node A... The value is only 0.152 and the load on the relay nodes along the routing path is low, making it possible to achieve this. Low position, cross-product term The value is extremely small, and the overall communication cost is low. far below The results show that although the physical distance between nodes A and B is only 30m, their combined communication cost is much higher than that between nodes A and C, which are 200m apart, due to the dual deterioration of severe channel attenuation and increased routing congestion. If nodes A and B are grouped into the same aggregate, a retransmission storm will be triggered during the dense issuance of control commands in high-frequency interaction scenarios such as peak shaving and frequency modulation, leading to communication paralysis. The weighted graph model and inter-node communication cost matrix generated based on the above nonlinear fusion mechanism can effectively avoid such high-risk pairings in subsequent partitioning optimization.

[0039] In the aforementioned collaborative optimization configuration system 100 based on a dual-layer edge-cloud structure of a virtual power plant, the modularity evaluation construction unit 123 is used to construct a modularity evaluation by fusing the communication cost matrix between nodes as an economic penalty term and the communication cost of the adjacency matrix and the degree of node connections in the weighted graph model, thereby obtaining an improved modularity objective function and an aggregation constraint matrix. It should be noted that although the weighted graph model and the communication cost matrix between nodes have fully quantified the comprehensive communication cost between each pair of nodes, this cost information has not yet been incorporated into the optimization objective function used to drive partitioning decisions. Traditional modularity indices only measure topological connectivity without considering communication economics; directly partitioning using traditional modularity may generate aggregates with excessively high internal communication costs. Based on this, the technical solution of this application further uses the communication cost matrix between nodes as an economic penalty term to construct a modularity evaluation by fusing the communication cost of the adjacency matrix and the degree of node connections in the weighted graph model, thereby obtaining an improved modularity objective function and an aggregation constraint matrix. Through the above processing, topological compactness and communication economy can be integrated into the same optimization objective framework, providing an objective function and hard constraint boundary that take into account both indicators for subsequent constraint optimization solutions.

[0040] More specifically, in a concrete example of this application, the process of constructing the improved modularity objective function and the aggregation constraint matrix includes three progressive steps: constructing the basic modularity expression, fusing communication cost penalty terms, and generating the aggregation constraint matrix. First, the adjacency matrix and the edge degree of each node are extracted from the weighted graph model, and a basic modularity expression representing the tightness of electrical connectivity is constructed by combining it with the community affiliation indicator variable. The core logic of this expression is: for any two nodes assigned to the same region aggregate, if there is an actual electrical connection edge between them, the node's contribution to the modularity is positive, equal to the actual connection weight minus the expected connection weight under the random connection assumption; if there is no actual connection edge between them, the node's contribution to the modularity is negative, representing the topology quality loss caused by forcibly assigning electrically distant nodes to the same region.

[0041] Based on this, an inter-node communication cost matrix is ​​introduced as an economic penalty term, which is then incorporated into the basic modularity expression through a deduction method. For any two nodes assigned to the same region aggregate, the higher their combined communication cost, the greater their negative contribution to improving modularity. This automatically suppresses the tendency to assign nodes with high communication costs to the same aggregate during the optimization process. The objective function expression for the improved modularity after integration is as follows:

[0042] in, To improve the objective function value of modularity, The total number of edges in the weighted graph model. The total set of nodes in the physical network. For nodes in a weighted graph model With nodes The adjacency matrix elements are set to 1 if the elements are connected and 0 otherwise. and They are nodes and nodes The degree of undirected connected edges, For community affiliation indicator function, when node and nodes The value is 1 if the elements are assigned to the same region aggregate, and 0 otherwise. The preset communication cost penalty weighting coefficient, For nodes in the inter-node communication cost matrix With nodes The corresponding comprehensive communication cost value. In the above formula, the first part... It inherits the ability of classic modularity metrics to evaluate the tightness of topological connectivity, and the second half... This introduces the overall communication cost as a penalty term, so that the optimization process is constrained by the economics of communication while pursuing topological compactness.

[0043] Simultaneously, based on the inter-node communication cost matrix, a mathematical equation matrix representing the total communication cost within any single aggregate is constructed. For each candidate region aggregate... The sum of communication costs among all flexible resources within the partition scheme must not exceed a preset maximum aggregated communication cost limit. This constraint is expressed in a structured form as an inequality matrix, generating an aggregated constraint condition matrix. This matrix is ​​injected into the solver as a hard inequality boundary during subsequent constraint optimization, ensuring that the internal communication cost of any generated candidate partition scheme strictly does not exceed the limit, thereby guaranteeing the communication feasibility and economic rationality of the edge partition scheme in actual deployment.

[0044] In the aforementioned collaborative optimization configuration system 100 based on a dual-layer edge-cloud structure of a virtual power plant, the partitioning optimization solution unit 124 is used to perform heuristic graph theory partitioning optimization within the constraint space to obtain an edge partitioning scheme based on the maximum aggregation communication cost limit set by the aggregation constraint matrix. It should be noted that since the optimization direction and feasible region boundary have been defined for the improved modularity objective function and the aggregation constraint matrix, but the actual solution operation has not yet been performed to generate specific partitioning results, and this optimization problem belongs to the category of combinatorial optimization, its solution space grows exponentially with the number of nodes, making precise solution computationally infeasible. Therefore, the technical solution of this application further performs heuristic graph theory partitioning optimization within the constraint space to obtain an edge partitioning scheme based on the maximum aggregation communication cost limit set by the aggregation constraint matrix. Through the above processing, near-optimal partitioning results that balance topological compactness and communication economy can be obtained within an acceptable computation time, providing a structurally reasonable edge control layer architecture for subsequent external characteristic identification and cloud alliance game theory.

[0045] More specifically, in a concrete example of this application, the solution process for the edge partitioning scheme includes three progressive steps: solver initialization configuration, heuristic iterative optimization, and constraint verification and scheme output. First, the objective function for improving modularity is set as a global optimization maximization objective in the solver. Simultaneously, an aggregation constraint matrix is ​​injected as a hard inequality boundary of the solution space, ensuring that the sum of the internal communication costs of each aggregate in any candidate partitioning scheme generated during the iteration process does not exceed a preset maximum aggregation communication cost limit threshold. The mathematical expression of this constraint relationship is: , in To improve the objective function value of modularity, This represents the combined communication cost of the corresponding node pairs in the inter-node communication cost matrix. For community affiliation indicator function, This is the preset maximum aggregated communication cost limit threshold.

[0046] Based on this, a heuristic graph theory partitioning algorithm is driven to iteratively optimize within the aforementioned constraint space. Specifically, an improved Louvain community detection algorithm is used as the core solution engine. This algorithm takes a weighted graph model as input, initially treating each node as an independent community. Then, in each iteration, it traverses each node sequentially, evaluating the gain change in the objective function value of improving modularity after moving the current node out of its own community and merging it into an adjacent community. If merging a node into an adjacent community results in a positive gain in modularity and the sum of internal communication costs within the merged community still meets the maximum aggregated communication cost limit, then the community migration operation for that node is executed. If, although the modularity gain is positive after merging, the internal communication cost exceeds the limit threshold, the migration operation is rejected to ensure constraint feasibility. When no node migration operation exists in a complete traversal that can improve modularity under the constraints, the algorithm compresses the current communities into super nodes, constructs a new compressed graph using the aggregated connection weights between super nodes, and repeats the above iterative process on the compressed graph. This multi-level compression and iteration mechanism continues to execute until the change in the objective function value of the improved modularity in consecutive iterations is lower than the preset convergence accuracy threshold. The algorithm then terminates and outputs the node cluster partitioning variables in the converged state.

[0047] Finally, the converged node cluster partitioning variables are constrained and verified, checking whether the sum of the internal communication costs of each candidate aggregate strictly satisfies the inequality boundaries set in the aggregation constraint matrix. For all aggregates that pass the verification, their node affiliation mapping relationship is instantiated and encapsulated to generate an edge partitioning scheme. This scheme explicitly records the regional aggregate number to which each flexible resource terminal node belongs, as well as the node set composition within each regional aggregate, serving as the input basis for subsequent external characteristic identification and feasible region mapping. Taking the aforementioned industrial park scenario as an example, after the above constraint optimization solution, distributed photovoltaic inverter access node A and electric vehicle charging pile metering node C are assigned to the same regional aggregate due to their lower overall communication costs and close topological connectivity. However, energy storage converter grid-connected node B is assigned to another regional aggregate due to its excessively high overall communication costs with node A and is managed by an independent edge controller. This avoids the risk of aggregates losing responsiveness due to communication bottlenecks in high-frequency interaction scenarios such as peak shaving and frequency regulation.

[0048] In the aforementioned collaborative optimization configuration system 100 based on a virtual power plant edge-cloud dual-layer structure, the external characteristic identification module 130 is used to identify the external characteristics and map the feasible domains of flexible resources in the resource status dataset based on the edge partitioning scheme to obtain an aggregate adjustable feature set. It should be noted that since the edge partitioning scheme only determines the regional affiliation of each flexible resource terminal, it has not yet identified and quantified the external characteristic parameters such as the power adjustable range, ramping capability, and duration when each regional aggregate is treated as a whole unit. However, the cloud platform layer needs to use the aggregate-level adjustable features as the basis for decision-making when conducting alliance games and market transaction matching. Based on this, the technical solution of this application further identifies the external characteristics and maps the feasible domains of flexible resources in the resource status dataset based on the edge partitioning scheme to obtain an aggregate adjustable feature set. Through the above processing, the dispersed bottom-level individual device parameters can be aggregated into equivalent external characteristics and power feasible domain boundaries at the aggregate level, providing a structured adjustable capability input for the calculation of market fit indicators in subsequent cloud alliance games.

[0049] More specifically, in a concrete example of this application, the generation process of the aggregate adjustable feature set includes three progressive steps: region placement, equivalent aggregation deduction, and feasible domain boundary fusion. In the region placement step, based on the node affiliation mapping relationship of each region in the edge partitioning scheme, the flexible resources in the resource status dataset are accurately delivered and placed by region. Specifically, the correspondence between the aggregate number of each region and its internal node set recorded in the edge partitioning scheme is read, the status data records of all flexible resource terminals in the resource status dataset are traversed, and the data is accurately delivered to the region aggregate data container according to the node number index of each terminal. Taking the aforementioned industrial park scenario as an example, the status data of distributed photovoltaic inverter access node A and electric vehicle charging pile metering node C are placed in the same region aggregate container, while the status data of energy storage converter grid-connected node B is placed in another region aggregate container. The data in all containers are instantiated and encapsulated to generate a region clustered physical status set that has both regional isolation attributes and physical operating status characteristics.

[0050] In the equivalent aggregation derivation stage, the power capacity and ramp rate of each individual device within the same aggregation body are equivalently aggregated and derived. For each aggregation body, all individual devices within it are considered as a generalized equivalent model. The upper limit of the rated active power capacity of each individual device is linearly superimposed and summed to obtain the upper limit of the maximum active power capacity of the aggregation body. The upper limit of the ramp rate of each individual device is linearly superimposed and summed to obtain the overall ramp rate of the aggregation body. Taking an aggregation body containing nodes A and C as an example, if the rated power capacity of the photovoltaic inverter at node A is 100kW and the ramp rate is 30kW / min, and the rated power capacity of the electric vehicle charging pile at node C is 60kW and the ramp rate is 15kW / min, then the upper limit of the maximum active power capacity of the aggregation body is 160kW, and the overall ramp rate is 45kW / min. The macroscopic equivalent values ​​obtained by the above superposition calculation are encapsulated as equivalent parameters of the aggregation external characteristics. Simultaneously, the original upper and lower limit constraints and ramp constraints of each individual device at the bottom layer are extracted and packaged independently into a set of node-level constraints.

[0051] In the feasible region boundary fusion stage, the equivalent parameters of aggregate external characteristics and the set of node-level constraints are fused by projection of the feasible region boundary of the power domain at the grid connection point. Using a high-dimensional polyhedral projection algorithm, with the grid connection point as the reference anchor point, the node-level local constraints and the equivalent characteristic parameters of the aggregate are projected and their spatial intersection is calculated into a low-dimensional space, constructing a set of dynamic power feasible region boundary constraint equations that characterize the current aggregate during the time-series scheduling process. The expression for this feasible region set is:

[0052] in, For the first The set of power feasible regions for each aggregate. For the first Regional aggregates in The planned value of active power output at any given time. For the first Minimum active power capacity limit for each regional aggregate For the first The maximum active power capacity limit of each regional aggregate. For the first The overall upward climbing rate of the aggregate in each region. Let be the discrete control time step of the system. Geometrically, this feasible region is represented as a low-dimensional polyhedron defined by power capacity constraints and ramping constraints. The boundary parameters of each face represent the upper and lower limits of the adjustable power of the aggregate at each time step and the maximum power change between adjacent time steps. The constraint equations are serialized into a high-dimensional space vector format to generate a multi-dimensional feature boundary matrix of the aggregate as the adjustable feature set of the aggregate. After encryption, compression, and encapsulation, this matrix is ​​reported across layers to the virtual power plant cloud platform layer.

[0053] In the aforementioned collaborative optimization configuration system 100 based on a virtual power plant edge-cloud dual-layer structure, the alliance game solving module 140 is used to perform dynamic alliance game solving on each aggregator based on the market fit index of the adjustable power range, response time, and duration of the aggregate's adjustable features as the alliance utility function, so as to obtain the cloud alliance structure. It should be noted that, given that the edge control layer has completed the partitioning and aggregation of flexible resources and generated adjustable feature sets for each aggregate, but each aggregate still exists as an independent entity, it has not yet formed a collaborative organizational structure capable of participating in grid dispatching and electricity market transactions as a whole virtual power plant. The core function of the virtual power plant cloud platform layer is to break geographical limitations, utilize communication technology to collaboratively control edge controllers in different regions, process adjustable information and operational status reported by the edge control layer, and interact with the trading center and dispatch center. If the combination of aggregates lacks market-demand-oriented optimization, the resulting virtual power plant will mismatch with the peak-shaving and frequency regulation ancillary service needs of the electricity market in key performance dimensions such as adjustable power range, response speed, and sustainable regulation time, weakening the virtual power plant's competitiveness and probability of winning bids in market competition. Based on this, the technical solution of this application further uses the market fit index of adjustable power range, response time, and sustainable regulation time in the adjustable feature set of aggregates as the alliance utility function, and performs dynamic alliance game solving on each aggregate to obtain the cloud alliance structure. Through the above processing, market fit can be used as the utility driver. Through adaptive optimization under the game theory framework, the aggregates of various regions can be combined into a virtual power plant alliance that maximizes the expected global return. This forms the optimal organizational structure of the cloud platform layer, enabling the constructed virtual power plants to be highly adapted to the needs of electricity market ancillary services in terms of adjustable power limits, response time, and duration. This provides a competitive cloud alliance structure foundation for receiving market transaction instructions and distributing them at multiple levels.

[0054] Figure 5 This is a block diagram of the alliance game solving module in a collaborative optimization configuration system based on a virtual power plant edge-cloud dual-layer structure, according to an embodiment of this application. Figure 5As shown, the coalition game solving module 140 includes: a feature extraction unit 141, used to decrypt and normalize the adjustable feature set of the aggregate based on a preset security key and cross-layer communication protocol specifications to obtain an aggregate extrinsic characteristic matrix; a fit degree calculation unit 142, used to perform weighted summation of the adjustable power range, response time, and duration in the aggregate extrinsic characteristic matrix based on preset preference weight coefficients and deduct the cloud communication cost penalty term to obtain a market fit degree index; a game model construction unit 143, used to map the market fit degree index to establish a coalition utility function, and instantiate a coalition game model for the participant set and strategy space of each aggregate to generate a coalition game model; and a coalition convergence solving unit 144, used to perform dynamic walks and Nash stable state convergence solutions in the strategy space of the coalition game model to obtain a cloud coalition structure.

[0055] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the feature extraction unit 141 is used to decrypt and normalize the adjustable feature set of the aggregates based on a preset security key and cross-layer communication protocol specifications to obtain the aggregate external characteristic matrix. It should be noted that since the adjustable feature set of the aggregates is reported from the edge control layer to the cloud platform layer via a dedicated wide area network line in encrypted and compressed ciphertext format, the cloud platform layer cannot directly read the external characteristic parameters. Furthermore, the physical dimensions and numerical ranges of parameters such as the adjustable power range, response time, and duration of each aggregate differ. Without normalization, a certain dimension parameter will dominate the scoring result in the subsequent market fit index calculation. Therefore, the technical solution of this application further decrypts and normalizes the adjustable feature set of the aggregates based on a preset security key and cross-layer communication protocol specifications to obtain the aggregate external characteristic matrix. Through the above processing, encrypted cross-layer transmission data can be restored to computable plaintext parameters, and the dimensional differences of parameters in each dimension can be eliminated, providing standardized numerical input for the subsequent weighted summation calculation of market fit index.

[0056] More specifically, in a concrete example of this application, the generation process of the aggregated extrinsic characteristic matrix includes two progressive steps: encrypted parsing and restoration, and normalized feature extraction. First, the communication gateway at the cloud platform layer receives encrypted data packets of the aggregated adjustable feature set from various edge controllers. It performs symmetric decryption on the data packets using a preset security key to restore the binary serialized data stream. Then, based on the data dictionary structure defined in the cross-layer communication protocol specification, it deserializes and parses the data stream to extract the key extrinsic characteristic parameters carried in the aggregated multidimensional feature boundary matrix of each regional aggregate, including the adjustable upper limit of power, the adjustable lower limit of power, response time, and duration. Taking the aforementioned industrial park scenario as an example, the regional aggregate containing distributed photovoltaic inverter access node A and electric vehicle charging pile metering node C, after decryption and parsing, has an adjustable upper limit of 160kW and an overall upward ramp rate of 45kW / min. The corresponding response time and duration, and other extrinsic characteristic parameters, are extracted simultaneously.

[0057] Based on this, the extracted external characteristic parameters of each aggregate across all dimensions are normalized to their minimum and maximum values, mapping each parameter to a unified numerical range of 0 to 1. For positive indicators such as adjustable power limit and duration, a larger normalized value indicates stronger market competitiveness of the aggregate in that dimension; for negative indicators such as response time, a smaller normalized value indicates faster response speed of the aggregate. The normalized external characteristic parameters of all aggregates are arranged in rows, with each row corresponding to a regional aggregate and each column corresponding to an external characteristic dimension, constructing a structured aggregate external characteristic matrix. Each element in this matrix is ​​a dimensionless normalized value that can directly participate in the weighted summation calculation of subsequent market fit indicators.

[0058] In the aforementioned collaborative optimization configuration system 100 based on a virtual power plant edge-cloud dual-layer structure, the matching degree calculation unit 142 is used to perform a weighted summation of the power adjustable range, response time, and duration in the aggregated external characteristic matrix based on preset preference weight coefficients, and deduct the cloud communication cost penalty term to obtain a market matching degree index. It should be noted that, although the normalized external characteristic parameters of each aggregate in the aggregated external characteristic matrix have eliminated dimensional differences, they have not yet been integrated into a single evaluation index that comprehensively reflects the aggregate's ability to participate in the power market. Furthermore, the cross-regional communication cost between each aggregate and the cloud platform layer is not included in the evaluation system. Ignoring cloud communication costs will lead to evaluation results deviating from the economic feasibility of actual operation. Therefore, the technical solution of this application further performs a weighted summation of the power adjustable range, response time, and duration in the aggregated external characteristic matrix based on preset preference weight coefficients and deducts the cloud communication cost penalty term to obtain a market matching degree index. Through the above processing, the multidimensional external characteristic parameters can be compressed into a single comprehensive score that takes into account both market adaptability and communication economy, providing a quantitative basis for the establishment of the alliance utility function in the subsequent alliance game model.

[0059] More specifically, in a concrete example of this application, the calculation process of the market fit index includes two progressive steps: preference weight loading and weighted summation, and cloud communication cost penalty deduction. First, preset preference weight coefficients are loaded. These coefficients reflect the differentiated demand preferences of the electricity market for the adjustable capabilities of virtual power plants across various dimensions. Normalized extrinsic parameters of each potential alliance are extracted row by row from the aggregated extrinsic characteristic matrix. The upper limit of adjustable power, the lower limit of adjustable power, response time, and duration are multiplied by their respective preference weight coefficients and then summed using weighted methods. A higher upper limit of adjustable power, a higher lower limit of adjustable power, and a longer duration result in a higher weighted sum score, indicating a stronger adjustable capability for the alliance in the market. A shorter response time indicates stronger market competitiveness; therefore, the weight coefficient before the response time term is introduced with a negative sign, making the weighted contribution greater for shorter response times.

[0060] Based on this, the cloud communication costs required for coordination and control at the cloud platform layer by each potential alliance are loaded, multiplied by a communication cost conversion factor, and then deducted from the weighted sum score as a penalty to obtain the final market fit index. The calculation expression for this index is as follows:

[0061] in, For potential alliances Market fit index For the alliance Normalized adjustable power limit, For the alliance The normalized adjustable lower limit of power. For the alliance Normalized response time For the alliance The normalization duration is For the alliance The cloud communication costs required for coordination and control at the cloud platform layer. , , , These are the preset preference weight coefficients that characterize the importance of each external characteristic's ability to participate in the market. This is a conversion factor used to convert cloud communication costs into a market fit penalty term. In the above formula, These three factors collectively demonstrate that the greater the adjustable upper and lower limits of alliance participation in the market and the longer the duration of such participation, the higher the market fit. This term represents a negative correlation between shorter response times and higher market fit. The project incorporates the economic costs of cross-regional cloud communication into the evaluation system, thus suppressing alliances with excessively high communication costs in the market fit score.

[0062] In the aforementioned collaborative optimization configuration system 100 based on a dual-layer edge-cloud structure of a virtual power plant, the game model construction unit 143 is used to map the market fit index as a coalition utility function and instantiate and construct a coalition game model for the participant set and strategy space of each aggregate to generate a coalition game model. It should be noted that while the market fit index quantifies the comprehensive market adaptability score of each potential coalition, it has not yet been embedded into the mathematical framework of game theory to drive adaptive combination optimization among aggregates. The lack of formal definitions for the three elements—participant set, strategy space, and characteristic function—prevents the subsequent game solving process. Therefore, the technical solution of this application further maps the market fit index as a coalition utility function and instantiates and constructs a coalition game model for the participant set and strategy space of each aggregate to generate a coalition game model. Through the above processing, the aggregate combination optimization problem can be transformed into a cooperative game problem with a complete mathematical structure, providing a computable game model carrier for subsequent adaptive optimization solutions.

[0063] More specifically, in a concrete example of this application, all regional aggregates reported by the edge control layer are first abstractly defined as a set of participants in game theory. Each element in the set corresponds to a regional aggregate, and the total number of participants is equal to the number of flexible resource aggregates within the edge control layer. Based on the participant set, all possible non-overlapping and non-exclusive aggregate combination states are generated, that is, all partitioning methods of the participant set are traversed. Each partitioning method corresponds to a coalition combination strategy. The set of all partitioning methods constitutes the strategy space of this game problem. On this basis, the market fit index calculated above is directly mapped and established as the characteristic function, i.e., the coalition utility function, in this game model. The mapping relationship is as follows: ,in For the alliance The characteristic function value represents the sum of expected benefits that members of the alliance can obtain through cooperation. The market fit index value corresponding to the alliance is used to ensure that the utility evaluation of each alliance in the game model is consistent with its actual adaptability to the electricity market. Finally, the above three elements are instantiated and assembled to generate a complete alliance game model and loaded into the cloud computing cluster to await subsequent calls from the optimization and solution engine.

[0064] In the aforementioned collaborative optimization configuration system 100 based on a virtual power plant edge-cloud dual-layer structure, the alliance convergence solution unit 144 is used to perform dynamic walks and Nash stable state convergence solutions within the strategy space of the alliance game model to obtain the cloud alliance structure. It should be noted that, since the alliance game model has completed the formal definition of the three elements—participant set, strategy space, and characteristic function—but has not yet performed actual solution operations to determine the optimal aggregate combination scheme that maximizes the global expected benefit, the number of partitioning methods in the strategy space increases exponentially with the number of participants, making exhaustive solutions computationally infeasible. Based on this, the technical solution of this application further performs dynamic walks and Nash stable state convergence solutions within the strategy space of the alliance game model to obtain the cloud alliance structure. Through the above processing, the aggregate combination strategy that maximizes the overall expected benefit of the entire system can be obtained within an acceptable computational time, forming the optimal organizational structure of the cloud platform layer.

[0065] More specifically, in a concrete example of this application, an instantiated coalition game model is loaded into a cloud computing cluster, and a merge and split iterative algorithm based on the Pareto criterion is deployed as the core solution engine. The global optimization objective of this algorithm is to find the optimal partitioning scheme among all possible partitioning structures that maximizes the sum of the utility function values ​​of each coalition. Its mathematical expression is as follows:

[0066] in, This is a unique alliance partitioning structure that avoids overlap and omission in all regional aggregates. To divide a specific alliance in the structure, Let be the characteristic function value of the alliance. In each iteration, the algorithm evaluates the utility changes under different alliance states. When the total utility of two mutually exclusive alliances after merging is strictly greater than the sum of the two before merging, the merge operation is performed. Conversely, if the sum of the utilities of the subsets after splitting an alliance is greater, the split operation is performed. The merge and split operations perform dynamic walks and state transitions in the policy space until there are no combinatorial transformations with Pareto improvement space. When the algorithm reaches a Nash stable state, the iteration terminates. The aggregation strategy in the convergent state is instantiated and encapsulated as the final solution, and the cloud alliance structure is output.

[0067] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the instruction decomposition and distribution module 150 is used to perform multi-level optimal decomposition and layer-by-layer distribution of global transaction instructions from the cloud platform layer to the edge control layer based on the collaborative matching relationship between the cloud alliance structure and the received market transaction instructions to obtain the terminal control instruction set. It should be noted that, given that the cloud alliance structure has determined the aggregate combination scheme that maximizes the overall expected benefits of the entire system, this structure is only a static organizational definition and has not yet been connected with the actual market transaction instructions issued by the power grid dispatch center and trading center. The core function of the virtual power plant cloud platform layer is to receive market transaction instructions and decompose and distribute them, decomposing and accurately transmitting the global power demand from the cloud platform layer to the edge control layer and then to the flexible resource terminal layer from top to bottom. Without a multi-level optimal decomposition mechanism, the global transaction instructions cannot be reasonably broken down into specific power allocation schemes that can be executed by each regional aggregate and its internal terminal equipment. This results in the virtual power plant being unable to effectively fulfill the power grid dispatch instructions in ancillary service scenarios such as peak shaving and frequency regulation. At the same time, after receiving regional instructions, the edge control layer also needs to perform secondary optimal decomposition based on the real-time available capacity limits of each flexible resource within the region to minimize the overall control cost within the region and ensure that the physical constraints of each terminal equipment are not violated. Based on this, the technical solution of this application further leverages the collaborative matching relationship between the cloud alliance structure and the received market transaction instructions. It performs multi-level optimal decomposition and layer-by-layer distribution of global transaction instructions from the cloud platform layer to the edge control layer to obtain the terminal control instruction set. Through the above processing, a top-down, cloud-to-edge two-layer collaborative optimization configuration and control closed loop can be achieved, accurately decomposing global market transaction instructions into action instructions that can be recognized and executed by terminal physical devices, thus completing the full-link response and fulfillment of the virtual power plant's demand for electricity market ancillary services.

[0068] Figure 6This is a block diagram of the instruction decomposition and distribution module in a virtual power plant edge-cloud dual-layer collaborative optimization configuration system according to an embodiment of this application. Figure 6 As shown, the instruction decomposition and distribution module 150 includes: a capacity allocation unit 151, used to calculate and encapsulate the total system target power in market transaction instructions according to capacity ratio and regional identifier based on the capacity weight coefficient of each regional aggregate in the cloud alliance structure to obtain regional control instructions; a constraint optimization decomposition unit 152, used to perform constraint optimization decomposition of regional control instructions based on the adjustable upper and lower limit constraints of each terminal node and the objective function of minimizing the overall regional control cost to obtain the node target power matrix; and a message encapsulation unit 153, used to perform proportional scaling quantization conversion and verification message encapsulation of the node target power matrix based on the physical communication register specification and cyclic redundancy check protocol of the terminal device to obtain the terminal control instruction set.

[0069] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the capacity allocation unit 151 is used to calculate and encapsulate the total system target power in the market transaction instruction according to capacity ratio and regional identifiers based on the capacity weight coefficients of each regional aggregate in the cloud alliance structure to obtain regional-level control instructions. It should be noted that since the market transaction instructions received by the cloud platform layer are the total system target power for the entire virtual power plant, while the actual control execution requires edge controllers distributed in different regions to drive their respective flexible resource terminals to complete the physical response, the global power instruction must be reasonably divided into the target power share that each regional aggregate should bear. Based on this, the technical solution of this application further calculates and encapsulates the total system target power in the market transaction instruction according to capacity ratio and regional identifiers based on the capacity weight coefficients of each regional aggregate in the cloud alliance structure to obtain regional-level control instructions. Through the above processing, the global power demand can be fairly and reasonably decomposed initially according to the actual adjustable capacity ratio of each aggregate, providing a clear regional-level power target for the subsequent secondary optimal decomposition of the edge control layer.

[0070] More specifically, in a concrete example of this application, the market transaction instruction issued by the power grid dispatch center is first received and parsed, the total system target power distribution within the current dispatch cycle is extracted, and the cloud alliance structure generated by the preceding steps is loaded. The capacity weight coefficients mapped to each regional aggregate in this structure are extracted, and the total system target power distribution is mapped and allocated to each regional aggregate using a capacity weight ratio allocation method. The calculation expression is as follows:

[0071] in, To be issued to the first Regional aggregates in The target power allocation value at time t. Extracted from market trading orders The total target power of the system at any given time. For the first extracted from the cloud consortium structure Capacity weighting coefficients for regional aggregates This refers to the complete set of regional aggregates contained in the cloud alliance structure. For set The aggregate traversal index is used. The target power allocation value of each region is appended with the corresponding region aggregate number identifier and high-precision timestamp, encapsulated to generate a region-level control command, and accurately sent to the corresponding edge controller.

[0072] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the constraint optimization decomposition unit 152 is used to perform constraint optimization decomposition on the regional-level control commands based on the adjustable upper and lower limits of each terminal node and the objective function of minimizing the overall regional control cost to obtain the node target power matrix. It should be noted that since the regional-level control commands only specify the target power that each regional aggregate should bear as a whole, but the same aggregate contains multiple heterogeneous flexible resource terminals with different real-time available capacity, ramp-up rate, and control cost coefficients, a simple equal distribution method would lead to some terminals exceeding physical constraints or an overall high control cost. Therefore, the technical solution of this application further performs constraint optimization decomposition on the regional-level control commands based on the adjustable upper and lower limits of each terminal node and the objective function of minimizing the overall regional control cost to obtain the node target power matrix. Through the above processing, fine-grained decomposition of regional-level commands can be completed with minimal control cost while satisfying the physical constraints of each terminal, providing an optimal power allocation scheme for the subsequent physical response of terminal equipment.

[0073] More specifically, in a specific example of this application, after receiving a regional-level control command belonging to its own region, the edge controller parses the overall target power of its region. Using minimizing the overall control cost within the region as the objective function, and satisfying the equality constraints of the overall target power of the region and the inequalities of the adjustable upper and lower limits of each terminal node as hard boundaries, a quadratic constraint optimization model is constructed. Its mathematical expression is as follows:

[0074] in, The objective function is the overall regional regulation cost. To belong to the The set of all underlying nodes for flexible resources within a region Number the underlying node devices within the region using an index. For the pre-set number The cost coefficient for individual power regulation of each node device The first calculated by the solver Each node The specific output target value at any given time. The objective function is a quadratic form that sets the overall target power value for the region based on the regional-level control commands input from upstream. This objective function allocates a smaller power share to terminal devices with higher control costs, while allowing lower-cost terminal devices to undertake more control tasks, thereby minimizing the overall control cost while maintaining regional power balance. The Lagrange multiplier method is used to solve the model, serializing the optimal power output planning values ​​of each terminal node according to the device communication network address, and integrating them to generate a structured node target power matrix.

[0075] In the aforementioned virtual power plant edge-cloud dual-layer collaborative optimization configuration system 100, the message encapsulation unit 153 is used to perform proportional scaling quantization conversion and verification message encapsulation on the node target power matrix based on the physical communication register specifications and cyclic redundancy check protocol of the terminal equipment to obtain the terminal control instruction set. It should be noted that since the planned output values ​​of each terminal in the node target power matrix are stored in floating-point form, and the physical controller of the flexible resource terminal can only recognize integer control instruction words conforming to specific communication register specifications, and the electromagnetic interference environment in industrial sites may cause data bit flipping during transmission, without format conversion and integrity verification, the terminal equipment will be unable to correctly parse the instructions or execute incorrect power responses. Based on this, the technical solution of this application further performs proportional scaling quantization conversion and verification message encapsulation on the node target power matrix based on the physical communication register specifications and cyclic redundancy check protocol of the terminal equipment to obtain the terminal control instruction set. Through the above processing, the optimal power allocation scheme at the numerical level can be transformed into communication messages that the terminal physical equipment can directly recognize and execute, completing a top-down, cloud-to-edge dual-layer collaborative optimization configuration and control closed loop.

[0076] More specifically, in a specific example of this application, the edge communication gateway sequentially reads the target power floating-point values ​​of each terminal in the node target power matrix. Based on the scaling factor defined in the specific physical communication register specification of the terminal device, the target power floating-point number is multiplied by the device scaling factor and then rounded to an integer. This is converted into an integer control instruction word that meets the register precision requirements and represented as a hexadecimal byte string. Subsequently, the hardware slave address of the target device, the function control write code, and the converted hexadecimal data word are concatenated to form the message body. A 16-bit cyclic redundancy check is performed on the entire byte stream, and the check code is appended to the end of the message. Finally, the complete and interference-resistant communication message is encapsulated into a terminal control instruction set and directly sent to each physical device in the flexibility resource terminal layer to trigger mechanical or electrical responses, completing the two-layer collaborative optimization configuration closed loop.

[0077] In summary, the edge-cloud dual-layer collaborative optimization configuration system based on the embodiments of this application is explained. It achieves full-link closed-loop control from the bottom layer resource status perception to the top layer market transaction response through five progressive collaborations: flexible resource terminal layer data acquisition, edge control layer partition aggregation, external characteristic identification and feasible domain mapping, cloud platform layer dynamic alliance game, and multi-level instruction decomposition and issuance. In the edge control layer partitioning stage, to address the issue of quantization distortion in inter-node communication costs, a heterogeneous channel nonlinear attenuation penalty quantization mechanism is introduced. This mechanism combines the media reference attenuation rate and environmental noise figure to characterize the real channel transmission quality of different communication media under complex electromagnetic environments. Simultaneously, a dynamic routing congestion assessment mechanism is introduced, capturing dynamic queuing delay risks on multi-hop forwarding paths based on shortest path search and real-time traffic throughput of relay nodes. Building upon this, a comprehensive cost fusion operator is employed to perform multi-dimensional nonlinear fusion of the channel attenuation penalty matrix and the dynamic routing congestion matrix. This effectively quantifies the retransmission storm amplification effect when channel degradation and routing congestion intensify simultaneously. Consequently, a communication cost matrix that truly reflects the channel quality differences and dynamic load risks between node pairs in a heterogeneous communication environment is generated. This prevents nodes with weak communication or congested paths from being incorrectly aggregated, ensuring the communication robustness and control response reliability of the virtual power plant in high-frequency control scenarios.

Claims

1. A collaborative optimization configuration system based on a virtual power plant edge-cloud dual-layer structure, characterized in that, include: The resource status acquisition module is used to collect the injected power, load power, and adjustable capability parameters including power capacity and ramp rate of each flexible resource terminal at each operating moment based on the topological connection relationship represented by the set of nodes and the set of branches in the power network to obtain a resource status dataset. The partition aggregation optimization module is used to perform partition aggregation optimization on the flexible resources in the resource status dataset to obtain an edge partitioning scheme while satisfying the constraint of the highest aggregation communication cost. The external characteristic identification module is used to identify the external characteristics and map the feasible region of flexible resources in the resource status dataset based on the edge partitioning scheme to obtain the adjustable feature set of the aggregate. The alliance game solving module is used to solve the dynamic alliance game of each aggregator by using the market fit index of adjustable power range, response time and duration of the adjustable features of the aggregator as the alliance utility function to obtain the cloud alliance structure. The instruction decomposition and distribution module is used to perform multi-level optimal decomposition and layer-by-layer distribution of global trading instructions based on the collaborative matching relationship between the cloud alliance structure and the received market trading instructions, so as to obtain the terminal control instruction set.

2. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 1, characterized in that, The resource status acquisition module includes: The topology instantiation unit is used to instantiate the spatial connectivity relationship of the node set and the branch set based on the connection status of the node set and the branch set in the power network to obtain the network topology physical information. The net power calculation unit is used to calculate the time-by-time difference of net power for each node and splice the time sequence based on the difference between injected power and load power at each operating time to obtain the real-time net power sequence of the node. The margin assessment binding unit is used to impose a dual constraint on the adjustment upper limit of each terminal based on power capacity and ramp rate, and to perform flexibility margin assessment and spatial attribute binding on the real-time net power sequence of the node and the network topology physical information to obtain multi-dimensional feature binding data of the terminal. The serialization encapsulation unit is used to perform unified structure transformation and serialization compression on the terminal's multi-dimensional feature binding data based on a preset lightweight communication protocol data dictionary standard to generate a resource status dataset.

3. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 2, characterized in that, The margin assessment binding unit includes: performing a flexibility margin assessment using the following formula: in, For the first Each terminal node The upward flexibility adjustment margin at any time, For the first The upper limit of the rated active power capacity of each terminal node. For the first Each node Real-time net power at any given moment For the first The upper limit of the ramp rate for each terminal device. To represent the discrete control time step of the system, It is a local minimum function.

4. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 1, characterized in that, The partition aggregation optimization module includes: The topology mapping unit is used to perform topology mapping and parameter parsing on the resource status dataset based on the node set and branch set in the resource status dataset to obtain a physical topology undirected graph. The communication cost quantification unit is used to quantify the communication cost between any two nodes in the physical topology undirected graph based on the spatial geographic coordinates of each node and the preset unit distance communication cost coefficient to obtain the weighted graph model and the inter-node communication cost matrix. The modularity evaluation construction unit is used to construct a modularity evaluation by fusing the communication cost matrix between nodes and the degree of edge connections in the weighted graph model with the communication cost of the node communication cost matrix as an economic penalty term, so as to obtain an improved modularity objective function and aggregation constraint matrix. The segmentation optimization unit is used to perform heuristic graph theory segmentation optimization within the constraint space of the improved modularity objective function based on the maximum aggregation communication cost limit set by the aggregation constraint matrix, so as to obtain the edge partitioning scheme.

5. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 1, characterized in that, The external characteristic identification module includes: The regional placement unit is used to accurately deliver and place flexible resources in the resource status dataset by region based on the node affiliation mapping relationship of each region in the edge partitioning scheme to obtain the regional clustered physical status set. The equivalent aggregation deduction unit is used to perform equivalent aggregation deduction on the power capacity and ramp rate of each individual device in the same aggregate within the physical state concentration of the regional cluster to obtain the equivalent parameters of the aggregate external characteristics and the set of node-level constraints. The feasible domain boundary fusion unit is used to project and fuse the equivalent parameters of aggregate external characteristics and the set of node-level constraints to obtain the multidimensional feature boundary matrix of the aggregate as the adjustable feature set of the aggregate.

6. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 1, characterized in that, The coalition game solving module includes: The feature extraction unit is used to decrypt and normalize the adjustable feature set of the aggregate based on the preset security key and cross-layer communication protocol specifications to obtain the aggregate external characteristic matrix. The fit calculation unit is used to perform a weighted summation of the adjustable power range, response time, and duration in the aggregated external characteristic matrix based on preset preference weight coefficients, and deduct the cloud communication cost penalty term to obtain the market fit index. The game model construction unit is used to map the market fit index into the coalition utility function and instantiate the coalition game model to generate the coalition game model by constructing the participant set and strategy space of each aggregate. The coalition convergence solution unit is used to perform dynamic walks in the policy space and Nash stable state convergence solutions on the coalition game model to obtain the cloud coalition structure.

7. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 1, characterized in that, The instruction decomposition and distribution module includes: The capacity allocation unit is used to calculate and encapsulate the total system target power in market trading instructions according to the capacity ratio and regional identifier based on the capacity weight coefficient of each regional aggregate in the cloud alliance structure to obtain regional control instructions. The constraint optimization decomposition unit is used to perform constraint optimization decomposition on the regional-level control command based on the adjustable upper and lower limit constraints of each terminal node and the objective function of minimizing the overall regional control cost to obtain the node target power matrix. The message encapsulation unit is used to perform scaling, quantization, and verification message encapsulation on the node target power matrix based on the physical communication register specifications and cyclic redundancy check protocol of the terminal device to obtain the terminal control instruction set.

8. The edge-cloud dual-layer collaborative optimization configuration system based on a virtual power plant according to claim 4, characterized in that, The communication cost quantification unit includes: The channel attenuation quantization subunit is used to perform channel nonlinear attenuation penalty quantization on any two nodes based on the spatial geographic coordinates of each node in the physical topology undirected graph and the medium reference attenuation rate and environmental noise coefficient in the heterogeneous medium attribute set to obtain the channel attenuation penalty matrix. The routing congestion assessment subunit is used to accumulate the congestion cost of hop count and dynamic queuing delay for each relay node on the routing path based on the shortest path search results of the physical topology undirected graph and the real-time traffic throughput of each relay node to obtain the dynamic routing congestion matrix. The integrated cost fusion subunit is used to perform integrated cost fusion calculation and edge weight binding on multi-dimensional communication features based on the channel attenuation penalty matrix, dynamic routing congestion matrix and fixed configuration cost of edge controller to obtain weighted graph model and inter-node communication cost matrix.

Images