A multi-element resource aggregation scheduling method and system for a virtual power plant
Patent Information
- Application Number
- CN202611247839.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-10-16
AI Technical Summary
然而，随着接入资源规模不断扩大，该架构暴露出日益严重的技术缺陷：首先，当资源规模达到万级时，通信复杂度为O(n)（n为资源设备总数），海量数据的采集、传输与处理导致调度响应延迟显著增加；其次，单层优化模型的优化变量维度为n，计算复杂度呈多项式增长，且未考虑不同资源在地理位置、响应特性、约束类型等方面的差异性；再次，中心化调度需收集所有资源的详细运行数据，造成巨大通信开销的同时暴露了资源用户的私有运行信息，损害了用户参与积极性；最后，现有技术多采用基于地理位置或资源类型的静态聚合规则，无法适应实际运行中资源出力特性和可调能力的动态变化，导致聚合形成的虚拟机组难以与电网调度需求高效匹配
本发明通过“区域自治—全局协同”双层多智能体架构设计，使全局协同层仅与区域智能体交互聚合能力边界信息而非所有资源的原始数据，通信复杂度从O(n)降低至O(m)，优化变量维度从n维降低至m维，有效解决了现有中心化架构在高并发场景下的通信瓶颈问题，显著降低了通信开销与计算复杂度，提升了系统在大规模资源接入场景下的可扩展性。
Smart Images
Figure CN122931778A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system automation, and specifically to a multi-resource aggregation and scheduling method and system for virtual power plants. Background Technology
[0002] With the continuous increase in installed capacity of new energy power generation facilities such as distributed photovoltaic and wind power, new flexible resources such as energy storage systems, controllable loads, and electric vehicles are also experiencing explosive growth. While promoting the consumption of clean energy, the rapid development of these diversified resources also brings unprecedented challenges to the safe and stable operation of the power system. On the one hand, the resources are numerous and diverse, encompassing controllable distributed power sources on the generation side, energy storage devices on the grid side, and massive distributed loads on the demand side, with significantly different response characteristics and complex and variable operating states. On the other hand, the resources are geographically widespread, with small individual capacity but large aggregate scale, making the traditional "source follows load" dispatching model difficult to apply directly.
[0003] To achieve effective integration and coordinated control of distributed resources, virtual power plant technology has emerged. It aggregates and encapsulates diverse distributed resources within a region using advanced information and communication technologies and intelligent optimization algorithms, forming adjustable and controllable "virtual machine groups" to participate in ancillary services such as grid peak shaving, frequency regulation, and voltage regulation. The core objective of virtual power plant dispatching technology is to rationally allocate the output plans of each resource based on grid dispatching instructions or market signals, minimizing dispatching costs while meeting system operational constraints.
[0004] Existing virtual power plant scheduling methods are primarily based on a centralized architecture, where the status data of all distributed resources must be uniformly aggregated to a central dispatch server, which then performs global status monitoring, optimization calculations, and command distribution. However, as the scale of accessed resources continues to expand, this architecture has exposed increasingly serious technical defects: First, when the resource scale reaches tens of thousands, the communication complexity is O(n) (where n is the total number of resource devices), and the collection, transmission, and processing of massive amounts of data leads to a significant increase in dispatch response latency; second, the optimization variable dimension of the single-layer optimization model is n, resulting in a polynomial increase in computational complexity, and it does not consider the differences between different resources in terms of geographical location, response characteristics, and constraint types; third, centralized scheduling requires the collection of detailed operational data for all resources, causing huge communication overhead while exposing the private operational information of resource users, thus undermining user participation; finally, existing technologies mostly adopt static aggregation rules based on geographical location or resource type, which cannot adapt to the dynamic changes in resource output characteristics and adjustability during actual operation, making it difficult for the aggregated virtual machine groups to efficiently match the grid dispatch requirements.
[0005] In summary, how to effectively reduce communication overhead and computational complexity, improve system scalability and real-time response capabilities, and achieve information encapsulation and privacy protection while ensuring global scheduling optimization has become a key technical challenge that urgently needs to be addressed in the field of virtual power plant scheduling technology. Summary of the Invention
[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a multi-resource aggregation and scheduling method and system for virtual power plants.
[0007] To achieve the above objectives, in a first aspect, the present invention provides a multi-resource aggregation and scheduling method for virtual power plants. The method is applied to a two-layer scheduling architecture based on multi-agent hierarchical collaboration, the architecture comprising a global collaboration layer and a regional autonomy layer composed of multiple regional agents. The method includes the following steps: The regional capability assessment module of each regional agent assesses the aggregation capability boundary of the region at the current moment based on the real-time operating status of local resources; the aggregation capability boundary includes minimum adjustable power, maximum adjustable power, response ramp rate, and aggregation energy capacity; Each of the aforementioned regional agents will report the obtained aggregation capability boundary information to the global aggregation capability evaluation module of the global collaboration layer; The scheduling decision center of the global coordination layer receives the scheduling target issued by the power grid dispatch center, performs global scheduling decision optimization based on the aggregation capability boundary of each regional intelligent agent, and generates the target scheduling power of each regional intelligent agent. The instruction distribution module of the global coordination layer distributes the target scheduling power of each regional agent to the corresponding regional agent. After receiving the scheduling instructions, each regional intelligent agent performs local optimization and redistribution by the local optimization solver, decomposing the overall regional scheduling target to each resource device under its jurisdiction; Each of the aforementioned regional agents executes a consensus protocol through a distributed coordination unit to synchronize and coordinate its state with neighboring regional agents; Each regional intelligent agent generates control commands for each resource device based on the local optimization and reallocation results and issues them for execution.
[0008] In some embodiments, the scheduling decision center of the global coordination layer performs global scheduling decision optimization based on an optimization model that includes power balance constraints, regional capacity constraints, ramp rate constraints, and energy balance constraints. The objective function of the optimization model is to minimize the sum of scheduling cost and deviation penalty, where the deviation penalty applies to both cases where the target power exceeds the upper limit and falls below the lower limit of regional capacity. Each regional agent executes a consensus protocol through a distributed coordination unit to achieve state synchronization and coordination between regions.
[0009] In some embodiments, the global collaboration layer is further deployed with a dynamic formation module for managing the dynamic formation relationships between the regional agents.
[0010] In some embodiments, the internal structure of the regional agent includes a perception module, a decision module, an execution module, and a communication interface; the perception module is used to collect the operating status of local resources in real time; the decision module includes a local optimization solver, a regional capability assessment module, and a distributed coordination unit.
[0011] In some embodiments, in a real-time frequency regulation scenario, when the regional agent detects that the grid frequency deviation exceeds a set threshold, it assesses the locally available frequency regulation resources and initiates a local frequency regulation response.
[0012] Secondly, this invention provides a multi-resource aggregation and scheduling system for virtual power plants, employing a two-layer scheduling architecture based on hierarchical multi-agent collaboration of regional autonomy and global collaboration, including: The global coordination layer includes a scheduling decision center, a global aggregation capability assessment module, an instruction distribution module, and a dynamic formation module. A regional autonomous layer, comprising multiple regional agents, each of which is communicatively connected to the global coordination layer; The resource device layer includes multiple distributed resource devices, each of which is communicatively connected to the intelligent agent of its respective region. The global aggregation capability assessment module is used to receive the aggregation capability boundary information reported by each of the regional agents; The scheduling decision center is used to receive the scheduling targets issued by the power grid scheduling center, perform global scheduling decision optimization based on the aggregation capability boundary of each regional intelligent agent, and generate the target scheduling power of each regional intelligent agent. The instruction distribution module is used to distribute the target scheduling power of each of the regional agents to the corresponding regional agents. The dynamic formation module is used to identify regional agents with similar resource characteristics or geographical proximity as candidate sets, initiate the alliance formation process, manage the election of coordinators within the alliance, and trigger the alliance dissolution process, so as to realize the dynamic aggregation and disaggregation among regional agents. Each of the aforementioned regional intelligent agents includes a regional capability assessment module, a local optimization solver, and a distributed coordination unit. The regional capability assessment module is used to assess the current aggregation capability boundary of the region based on the real-time operating status of local resources. The local optimization solver is used to decompose the overall regional scheduling target to the various resource devices under its jurisdiction. The distributed coordination unit is used to synchronize and coordinate the state with neighboring regional intelligent agents through a consensus protocol.
[0013] In some embodiments, the aggregation capacity boundary includes minimum adjustable power, maximum adjustable power, response ramp rate, and aggregated energy capacity; the global coordination layer also includes a real-time forecasting module for forecasting ultra-short-term load and renewable energy output.
[0014] In some embodiments, the dynamic formation module includes an aggregation alliance identification unit, an alliance formation unit, an alliance initialization unit, an alliance operation unit, and an alliance disbandment unit.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention employs a two-layer, multi-agent architecture design of "regional autonomy - global collaboration," enabling the global collaboration layer to interact only with regional agents to aggregate capability boundary information rather than the raw data of all resources. This reduces communication complexity from O(n) to O(m) and the optimization variable dimension from n-dimensional to m-dimensional, effectively solving the communication bottleneck problem of existing centralized architectures in high-concurrency scenarios. It significantly reduces communication overhead and computational complexity, and improves the scalability of the system in scenarios with large-scale resource access.
[0016] This invention enables regional agents to independently complete resource reallocation and decision-making locally through the local optimization solver and fast response mechanism of the regional autonomous layer, without waiting for the global optimization calculation to be completed, thus effectively shortening the scheduling response time.
[0017] This invention realizes dynamic aggregation and deaggregation among regional intelligent agents through a dynamic formation module, enabling virtual power plants to flexibly adjust their aggregation structure according to real-time changes in resource characteristics and operating status. This effectively overcomes the problem of poor adaptability of static aggregation methods and supports plug-and-play and dynamic expansion of resources.
[0018] This invention implements a consistency protocol through a distributed coordination unit, thereby achieving state synchronization and coordination between regions under abnormal operating conditions such as communication delays or packet loss, and improving the reliability of system operation.
[0019] This invention uses an aggregation capability boundary reporting mechanism to ensure that the information reported by the regional agent to the global layer is only the aggregation capability boundary parameters, without including the detailed operating parameters of individual resource devices. This achieves information encapsulation and privacy protection while meeting global optimization requirements. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram of the overall process of the multi-resource aggregation and scheduling method provided by the present invention; Figure 3 This is a schematic diagram of the multi-level interaction relationships and data flow between the global collaboration layer, the regional autonomy layer, and the resource device layer in this invention; Figure 4 This is a schematic diagram of the internal structure and data flow of the regional intelligent agent in this invention; Figure 5 This is a flowchart illustrating the dynamic formation function in this invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of this invention clearer and more complete, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention in any way. Example 1
[0022] This embodiment provides a multi-resource aggregation scheduling system and method for virtual power plants. The system adopts a two-layer scheduling architecture of "regional autonomy - global collaboration" based on multi-agent hierarchical collaboration. The virtual power plant scheduling task is decomposed into two levels: local optimization at the regional autonomy layer and cross-regional coordination at the global collaboration layer. Information encapsulation and computation decoupling are achieved through the aggregation capability boundary constraint mechanism.
[0023] like Figure 1 As shown, the scheduling system comprises three core layers: a global coordination layer, a regional autonomy layer, and a resource and equipment layer.
[0024] The global coordination layer, as the highest decision-making level of the entire dispatching system, is responsible for receiving dispatching instructions from the power grid dispatching center and making global optimization decisions. The global coordination layer includes a dispatching decision center, a global aggregation capability assessment module, an instruction distribution module, a real-time prediction module, and a dynamic formation module.
[0025] The dispatch decision center, as the core control unit of the global coordination layer, receives day-ahead dispatch plans or real-time dispatch instructions from the power grid dispatch center and triggers corresponding dispatch optimization processes based on the received dispatch objectives. The dispatch decision center integrates a global optimization solver, which employs a mixed-integer programming-based optimization algorithm, using the primal-dual interior-point method as its core solution engine. The dispatch decision center connects to the power grid dispatch center via a communication interface, with the communication protocol conforming to the IEC 60870-5-104 standard.
[0026] The global aggregation capability assessment module receives aggregation capability boundary information reported by agents in each region and builds and dynamically maintains the overall aggregation capability model of the virtual power plant based on this information. Internally, this module maintains an aggregation capability database, employing a time-series database structure. It stores the capability boundary parameters reported by all agents in each region according to a time series and updates them in real time to reflect the latest changes in the resource status of each region.
[0027] The instruction distribution module encapsulates the optimized scheduling instructions into a standard message format and distributes them to the corresponding regional agents via the communication network. The module supports a priority queue mechanism, which determines the distribution order based on the urgency of the scheduling instructions. The scheduling instruction message includes the target power value, instruction activation time, duration, and priority parameters. The module also supports an instruction issuance confirmation mechanism; if no confirmation message is received from a regional agent within a set time, a retransmission process is automatically triggered.
[0028] The real-time forecasting module is used to forecast ultra-short-term load and renewable energy output based on historical operating data and real-time collected environmental information. This module integrates a time series forecasting model (ARIMA model) and a machine learning forecasting model (Long Short-Term Memory network LSTM), with a forecasting time window of 5 to 15 minutes, providing data support for scheduling optimization.
[0029] The dynamic formation module manages the dynamic formation relationships between regional agents, enabling them to form temporary aggregation alliances with similar resource characteristics or geographical proximity. Within these alliances, scheduling information is shared, and the global coordination layer treats the alliance as a whole when allocating scheduling targets. The specific workflow of the dynamic formation module will be described in detail in Example 2.
[0030] The Regional Autonomy Layer, acting as an intermediate layer connecting the Global Collaboration Layer and the Resource and Device Layer, is responsible for managing all distributed resources within a specific geographical area or resource type. The core components of the Regional Autonomy Layer are multiple regional intelligent agents. Each regional intelligent agent corresponds to a specific geographical area or resource type, including load aggregation agents, energy storage aggregation agents, distributed energy aggregation agents, and electric vehicle aggregation agents. Each resource device belongs to a unique regional intelligent agent based on its geographical location and resource type. This affiliation is configured during system initialization and can be adjusted during operation through the alliance management of the dynamic formation module. Regional intelligent agents are key components for achieving local autonomous control, possessing independent information collection, decision optimization, and command execution capabilities.
[0031] like Figure 4 As shown, the internal structure of the regional intelligent agent includes four core sub-modules: perception module, decision-making module, execution module, and communication interface.
[0032] The sensing module is used to collect real-time operating status, power data, environmental parameters, and interaction information of neighboring intelligent agents from local resource devices. The sensing module periodically acquires device operating data through data acquisition terminals deployed on the resource devices. The acquisition period can be flexibly configured according to the real-time requirements of the scheduling scenario: 15 minutes in day-ahead scheduling scenarios and 100 milliseconds in real-time frequency modulation scenarios. The data types collected by the sensing module include parameters such as real-time device power, device operating status, device health indicators, ambient temperature, and irradiance. After preprocessing such as filtering, noise reduction, and outlier removal, the data is stored in a local status database.
[0033] The decision-making module is the core computing unit of the regional intelligent agent, which includes three sub-components: a local optimization solver, a regional capability assessment module, and a distributed coordination unit.
[0034] The local optimization solver decomposes the overall regional scheduling objective into the resources allocated to its subordinate equipment, employing a distributed optimization algorithm to achieve optimal resource allocation. Internally, the local optimization solver integrates a quadratic programming solver and a mixed-integer programming solver, enabling it to handle optimization problems with both linear and nonlinear constraints. The local optimization solver uses the minimum sum of the total operating costs of all resource equipment over the entire scheduling cycle as its objective function, with constraints including the sum of the target powers of all equipment equaling the overall regional scheduling objective and the target powers of each equipment not exceeding the equipment's capacity boundary. The local optimization solver iteratively updates the target powers of each equipment using the Lagrange relaxation method and the Alternating Direction Multiplier Method (ADMM) until convergence to the optimal solution.
[0035] Specifically, for a region containing M resource devices, the mathematical expression of the local optimization problem is: Objective function:
[0036] The constraints are as follows: Power balance constraints:
[0037] Equipment capacity constraints:
[0038] Climbing speed constraint:
[0039] in, Let the target power of the j-th resource device be in time period t. This is a function representing the operating cost of the equipment. The overall scheduling target for the region during time period t. and These are the minimum and maximum allowable power values for the j-th resource device (for energy storage devices). Take the maximum discharge power value (negative value). Take the maximum charging power value (positive value)). Let the ramp rate be the rate of the j-th resource device. The duration of each time period.
[0040] The local optimization solver incorporates power balance constraints into the objective function to form an augmented Lagrangian function, and iteratively updates the target power and Lagrangian multipliers of each device using the alternating direction multiplier method. The convergence condition for the iteration is that the change in power allocation scheme between two consecutive iterations is less than a set threshold.
[0041] The regional capacity assessment module is used to assess the current aggregation capacity boundary of a region based on the real-time operating status and historical data of local resources. Output parameters include the minimum adjustable aggregation power. Maximum adjustable aggregation power Aggregate response ramp rate and aggregated energy capacity .
[0042] The minimum adjustable power is the total adjustable power of all resource devices in the area relative to the current power in the current state, and its calculation formula is as follows:
[0043] in, Let j be the current actual power of the j-th resource device. Let be the algebraic lower limit of the adjustable power of the j-th resource device. For adjustable load devices, Take the minimum allowable operating power of the equipment; for energy storage equipment, Take the maximum discharge power (negative value); for distributed power supply equipment, Take the minimum technical output value.
[0044] The maximum adjustable power is the total adjustable power of all resource devices in the area relative to the current power under the current state, and its calculation formula is as follows:
[0045] in, Let be the upper limit of the adjustable power of the j-th resource device (algebraic value). For adjustable load devices, Take the maximum allowable operating power; for energy storage devices, Take the maximum charging power (positive value); for distributed power devices, Take the maximum available power limit.
[0046] The polymerization response ramp rate is the maximum change in polymerization power per unit time in this region, and its calculation formula is:
[0047] in, Let be the ramp rate of the j-th resource device.
[0048] Aggregated energy capacity is the cumulative dispatchable energy of the region during the scheduling cycle, and its calculation formula is as follows:
[0049] in, Let j be the cumulative dispatchable power that the j-th resource device can provide within the scheduling period in its current state. For energy storage devices, For adjustable load equipment, This represents the maximum cumulative electricity consumption that can be saved by reducing power consumption within the scheduling cycle; for distributed power supply equipment... This represents the maximum cumulative power generation that can be provided within the dispatch cycle.
[0050] The distributed coordination unit is used to synchronize the state with neighboring agents through a consensus protocol, achieving coordination and optimization between regions. The distributed coordination unit exchanges state information with neighboring agents via a point-to-point communication network. Each agent updates its local state according to the consensus iteration formula:
[0051] Among them, state variables Take the marginal cost or total scheduling power of each region as a coordination measure. The convergence step size (within the range of values) ), Let i be the set of agents adjacent to the regional agent i. Iterate until the state variables of all agents converge to a consistent value. The convergence condition is that the difference between the state variables of all agents is less than a set threshold (which is [value]). to ).
[0052] The execution module includes an instruction parsing unit and a control instruction generation unit. The instruction parsing unit parses the scheduling instructions received from the global coordination layer, extracting key parameters such as the target power value, effective time, duration, and priority. The control instruction generation unit generates direct control instructions for each resource device based on the local optimization and reallocation results, including power setpoint instructions, start / stop instructions, and mode switching instructions.
[0053] The communication interface is used for bidirectional information exchange with the global coordination layer and for point-to-point communication with neighboring agents. It supports multiple communication protocols, including MQTT, IEC 61850, and IEC 60870-5-104, ensuring compatibility with different types of devices. The interface also supports TLS encrypted transmission to ensure data transmission security.
[0054] The resource equipment layer includes various types of distributed resource equipment, such as adjustable load devices, energy storage devices, distributed photovoltaic devices, gas turbine devices, and electric vehicle charging piles. Each resource device connects to its respective regional agent through a standardized communication interface, enabling bidirectional transmission of status data and the issuance and execution of control commands. Regional agents share status information and coordinate interactions through a point-to-point communication network. Each regional agent only reports its aggregation capability boundary information, not raw equipment operating data, to the global coordination layer; the global coordination layer issues scheduling commands to each regional agent, which then performs local optimization and redistribution before issuing them to the resource devices.
[0055] The following is combined Figure 2 Taking a virtual power plant containing five regional intelligent agents as an example, the complete process of the method of the present invention will be described in detail.
[0056] The virtual power plant comprises five regional intelligent entities: Region 1 (industrial load aggregation, containing 200 adjustable load devices), Region 2 (energy storage aggregation, containing 150 energy storage units, each with a rated capacity of 1MWh and a rated power of 0.5MW), Region 3 (distributed photovoltaic aggregation, containing 500 photovoltaic inverters), Region 4 (commercial load aggregation, containing 300 adjustable devices), and Region 5 (electric vehicle aggregation, containing 1000 charging piles). In the day-ahead dispatch scenario, the dispatch cycle is 24 hours, with each time period lasting 15 minutes, totaling 96 dispatch periods.
[0057] In step S1, the capability assessment module of each regional agent starts the aggregation capability assessment process at 8:00 a.m. every day. Based on the real-time operating status of each resource device and meteorological forecast data, it assesses the aggregation capability boundary of the region in each time period of the next 24 hours.
[0058] Taking Region 2 (energy storage aggregate) as an example, the current SOC of the 150 energy storage units is distributed between 30% and 80%, and the current power is 0 (standby state). Each energy storage unit's... MW (discharge is negative) MW (charge is positive), ramp rate Each energy storage unit Unified at 20%, available energy . Indicates the current state of charge. This represents the minimum permissible state of charge. Calculated by the capability assessment module:
[0059]
[0060]
[0061]
[0062] Assuming the average SOC is 60%, then The energy capacity constraint and power constraint are independent of each other: the maximum adjustable power of 75MW limits the upper limit of instantaneous charging and discharging power at any given time, while the aggregated energy capacity of 60MWh limits the total dispatchable electricity throughout the day. Together, they constitute the complete dispatchable space of the energy storage aggregate. In the day-ahead dispatch of this embodiment, if a dispatch command requires Region 2 to discharge at a power of 75MW during a certain period, since the energy capacity is only 60MWh, this power level can only last for a maximum of 0.8 hours (60MWh / 75MW). The energy balance constraint in the actual optimization model will automatically limit the power allocation for each period to ensure that the total dispatchable electricity throughout the day does not exceed the limit. . The specific values for each time period are adjusted accordingly based on the predicted SOC for that time period, resulting in 96 time periods. , , , sequence.
[0063] In step S2, at 8:15, each regional agent encapsulates the evaluated aggregated capability boundary information into a standard data message and reports it to the global aggregated capability evaluation module of the global coordination layer via a TLS-encrypted communication connection. The message format includes a message header (protocol version number, message sequence number, timestamp), regional identifier, capability boundary data, and a checksum field. The information reported by each regional agent does not include detailed operating parameters of individual resource devices, but only the aggregated capability boundary.
[0064] After receiving data from each region, the aggregation capability assessment module of the global collaboration layer stores the data in the aggregation capability database (time series database) and updates the overall aggregation capability model of the virtual power plant.
[0065] In step S3, the power grid dispatch center issues the dispatch targets for the next day (0:00 to 24:00) to the virtual power plant at 9:00, including the total power demand for each time period. (t=1 to 96). After receiving the target, the scheduling decision center reads the aggregation capacity boundaries of each region provided by the aggregation capacity assessment module and the load and photovoltaic output forecasts provided by the real-time forecasting module, and triggers the global optimization process.
[0066] The optimization model for the scheduling decision center is as follows: Objective function:
[0067] in The coefficient is determined by the cost parameters reported by each region (Region 1: , Area 2: , Area 3: , Area 4: , Area 5: , ), This is the penalty coefficient (taken as 1000). and These represent the minimum and maximum adjustable power of the i-th region during time period t, respectively.
[0068] The constraints include: Power balance constraints:
[0069] Regional capability constraints:
[0070] Climbing speed constraint:
[0071] Energy balance constraints:
[0072] in Let be the aggregated energy capacity of the i-th region.
[0073] The global optimization solver uses the primal-dual interior-point method, with a total of 5 × 96 = 480 variables. The solver completes the solution in 5 minutes, outputting the target power matrix for 5 regions over 96 time periods. .
[0074] In step S4, at 9:05, the instruction distribution module encapsulates the optimized target scheduling power for each region into scheduling instruction messages and sends them to the corresponding regional agents via the MQTT protocol. Each region's instruction message contains the target power values for 96 time periods, the instruction's effective time (0:00 the next day), duration (24 hours), and priority. Upon receiving the instruction, the regional agent replies with an acknowledgment message; if no acknowledgment is received, a retransmission mechanism is triggered.
[0075] In step S5, each regional agent performs local optimization and reallocation between 9:06 and 9:20, decomposing the total scheduling target for 96 time periods to the various resource devices under its jurisdiction.
[0076] Taking region 1 (200 adjustable loads) as an example, for a single time period t, the local optimization problem is: Objective function:
[0077] The constraints are as follows:
[0078]
[0079]
[0080] This embodiment uses the interior-point method to solve the problem. To handle the ramping constraints between adjacent time periods, a decomposition and coordination strategy is adopted: first, each time period is solved independently, and then the ramping constraints between adjacent time periods are coordinated through Lagrange relaxation. The ADMM algorithm is used to iteratively update the target power of each device. The convergence condition of the iteration is that the change in the power allocation scheme between two adjacent iterations is less than a set threshold. Parallel computation was used for 96 time periods, with a total computation time of approximately 5 seconds. The local optimization solver outputs the power plans for each device in each time period. .
[0081] In step S6, after local optimization is completed, each region's agents execute the consensus protocol through the distributed coordination unit. Each region initializes its state variables. (Marginal cost obtained through local optimization), exchanges states with neighbors through a circular topology, according to the iterative formula. Update here The value is set to 0.25. After approximately 18 iterations, the state variables of each region converge to a consistent value. If the state deviation of each region exceeds the set threshold after iteration, the process returns to step S3 to adjust the global optimization parameters and recalculate.
[0082] In step S7, the control instruction generation unit of the regional agent generates control instructions for each resource device based on the local optimization and reallocation results. For day-ahead scheduling, the control instructions are the power setpoint plans for each time period of the next day, which are sent to each device in advance. Taking an industrial motor (rated power 100kW) as an example, the control instructions contain a sequence of power setpoints for 96 time periods, which are encapsulated as MODBUS write register instructions and sent to the device controller.
[0083] In step S8, after each device executes its power plan, it feeds back its actual power, operating status, and fault information to the regional agent. The regional agent collects the feedback data and calculates the deviation between the actual output and the target power. If the deviation exceeds a set threshold, the deviation event is recorded and a correction mechanism is triggered. During actual operation the following day, after each time period, each regional agent updates its local state database, reassesses the aggregation capability boundary, and reports it to the global coordination layer before the start of the next time period, forming a rolling optimization closed loop.
[0084] Example 2 This embodiment provides an extended implementation scheme that includes dynamic formation functionality to meet the actual operational needs of large-scale virtual power plants for flexible resource aggregation.
[0085] like Figure 5 As shown, the dynamic formation module is deployed in the global coordination layer to manage the dynamic formation relationships between regional agents. Traditional static regional partitioning methods are difficult to adapt to dynamic changes in resource types. When a specific type of resource is highly concentrated in a certain region, static partitioning may lead to low resource utilization efficiency. The dynamic formation module analyzes the resource type structure and aggregation capability characteristics of each regional agent to automatically identify scenarios where temporary aggregation alliances can be formed. An aggregation alliance consists of multiple regional agents with similar resource characteristics or geographical proximity. Within the alliance, scheduling information and coordination strategies are shared, forming a larger-scale resource aggregation unit.
[0086] During alliance operation, the aggregation capability boundaries of each member region are still independently reported to the global aggregation capability assessment module of the global coordination layer. However, the scheduling decision center of the global coordination layer treats all member regions as a whole during global optimization and issues the overall alliance scheduling target. After receiving the overall target, the alliance coordinator performs distributed optimization within the alliance, decomposing the overall target to each member region. Alliance identification is triggered periodically by the dynamic formation module at a preset cycle (e.g., every 15 minutes), or by regional agents actively reporting when aggregation capability parameters change significantly.
[0087] The workflow of dynamic formation includes the following steps: Aggregation Alliance Identification: The aggregation alliance identification unit of the dynamic formation module analyzes the resource type distribution and aggregation capability trends of agents in each region, identifying agents with similar resource characteristics or geographical proximity as candidate sets. The criteria for identifying similar resource characteristics include the proportion of the same resource type in the resource composition of each agent and whether the trends in aggregation capability parameters are similar. The criteria for identifying geographical proximity include whether the communication latency between agents in each region is less than a preset threshold and whether the distance between regions is within the effective communication range.
[0088] Alliance Formation: The alliance formation unit of the dynamic formation module sends alliance formation invitations to candidate regional agents. Each regional agent evaluates the expected benefits of alliance formation and then replies with confirmation or rejection. The alliance is successfully formed once the number of confirmed joining regional agents reaches a set threshold.
[0089] Alliance Initialization: Regional agents approved to join the alliance initialize alliance coordination parameters via the alliance initialization unit, including internal communication protocols, collaborative optimization objectives, and coordination cycles. An alliance coordinator is elected internally to handle internal optimization scheduling.
[0090] Alliance Operation: Under the control of the alliance operation units, the alliance coordinator receives the overall alliance scheduling target from the global coordination layer, performs distributed optimization within the alliance, and decomposes the overall alliance scheduling target to the intelligent agents in each member region. Each member region intelligent agent interacts with the alliance coordinator to coordinate information during local optimization. During alliance operation, the global coordination layer treats the alliance as a whole when allocating scheduling targets.
[0091] Alliance Dissolution: The alliance dissolution unit of the dynamic formation module continuously monitors the alliance's operational status. When the alliance's operational conditions are no longer met, the dynamic formation module triggers the alliance dissolution process, sending dissolution notifications to each member region, and each region's agents resume independent operation.
[0092] Let's take an industrial park scenario as an example. Assume a virtual power plant comprises 8 regional agents, each corresponding to one of the 8 industrial parks. During the daytime working hours, regions 1 through 5 have similar resource characteristics (primarily adjustable loads) and are geographically close. The dynamic formation module identifies that the proportion of resources of the same type exceeds half in these regions, triggering the alliance formation process. Regions 1 through 5 confirm their joining the alliance, and region 1 is elected as the alliance coordinator. The global coordination layer merges the independent scheduling targets of regions 1 through 5 into a unified alliance scheduling target and distributes it. In the evening, after the load decreases, the differences in resource characteristics among the regions increase, and the alliance's operational efficiency assessment shows that it no longer meets the conditions for continued operation, triggering the alliance dissolution.
[0093] Example 3 This embodiment describes the specific application of the present invention in a real-time frequency modulation scenario.
[0094] When power system frequencies fluctuate, virtual power plants need to respond quickly to provide frequency regulation ancillary services. The regional agent monitors the grid frequency in real time via a sensing module. The sampling period for frequency acquisition is set to 100 milliseconds in the frequency regulation scenario. When a deviation of the grid frequency from the nominal value (50Hz) is detected and the deviation exceeds a set threshold (±0.05Hz in this embodiment), the regional agent immediately initiates the local frequency regulation response process.
[0095] The regional agent first assesses the available frequency regulation resources locally and ranks them according to their response speed. Energy storage units have the fastest response speed, followed by rapidly adjustable loads, while ordinary adjustable loads have a relatively slower response speed. Based on the direction and magnitude of the frequency deviation, the regional agent determines whether to increase or decrease the output and the required adjustment amount.
[0096] Then, the regional agent reports its frequency modulation capability to the global coordination layer and requests frequency modulation authorization. While waiting for authorization from the global coordination layer, the regional agent directly initiates its local frequency modulation response. The energy storage unit immediately adjusts its output according to the preset droop characteristic curve, calculating the target power change based on the droop coefficient and frequency deviation. The mathematical expression for the droop characteristic curve is: ,in Where K is the power regulation amount, and K is the droop coefficient. This is for frequency deviation. The energy storage unit completes output adjustment within 100 milliseconds. Rapidly adjustable load synchronous response completes power change within 1 second. Ordinary adjustable loads complete power adjustment within seconds to minutes.
[0097] After initiating a local frequency modulation response, the regional agent reports its frequency modulation action information (including adjustment amount, response time, and resource types involved in the adjustment) to the global coordination layer. Based on the frequency modulation information reported by each region, the global coordination layer performs unified global optimization and adjustment in the next scheduling cycle to ensure that the frequency modulation actions of each region are coordinated and consistent.
[0098] Taking a certain energy storage cluster as an example, this area contains 120 available energy storage units with a total adjustable power of 60MW. When a frequency drop to 49.92Hz (deviation -0.08Hz) is detected, the regional agent calculates the required up-adjustment power based on the droop characteristic curve (droop coefficient of 0.05pu / Hz). After being limited, the actual output was adjusted to 60MW. The 120 energy storage units completed the output adjustment synchronously within 100 milliseconds, and then the regional intelligent agent reported the frequency regulation information to the global coordination layer.
[0099] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A multi-resource aggregation and scheduling method for virtual power plants, characterized in that, The method is applied to a two-layer scheduling architecture based on multi-agent hierarchical cooperation. The architecture includes a global cooperation layer and a regional autonomy layer composed of multiple regional agents. The method includes the following steps: The regional capability assessment module of each regional agent assesses the current aggregation capability boundary of the region based on the real-time operating status of local resources. The aggregation capability boundary includes minimum adjustable power, maximum adjustable power, response ramp rate, and aggregation energy capacity; each regional agent reports the evaluated aggregation capability boundary information to the global aggregation capability evaluation module of the global collaboration layer; The scheduling decision center of the global coordination layer receives the scheduling target from the power grid scheduling center, optimizes the global scheduling decision based on the aggregation capability boundary of each regional intelligent agent, and generates the target scheduling power of each regional intelligent agent. Each regional agent executes a consensus protocol through a distributed coordination unit to synchronize and coordinate its state with neighboring regional agents, so that the state variables of each regional agent converge to a consistent value. The instruction distribution module of the global coordination layer distributes the target scheduling power of each regional agent to the corresponding regional agent. After receiving the scheduling instructions, each regional intelligent agent performs local optimization and redistribution by the local optimization solver, decomposing the overall regional scheduling target to each resource device under its jurisdiction; Each of the regional intelligent agents generates control commands for each of the resource devices based on the local optimization and reallocation results and issues them for execution.
2. The multi-resource aggregation and scheduling method for virtual power plants according to claim 1, characterized in that, The minimum adjustable power is the sum of the lower limit of the adjustable power of all resource devices in the region under the current state relative to the current power. The maximum adjustable power is the sum of the upper limit of the adjustable power of all resource devices in the region under the current state relative to the current power. The response ramp rate is the maximum change in aggregated power per unit time in the region. The aggregated energy capacity is the cumulative schedulable energy of the region within the scheduling cycle.
3. The multi-resource aggregation and scheduling method for virtual power plants according to claim 1, characterized in that, The scheduling decision center of the global coordination layer optimizes global scheduling decisions based on an optimization model that includes power balance constraints, regional capacity constraints, ramp rate constraints, and energy balance constraints.
4. The multi-resource aggregation and scheduling method for virtual power plants according to claim 1, characterized in that, The dynamic formation module is deployed in the global coordination layer and is used to manage the dynamic formation relationship between the regional agents. The dynamic formation module performs the following steps: identifying regional agents with similar resource characteristics or geographical proximity as a candidate set; sending alliance formation invitations to the candidate regional agents; initializing alliance coordination parameters for the regional agents that are approved to join the alliance; electing an alliance coordinator within the alliance to be responsible for the optimization and scheduling within the alliance; and triggering the alliance dissolution process when the alliance operating conditions are no longer met, with each regional agent resuming independent operation.
5. The multi-resource aggregation and scheduling method for virtual power plants according to claim 1, characterized in that, The internal structure of the regional intelligent agent includes a perception module, a decision-making module, an execution module, and a communication interface; the perception module is used to collect the operating status of local resource devices in real time; the decision-making module includes a local optimization solver, a regional capability assessment module, and a distributed coordination unit.
6. The multi-resource aggregation and scheduling method for virtual power plants according to claim 1, characterized in that, In a real-time frequency regulation scenario, when the regional agent detects that the grid frequency deviation exceeds a set threshold, it assesses the available local frequency regulation resources and initiates a local frequency regulation response.
7. A multi-resource aggregation and scheduling system for virtual power plants, characterized in that, A two-tier scheduling architecture based on multi-agent hierarchical collaboration and regional autonomy and global collaboration is adopted, including: The global coordination layer includes a scheduling decision center, a global aggregation capability assessment module, an instruction distribution module, and a dynamic formation module. A regional autonomous layer, comprising multiple regional agents, each of which is communicatively connected to the global coordination layer; The resource device layer includes multiple distributed resource devices, each of which is communicatively connected to the intelligent agent of its respective region. The global aggregation capability assessment module is used to receive the aggregation capability boundary information reported by each of the regional intelligent agents; the scheduling decision center is used to receive the scheduling target of the power grid scheduling center, perform global scheduling decision optimization based on the aggregation capability boundary of each of the regional intelligent agents, and generate the target scheduling power of each of the regional intelligent agents. The instruction distribution module is used to distribute the target scheduling power of each of the regional agents to the corresponding regional agents; the dynamic formation module is used to manage the dynamic formation relationship between the regional agents. Each of the aforementioned regional intelligent agents includes a regional capability assessment module, a local optimization solver, and a distributed coordination unit. The regional capability assessment module is used to assess the current aggregation capability boundary of the region based on the real-time operating status of local resources. The local optimization solver is used to decompose the overall regional scheduling target to each resource device under its jurisdiction. The distributed coordination unit is used to synchronize and coordinate the state with neighboring regional intelligent agents through a consensus protocol. The polymerization capacity boundary includes the minimum adjustable power, the maximum adjustable power, the response ramp rate, and the polymerization energy capacity.
8. The multi-resource aggregation and scheduling system for virtual power plants according to claim 7, characterized in that, The global collaboration layer also includes a real-time forecasting module for forecasting ultra-short-term load and renewable energy output.
9. The multi-resource aggregation and scheduling system for virtual power plants according to claim 7, characterized in that, The global collaboration layer is also equipped with a dynamic formation module for managing the dynamic formation relationships between the regional agents; the dynamic formation module includes an aggregation alliance identification unit, an alliance formation unit, an alliance initialization unit, an alliance operation unit, and an alliance disbandment unit.