Flexibility-based computing center virtual power plant hierarchical control system and method

Abstract

The application relates to the technical field of electric power, in particular to a computing power center virtual power plant hierarchical control system and method based on a flexible feasible region, wherein the cloud server receives power grid dispatching requirements and issues the requirements to a computing power center coordination controller; the coordination controller calculates adjustment targets of each computing power center according to flexible capability credentials reported by each edge autonomous controller and issues the adjustment targets to edge autonomous controllers of corresponding computing power centers; the edge autonomous controllers are arranged in each computing power center, establish a target model according to an operating state, generate a flexible feasible region and compress the flexible feasible region into the flexible capability credentials for reporting, and generate terminal equipment control instructions according to the adjustment targets; and the terminal equipment controller is arranged at the terminal equipment end and executes the terminal equipment control instructions. Thus, the problems that in the related art, the flexible capability description dimension is single, the aggregated scheduling lacks a feasibility boundary check, the scheduling instructions are prone to exceeding the actual safety boundary of the computing power center, and there are business default and equipment heat risk and the like are solved.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a hierarchical control system and method for a virtual power plant based on a flexible feasible domain computing center. Background Technology

[0002] As the proportion of new energy sources in the power system continues to rise, the inherent volatility and intermittency of wind and solar power pose serious challenges to the frequency stability, voltage control, and power balance of the power grid. At the same time, the booming development of the digital economy has spurred massive demand for computing power, and computing centers (including cloud data centers and intelligent computing centers) have become a key growth point in energy consumption, characterized by large load base, rapid growth, and significant adjustability potential. Transforming computing centers from purely rigid loads into flexible resources capable of participating in system regulation is of great significance for improving the flexibility of the power system and promoting the efficient absorption of new energy sources.

[0003] In related technologies, existing demand response schemes for computing centers typically describe flexibility using a single power capacity or a fixed reduction curve, making it difficult to express the coupling relationship between response time, duration, ramp rate, reliability, business impact, and virtual inertia. Existing virtual power plant aggregation control also often lacks online verification of computing service quality, data center thermal safety, and energy storage state of charge. Dispatch instructions may exceed the actual feasible boundaries of the computing center, leading to business defaults, equipment thermal risks, or substandard grid response.

[0004] Furthermore, existing solutions typically model and schedule computing load, cooling load, and energy storage resources separately, failing to unify server computing power service quality constraints, thermal inertia constraints, energy storage state of charge constraints, and grid fast frequency response constraints into the same capability boundary. This results in a discrepancy between the adjustable capacity obtained by the upper-layer aggregator and the actual executable capacity on the edge side. When the task queue suddenly increases, cooling equipment fails, or the energy storage state of charge changes, the original static capacity declaration is prone to becoming invalid, making it difficult to meet the requirement for a verifiable correspondence between technical problems and technical effects in patent applications. Summary of the Invention

[0005] This application provides a hierarchical control system and method for a virtual power plant in a computing center based on a flexible feasible domain, in order to solve the problems in related technologies, such as the single dimension of flexible capability description and the lack of feasibility boundary verification in aggregated scheduling, which leads to scheduling instructions easily exceeding the actual safety boundary of the computing center, resulting in business default and equipment thermal risks.

[0006] The first aspect of this application provides a hierarchical control system for a virtual power plant based on a flexible feasible domain in a computing center, comprising: a cloud server, a computing center coordination controller, edge autonomous controllers, and terminal device controllers; wherein, the cloud server is used to receive scheduling requests issued by the power grid dispatch center and send the scheduling requests to the computing center coordination controller, converting the scheduling requests into structured scheduling tasks containing target power, response start time, duration, ramp constraints, and ancillary service types, and then sending them to the computing center coordination controller; the computing center coordination controller is communicatively connected to the cloud server and at least one edge autonomous controller in the computing center, and is used to receive flexible capability credentials reported by each edge autonomous controller, and based on the constraint matrix, boundary vector, state summary, and validity information in the flexible capability credentials, to perform rolling optimization of the structured scheduling tasks with feasible domain boundary constraints, generating adjustment targets for each computing center in the future scheduling time domain, and sending the adjustment targets to the corresponding edge autonomous controllers of the computing center; the edge autonomous controllers are deployed in each computing center and are used to establish descriptions of load, cooling system, etc., based on the operating status. A target model of the coupling relationship between the energy storage system and the data center thermal environment is constructed. A multi-dimensional flexible capability vector is built, consisting of up-adjustable power, down-adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient. Based on the target model and the multi-dimensional flexible capability vector, the flexible feasible domain and flexible capability certificate at the current moment are generated. The flexible feasible domain is compressed into a flexible capability certificate and reported to the computing center coordination controller. The flexible capability certificate contains the constraint representation of the flexible feasible domain, state summary, validity information, and integrity verification information, as well as terminal device control instructions generated according to the adjustment target, operating status, and target model. The terminal device controller is deployed at the server clusters, cooling equipment, energy storage converters, and power distribution equipment of each computing center. It is used to execute the terminal device control instructions issued by the edge autonomous controller and to feed back the corresponding actual response power, equipment status, and computing power service quality indicators to the edge autonomous controller. This allows the edge autonomous controller to perform online correction of the target model based on the actual response power, equipment status, and computing power service quality indicators, and update the flexible feasible domain and flexible capability certificate for the next cycle.

[0007] The state summary in the flexible capability credential includes a summary of server load rate, deferred task volume, data center temperature, energy storage state of charge, and available cooling capacity corresponding to the time the flexible feasible domain is generated; validity information includes the credential generation time, validity duration, and applicable scheduling time domain; integrity verification information includes verification values ​​calculated from the constraint matrix, boundary vector, and state summary. Through these limitations, the flexible capability credential is no longer an abstract capacity declaration, but a boundary data structure that can be directly verified and invoked by the coordination layer.

[0008] Optionally, the edge autonomous controller is further used to: acquire the current operating status information and safe operating constraints of the edge autonomous controller, wherein the current operating status information includes at least server load rate, deferred task queue, data center temperature, server chip temperature, cooling equipment operating status, energy storage state of charge, and grid frequency measurement; and the safe operating constraints include at least temperature upper limit, energy storage state of charge upper and lower limits, power supply capacity constraint, network latency constraint, and task completion constraint; using the operating status information as the starting benchmark for boundary search, with the goal of maximizing the value of the flexibility dimension, in multi-dimensional flexibility... In the space formed by the capability vectors, the limit boundary points of each dimension that satisfy the safe operation constraints are traversed and searched. Using the limit boundary points of each dimension as initial vertices, multiple flexible capability dimensions are combined and traversed to eliminate candidate points that do not satisfy the safe operation constraints, resulting in combined boundary vertices that satisfy the safe operation constraints. Based on the limit boundary points of each dimension and the combined boundary vertices, all boundary points obtained from the traversal search are used as fitting data, and polyhedral fitting processing is performed on the fitting data to obtain the target multidimensional flexible feasible region, which is jointly defined by the constraint matrix and the boundary vectors.

[0009] Optionally, the edge autonomous controller also includes a verification module. This module is used to input the current operating state into the target model before the edge autonomous controller generates control instructions for the terminal devices based on the adjustment target. The target model outputs the predicted values ​​corresponding to the execution of the adjustment target. These predicted values ​​include: data center temperature, server chip temperature, energy storage state of charge, network latency, and task completion rate. The target model includes: server thermal model, cooling system model, energy storage model, and task queuing model. If any predicted value reaches or exceeds the corresponding preset safety boundary, the flexible capability vector corresponding to the adjustment target is projected to the nearest feasible point within the current flexible feasible domain according to a preset distance metric. The terminal device control instructions are then regenerated based on the nearest feasible point. If, after multiple iterations of projection, a predicted value still exceeds the safety boundary, or the deviation between the nearest feasible point and the adjustment target exceeds the allowable deviation threshold, the adjustment target is rejected, and the corrected flexible feasible domain is re-reported to the computing center coordination controller.

[0010] Optionally, the computing center coordination controller is further used to: obtain the scheduling requirements at each moment in the rolling optimization time domain, using multiple consecutive future moments as the rolling optimization time domain; and solve for the optimization objectives, with the equivalent regulation power of each computing center at each moment and the corresponding response time, duration and ramp rate as decision variables, under the premise of satisfying the capability boundary constraints of each computing center, the total value of the regulation target covering the scheduling requirements and each individual regulation target being located within the corresponding flexible feasible domain, with the optimization objectives being the minimum total power tracking deviation, regulation ancillary service cost and computing power business service quality loss, and the maximum power market regulation benefit, and then distribute the solved regulation objectives of each computing center to the corresponding edge autonomous controller.

[0011] Optionally, the edge autonomous controller is further configured to: construct a temporary feasible region at the current moment based on the operating status between two adjacent flexible capability certificate updates; calculate the flexible safety margin based on the minimum normalized distance between the flexible capability vector corresponding to the adjustment instruction to be executed and each constraint boundary of the temporary feasible region; and execute the corresponding early warning action based on the flexible safety margin.

[0012] Optionally, the edge autonomous controller is further configured to: if the flexible safety margin is greater than the safety threshold, not generate an alarm and continue to execute the adjustment command normally; if the flexible safety margin is less than or equal to the safety threshold but greater than the warning threshold, generate a first-level alarm and actively adjust the operating strategy to increase the flexible safety margin; if the flexible safety margin is less than or equal to the warning threshold, generate a second-level alarm, report the flexible capability degradation to the computing center coordination controller, and request the computing center coordination controller to regenerate the adjustment target.

[0013] Optionally, the edge autonomous controller is further configured to: calculate the virtual inertial response power based on the energy storage state of charge, the length of the deferred task queue, the current room temperature and grid frequency measurements; generate virtual inertial response power instructions based on the virtual inertial response power, and decompose and execute the corresponding actions according to the priority order of the energy storage system, server load and cooling system.

[0014] The second aspect of this application provides a hierarchical control method for a computing center virtual power plant based on a flexible feasible domain. This method is applied to the hierarchical control system of the computing center virtual power plant based on a flexible feasible domain as described in the above embodiments. The method includes the following steps: obtaining dispatch requirements issued by the power grid dispatch center; establishing a target model describing various adjustable resources of the computing center based on the operating status; constructing a multi-dimensional flexible capability vector consisting of up-adjustable power, down-adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient; and generating the flexible feasible domain and flexible capability certificate at the current moment based on the target model; wherein the flexible capability certificate at least includes the constraint representation of the flexible feasible domain, state summary, validity information, and integrity verification information; calculating the adjustment target of each computing center in the future dispatch time domain based on the flexible capability certificate and dispatch requirements, under the constraint that the adjustment target of each computing center is located within the corresponding flexible feasible domain; generating terminal equipment control instructions based on the adjustment target, operating status, and target model; and updating the target model and the flexible capability certificate for the next cycle based on the actual response power, service quality impact, and equipment status after executing the terminal equipment control instructions.

[0015] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to perform the hierarchical control method for a virtual power plant based on a flexible feasible domain, as described in the above embodiments.

[0016] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to perform the hierarchical control method for a computing center virtual power plant based on a flexible feasible domain as described in the above embodiments.

[0017] The fifth aspect of this application provides a computer program product, including a computer program or instructions, which, when executed, implement the hierarchical control method for a virtual power plant based on a flexible feasible domain as described in the above embodiments.

[0018] Therefore, this application has at least the following beneficial effects: Compared to schemes that only report a single power reduction capability or a fixed response curve, this application incorporates power, time, ramp-up, and high-frequency support capabilities into the same feasible domain boundary, and provides this boundary as a certificate for optimization by the coordination layer. Compared to schemes that only perform scheduling and allocation at the upper-layer aggregator, this application sets up pre-execution projection verification and flexible safety margin hierarchical response at the edge side, which can handle boundary drift caused by state changes after certificate reporting. Therefore, the distinguishing features of this application have a direct technical causal relationship with avoiding out-of-bounds scheduling, reducing thermal risks, and reducing computing power service quality defaults.

[0019] This application embodiment utilizes a four-layer hierarchical architecture consisting of a cloud server, a computing center coordination controller, an edge autonomous controller, and a terminal device controller. This architecture enables the virtual power plant at the computing center to achieve a hierarchical decomposition and reliable closed-loop process from receiving grid dispatch requirements to executing terminal devices. The edge autonomous controller establishes a target model based on its local operating status, uniformly representing adjustable power, adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient as a flexible feasible region. This region is then compressed into a flexible capability certificate containing constraint representation, state summary, and validity information, which is reported to the coordination controller. This solves the problem in existing technologies where the flexible capability description dimension is singular and upper-level scheduling cannot accurately assess the true flexible boundary. The coordination controller, based on the certificates from each computing center… After global optimization, the adjustment target is issued. The edge autonomous controller does not execute it mechanically, but combines the real-time operating status and target model to perform local optimization solutions and security verification. Before execution, the feasibility of the adjustment target under the current actual state is verified, and during the execution process, a hierarchical response is achieved by calculating the flexible safety margin in real time. This allows for adaptive adjustment when encountering sudden disturbances such as equipment failure, sudden increase in tasks, or environmental changes, avoiding the risk of exceeding the boundary caused by the time delay between the submission of credentials and the arrival of the target. Ultimately, without exceeding the boundaries of server temperature, data center thermal safety, energy storage charge status, power supply reliability, and computing power service quality, the computing power center can reliably respond to various types of power grid demands such as peak shaving, backup, frequency regulation, and virtual inertial support.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a block diagram of a hierarchical control system for a computing center virtual power plant based on a flexible feasible domain, according to an embodiment of this application. Figure 2 This is a schematic diagram of the hierarchical control system architecture for a computing center virtual power plant provided according to an embodiment of this application; Figure 3 This is a schematic diagram of the flexible capability certificate generation process provided according to an embodiment of this application; Figure 4 This is a schematic diagram of the construction of a flexible feasible domain polyhedron according to an embodiment of this application; Figure 5 This is a schematic diagram of the regional coordination aggregation layer rolling optimization and local execution closed-loop process provided according to the embodiments of this application; Figure 6This is a schematic diagram of the digital twin security verification and flexible security margin alarm process provided according to the embodiments of this application; Figure 7 This is a schematic diagram of the virtual inertial response power decomposition process provided according to an embodiment of this application; Figure 8 This is a schematic diagram of the hierarchical control method for a computing center virtual power plant based on a flexible feasible domain, according to an embodiment of this application. Figure 9 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0022] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0023] To improve the clarity and feasibility of the technical solution, the flexible feasible region in this application is represented as P={x|Ax≤b,g(x,s)≤0}, where x is a multidimensional flexible capability vector, A is a constraint matrix obtained by boundary search and polyhedral fitting, b is a boundary vector, s is the current operating state of the computing center, and g(x,s) is a safety constraint function formed by the server thermal model, cooling system model, energy storage model, and task queuing model. The coordination layer generates adjustment targets only when x belongs to P, and the edge layer performs projection verification again based on the latest state before execution, thus forming a closed-loop control chain of "credential constraint - rolling optimization - local projection - feedback update".

[0024] The following describes a hierarchical control system for a computing center virtual power plant based on a flexible feasible domain, according to an embodiment of this application, with reference to the accompanying drawings.

[0025] Figure 1 This is a block diagram of a hierarchical control system for a computing center virtual power plant based on a flexible feasible domain, according to an embodiment of this application.

[0026] like Figure 1 As shown, the hierarchical control system 10 of the computing center virtual power plant based on the flexible feasible domain includes: a cloud server 100, a computing center coordination controller 200, an edge autonomous controller 300, and a terminal device controller 400.

[0027] The cloud server 100 receives scheduling requests from the power grid dispatch center and sends these requests to the computing center coordination controller 200. It then converts the scheduling requests into structured scheduling tasks containing target power, response start time, duration, ramp constraints, and auxiliary service types before sending them to the computing center coordination controller. The computing center coordination controller 200 is communicatively connected to the cloud server 100 and at least one edge autonomous controller 300 in the computing center. It receives flexible capability credentials reported by each edge autonomous controller 300, performs rolling optimization with feasible domain boundary constraints on the structured scheduling tasks based on the constraint matrix, boundary vector, state summary, and validity information in the flexible capability credentials, generates adjustment targets for future scheduling in the time domain, and sends these targets to the corresponding edge autonomous controllers in the computing center. The edge autonomous controllers 300 are deployed in each computing center and are used to establish target models describing the coupling relationship between load, cooling system, energy storage system, and data center thermal environment based on the operating status, constructing a system for adjusting power, etc. A multi-dimensional flexible capability vector, composed of adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient, is used to generate the current flexible feasible region based on the target model and the multi-dimensional flexible capability vector. This flexible feasible region is then compressed into a flexible capability certificate and reported to the computing center coordination controller 200. The flexible capability certificate includes the constraint representation, state summary, validity information, and integrity verification information of the flexible feasible region. Terminal device control instructions are generated based on the adjustment target, operating status, and target model. Terminal device controllers 400 are deployed at the server clusters, cooling equipment, energy storage converters, and power distribution equipment in each computing center. They execute the terminal device control instructions issued by the edge autonomous controller and feed back the corresponding actual response power, equipment status, and computing power service quality indicators to the edge autonomous controller. This allows the edge autonomous controller 300 to perform online correction of the target model based on the actual response power, equipment status, and computing power service quality indicators, and update the flexible feasible region and flexible capability certificate for the next cycle.

[0028] It is understood that the embodiments of this application can achieve a hierarchical decomposition and reliable closed loop from grid dispatch demand reception to terminal device execution in the virtual power plant of the computing center through a four-layer hierarchical architecture consisting of a cloud server, a computing center coordination controller, an edge autonomous controller, and a terminal device controller: The edge autonomous controller establishes a target model based on the local operating status, and uniformly represents the adjustable power, adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient as a flexible feasible region, and compresses it into a flexible capability certificate containing constraint representation, state summary, and validity information, which is then reported to the coordination controller. This solves the problem of the single dimension of flexible capability description and the difficulty of upper-level scheduling to accurately assess the real flexible boundary in the prior art; the coordination controller performs global optimization based on the certificates of each computing center. The edge autonomous controller does not mechanically execute the issued adjustment targets, but rather performs local optimization and security verification by combining real-time operating status and target models. Before execution, it verifies the feasibility of the adjustment targets under the current actual conditions, and during execution, it achieves a graded response by calculating flexible safety margins in real time—normal execution, proactive contraction, or reporting requests for redistribution. This allows it to adaptively adjust when encountering sudden disturbances such as equipment failure, sudden increase in tasks, or environmental changes, avoiding the risk of exceeding boundaries caused by state changes due to time delays between credential reporting and target arrival. Ultimately, without exceeding the boundaries of server temperature, data center thermal safety, energy storage charge status, power supply reliability, and computing power service quality, it enables the computing center to reliably respond to various types of grid demands, such as peak shaving, backup, frequency regulation, and virtual inertial support.

[0029] It should be noted that, as Figure 2 As shown, the cloud server is responsible for exchanging day-ahead, intraday, and real-time dispatch information with the power grid dispatch center or electricity market platform. The computing center coordination controller is responsible for receiving flexible capability credentials from multiple computing centers, performing aggregate modeling and rolling optimization. The edge autonomous controller is deployed locally at each computing center and is responsible for status acquisition, flexible feasible domain generation, digital twin security verification, local model predictive control, and fault self-offset. Terminal device controllers include physical devices such as servers or cluster managers, task schedulers, cooling controllers, energy storage converters, backup power supplies, power distribution monitoring devices, temperature and humidity sensors, and meters, without specific limitations.

[0030] The flexible feasible region is a set of multi-dimensional flexible capability vectors that the computing center can safely realize under a given state vector and constraint set. The multi-dimensional flexible capability vectors include at least adjustable power, adjustable power, response time, duration, ramp rate, response reliability, virtual inertia constant, and virtual damping coefficient.

[0031] The dispatch requirements include: the total net power target of virtual power plants in each time period, the power tracking error threshold, the minimum requirement for increasing / decreasing reserve capacity, the minimum requirement for aggregated virtual inertia constant and virtual damping coefficient, the power ramp rate constraint, the regulation response time and duration constraint, and the state of charge constraint of energy storage terminals.

[0032] like Figure 3 As shown, a flexible capability credential includes a certificate header, capability payload, and verification payload. The certificate header includes the site identifier, certificate serial number, generation time, effective start time, and effective end time. The capability payload includes the constraint matrix, boundary vector, adjustable power range, adjustable power range, maximum ramp rate, shortest response time, longest duration, maximum virtual inertia constant, maximum virtual damping coefficient, reliability indicators, and service impact cost. The verification payload includes the state digest, security margin, confidence level, hash digest, and digital signature or message authentication code. For example, a flexible capability credential can use the following field structure: the certificate header is {site identifier, certificate number, generation time, validity period, version number}, and the capability payload is... The state load is The verification payload consists of {security margin, confidence level, hash digest, signature or message authentication code}.

[0033] like Figure 4 As shown, the flexible feasible region is defined as follows: ; Where i represents the i-th computing center, and t represents the current scheduling time. Let be the multidimensional flexible feasible region of the i-th computing center at time t. Let be the multidimensional flexibility capability vector of the i-th computing center at time t. Let be the linear constraint matrix of the flexible feasible region of the i-th computing center at time t. Let be the boundary vector corresponding to the linear constraint at time t. Let be the current state vector of the i-th computing center at time t. These are local control action variables for the computing power center.

[0034] in,

[0035] In the formula, To reduce net electricity consumption or to provide an equivalent increase in power to the grid, To increase net electricity consumption or the equivalent power reduction to the grid, The response time from receiving the command to reaching 90% of the target power. For the duration that is sustainable under the target adjustment range, Power ramp rate, This represents the confidence level that the capability can be fulfilled under uncertain disturbances. It is a virtual inertial constant. This is the virtual damping coefficient. The cost of the impact on the quality of computing services (expressed on a normalized scale, such as 0 to 1). The original set of nonlinear constraints includes at least thermal safety, energy storage, power supply redundancy, cooling capacity, task latency, and throughput constraints.

[0036] In this embodiment, the edge autonomous controller is further configured to: acquire the current operating status information and security constraints of the edge autonomous controller, wherein the current operating status information includes at least server load rate, deferred task queue, data center temperature, server chip temperature, cooling equipment operating status, energy storage state of charge, and grid frequency measurement; and the security constraints include at least temperature upper limit, energy storage state of charge upper and lower limits, power supply capacity constraint, network latency constraint, and task completion constraint; using the operating status information as the starting benchmark for boundary search, with the goal of maximizing the value of the flexibility dimension, in multiple... In the space formed by the dimensional flexibility capability vectors, the limit boundary points of each dimension that satisfy the safe operation constraints are traversed and searched. Using the limit boundary points of each dimension as initial vertices, multiple flexibility capability dimensions are combined and traversed to eliminate candidate points that do not satisfy the safe operation constraints, resulting in combined boundary vertices that satisfy the safe operation constraints. Based on the limit boundary points of each dimension and the combined boundary vertices, all boundary points obtained from the traversal search are used as fitting data, and polyhedral fitting processing is performed on the fitting data to obtain the target multidimensional flexible feasible region, which is jointly defined by the constraint matrix and the boundary vectors.

[0037] In one implementation, the step size of the boundary search is adaptively determined based on the scheduling time domain length and the model prediction error: when the prediction error increases or state fluctuations intensify, the boundary search step size is reduced and the validity period of the credentials is shortened; when the prediction error is less than a preset error threshold and the state is stable, the boundary search step size is increased to reduce computational load. This balances real-time performance and boundary accuracy, avoiding execution risks caused by an excessively wide flexible feasible region, or waste of adjustable resources due to an excessively narrow one.

[0038] It is understood that in this embodiment, the edge autonomous controller obtains the current operating status information and safety constraints, uses the operating status information as the starting benchmark for boundary search, traverses and searches for limit boundary points and combined boundary vertices in each dimension in the flexible capability space, and fits all boundary points into a polyhedral feasible region, thus achieving an accurate characterization of the multidimensional flexible capability boundary of the computing center. This scheme uses the current actual state as the benchmark and safety constraints as hard boundaries for capability space search, ensuring that the generated flexible feasible region truly reflects the actual adjustment capability of the computing center under the current operating conditions, avoiding the problem of capability boundaries deviating from reality due to static preset parameters or simplified models. At the same time, by combining single-dimensional limit search with combined traversal, the nonlinear coupling relationship between each flexible capability dimension is fully captured, so that the finally fitted polyhedron not only covers the key boundary features of the real feasible region, but also maintains the simplicity of the linear inequality description, making it easy for the upper-level coordination controller to directly call for solution, thereby achieving a balance between ensuring boundary characterization accuracy and control computation efficiency.

[0039] It should be noted that, as Figure 5 As shown, the edge autonomous controller performs sampling during each certificate update cycle. First, several key dimensions are selected, such as... , , , , and Then, the constrained optimization problem is solved by maximizing a single dimension to obtain the single-dimensional limit point; next, one dimension is fixed as a predetermined proportion of its limit value, and the other dimension is optimized to obtain the combined vertex; subsequently, perturbation sampling is performed in the neighborhood of the limit point, and the feasibility is verified using a digital twin model; finally, convex hull fitting, polyhedral body approximation, or support function approximation are used to obtain... and For scenarios with limited computing resources, a pre-trained proxy model can be used to output boundary points, and then the rule verification module can filter out infeasible points.

[0040] In this embodiment, the edge autonomous controller further includes a verification module.

[0041] The verification module is used by the edge autonomous controller to input the current operating status into the target model before generating control instructions for terminal devices based on the adjustment target. The target model outputs the predicted values ​​corresponding to the execution of the adjustment target. The predicted values ​​include: data center temperature, server chip temperature, energy storage state of charge, network latency, and task completion amount. The target model includes: server thermal model, cooling system model, energy storage model, and task queuing model. If any predicted value reaches or exceeds the corresponding preset safety boundary, the flexible capability vector corresponding to the adjustment target is projected to the nearest feasible point in the current flexible feasible domain according to the preset distance metric, and the terminal device control instructions are regenerated based on the nearest feasible point. If, after multiple iterations of projection, the predicted value still exceeds the safety boundary, or the deviation between the nearest feasible point and the adjustment target exceeds the allowable deviation threshold, the adjustment target is rejected, and the corrected flexible feasible domain is re-reported to the computing center coordination controller.

[0042] Among them, the preset safety boundary and preset threshold can be set according to actual needs or experience values, without specific limitations.

[0043] It is understood that the target model in this application, composed of a server thermal model, a cooling system model, an energy storage model, and a task queuing model, simulates and predicts multi-dimensional operating indicators such as room temperature, chip temperature, energy storage state of charge, network latency, and task completion volume after the execution of the adjustment target based on the real-time operating status of the data center before generating equipment control commands. When any indicator touches the safety boundary, the flexible capability vector is automatically iterated and corrected to the nearest boundary of the flexible feasible domain until all predicted indicators meet the safety limits. If multiple adjustments still fail to meet the standards, the scheduling target is directly rejected and the corrected flexible feasible domain is updated and uploaded to the computing power center coordination controller. This achieves pre-emptive safety verification before the issuance of scheduling commands, avoiding safety and business default risks such as data center overheating, energy storage exceeding limits, and computing power service latency exceeding standards from the source. It ensures that the computing power center can safely respond to the grid scheduling needs without breaking the constraints of heat, energy storage, and computing power services. At the same time, it provides real-time updated reliable and adjustable capability boundaries for upper-layer aggregation optimization, improving the overall control reliability and performance security of the virtual power plant.

[0044] It should be noted that, as Figure 6 As shown, the digital twin security verification module is used to simulate and predict the room temperature, server chip temperature, energy storage state of charge, network latency and task completion amount after execution based on the server thermal model, cooling system model, energy storage model and task queuing model before executing the target issued by the regional coordination aggregation layer. If any predicted value exceeds the preset security boundary, the target flexible vector is projected, reduced or rejected, and the shrunken flexible feasible domain is re-reported.

[0045] Specifically, the edge autonomous controller performs digital twin security verification upon receiving the target. The digital twin model predicts the room temperature for several control steps after the target is executed. Chip temperature Energy storage state of charge Task queuing delay Business completion volume And power supply redundancy N+1 state. If the prediction result satisfies... , , , If the completed business volume is not lower than the minimum threshold, then execution will proceed; otherwise, the target multidimensional flexible capability vector will be projected to the nearest feasible domain boundary, or execution rejection and capability degradation reporting will be triggered.

[0046] Construction and operation mechanism of digital twin security verification module: The digital twin security verification module adopts a hybrid modeling method of physical mechanism and data-driven approach to construct the twins of each subsystem.

[0047] The specific details of each sub-model are as follows: Server thermal model: Based on RC thermal network equations: ; in, For chip thermal capacity, For equivalent thermal resistance, Real-time power consumption of the server Real-time temperature of server chips. This provides the real-time temperature of the computer room environment. The parameter is identified online using historical operational data.

[0048] Cooling system model: A simplified heat exchanger model is used. ; in, This refers to the return water temperature. For cold water supply temperature, This refers to the mass flow rate of water in the chilled water pipeline. The specific heat capacity of water at constant pressure. This represents the real-time cooling capacity carried away from the computer room by the chiller units. It describes the dynamics of the chiller units, cooling towers, pumps, and valves, and correlates them with changes in the computer room temperature. ,in, Real-time temperature of the computer room environment. This refers to the real-time cooling capacity carried away by the chiller unit from the computer room. This represents the total heat load inside the computer room. The equivalent heat capacity of the entire computer room.

[0049] Energy storage model: A first-order equivalent circuit model is adopted, considering charging / discharging efficiency. The SOC dynamics are as follows: ; in, Let t be the current state of charge of the energy storage battery. The current state of charge of the energy storage battery at time t+1 is... For energy storage charging efficiency, For energy storage and discharge efficiency, For energy storage charging power, For energy storage discharge power, The rated total capacity of the energy storage system This is the simulation / control step time interval.

[0050] Task queuing model: The M / G / 1 queuing model is used as an approximation to predict task latency within the current and several future 15-second steps. and completion amount .

[0051] The simulation step size of the digital twin module is set to match the predictive control cycle of the local model (1–5 seconds), and the prediction time domain length is 6–12 control cycles. Upon receiving a target instruction from the regional coordination aggregation layer, the module uses the state summary from the reported certificate as initial conditions and performs rolling simulations to predict the evolution trajectory of each state variable after executing the target instruction. If the prediction results satisfy the safety boundary and the safety margin is greater than a preset threshold throughout the entire prediction time domain, execution is allowed; otherwise, the module automatically projects the target vector to the current feasible domain boundary. The projection operator is defined as... ,in, Z represents the original multidimensional flexible adjustment command vector issued by the upper-level regional coordination controller, which includes multidimensional adjustment requirements such as adjustable power, adjustable power, response time, and virtual inertia constant. Z represents any legal multidimensional flexible capability vector within the feasible region. Let i be the multidimensional flexible feasible region of the i-th computing center at time t. The corrected vector for the legitimate flexible adjustment command is the optimized output; and a new target after projection is issued. If no feasible solution can be satisfied after projection, execution is directly rejected and a capability degradation signal is reported, requesting the regional coordination aggregation layer to re-optimize the allocation.

[0052] Furthermore, the digital twin model utilizes measured feedback (power, temperature, SOC, latency, etc.) after each execution cycle for online calibration, and employs recursive least squares to update model parameters, ensuring that the deviation between the twin and the actual system converges over time, thereby guaranteeing the reliability of security verification. The version number of the digital twin model is bound to the certificate validity period; when the certificate is updated, the model state is reset synchronously to ensure that the initial conditions of each cycle are consistent.

[0053] During the execution cycle, the local model predictive controller operates with a cycle of 1 to 5 seconds. Its objective function includes net power tracking error, control action change, service impact cost, temperature safety margin penalty, and energy storage cycle loss. Control variables include server frequency, task migration amount, proportion of deferred tasks paused, chiller start / stop, pump or fan speed, supply and return water temperature setpoints, energy storage charging / discharging power, and backup power status. The controller executes only the first control step, and then recalculates based on the latest sensor feedback. Specifically, after receiving the net power target sequence and flexible target vector sequence, the edge autonomous controller executes local model predictive control with a control cycle of 1 to 5 seconds, decomposing the net power target sequence and flexible target vector sequence into server cluster frequency setpoints, task migration amount, task delay amount, cooling equipment start / stop status, pump or fan speed, supply and return water temperature setpoints, energy storage charging / discharging power, and backup power start / stop strategy; and only executes the first control variable of the current control cycle, resolving it based on the latest status in the next control cycle.

[0054] In this embodiment, the computing center coordination controller is further configured to: obtain the scheduling requirements at each moment within the rolling optimization time domain by using multiple consecutive future moments as the rolling optimization time domain; use the flexible feasible domain corresponding to the constraint matrix and boundary vector in the flexible capability certificate of each computing center as the capability boundary constraint of each computing center; use the equivalent adjustment power of each computing center at each moment and the corresponding response time, duration and ramp rate as decision variables, and under the premise of satisfying the capability boundary constraints of each computing center, the total value of the adjustment target covers the scheduling requirements and each individual adjustment target is located within the corresponding flexible feasible domain, solve for the optimization objective of minimizing the total power tracking deviation, the cost of regulation and auxiliary services and the loss of computing power business service quality, and maximizing the benefits of power market regulation, and distribute the solved adjustment objectives of each computing center to the corresponding edge autonomous controller.

[0055] It is understood that in this embodiment, the computing center coordination controller conducts collaborative scheduling based on multi-time-period rolling optimization in the time domain. Relying on the constraint matrix and boundary vector carried by the flexible capability certificate of each computing center, it accurately delineates the adjustable capability boundary of a single site. Using the output power of each computing center at each time step as the optimization variable, it simultaneously completes multi-objective collaborative optimization while ensuring that the adjustable range constraints of each site and the aggregated control index are not lower than the grid dispatch requirements. It takes into account minimizing the total power tracking deviation, the cost of ancillary service control and the loss of computing power business service quality, and maximizing the control benefits of virtual power plants participating in the electricity market. Finally, it outputs the adjustment target adapted to the carrying capacity of each computing center and sends it to the corresponding edge autonomous controller. This can accurately respond to the multi-dimensional dispatch requirements of the grid, reduce the overall output tracking error, reduce the economic loss of control and computing power business, and fully explore the aggregated control potential of multiple computing centers, thereby improving the overall market benefits of virtual power plants and the economy and stability of multi-site collaborative control.

[0056] Specifically, such as Figure 5 As shown, the computing center coordination controller uses the future rolling time domain. To optimize the time domain, solve the following optimization problem: And satisfy the following constraints:

[0057] in, This represents the total net power of the virtual power plant. In order to schedule the target power, For ancillary service costs, For market profits, For the first The service quality of each computing center is affected by losses. These are the weighting coefficients. This is a penalty item for insufficient flexibility margin.

[0058] in, Let be the output power of the i-th computing center at time k. The maximum tracking error allowed by the power grid for total power. Let be the linear constraint matrix of the i-th computing center at time k. Let be the multidimensional flexibility capability vector of the i-th computing center at time k. For boundary vectors, The reserve capacity can be increased for virtual power plant aggregation. To meet the minimum increase in reserve requirements for the power grid, The reserve requirement can be reduced for virtual power plant aggregation. To reduce the power grid's reserve requirements to a minimum, The virtual inertial constant is the overall equivalent virtual inertial constant of the virtual power plant. To meet the minimum required aggregation inertia of the power grid, This represents the overall equivalent virtual damping coefficient of the virtual power plant. This is the minimum damping requirement for the power grid.

[0059] Among these constraints are power balance constraints. Power tracking constraints Flexible feasible region constraints The constraints include standby capacity constraints, virtual inertia and damping capacity constraints, energy storage terminal SOC constraints, and task completion constraints. The optimization results include the net power target for each computing center, adjustments to standby commitments, and virtual inertia and damping setpoints.

[0060] In this embodiment, the edge autonomous controller is further configured to: construct a temporary feasible region at the current moment based on the operating status between two adjacent flexible capability credential updates; calculate the flexible safety margin based on the minimum normalized distance between the flexible capability vector corresponding to the adjustment instruction to be executed and each constraint boundary of the temporary feasible region; and execute the corresponding early warning action based on the flexible safety margin.

[0061] It is understood that in this embodiment, the edge autonomous controller dynamically constructs a temporary feasible domain based on the real-time operating status during the time interval between two adjacent flexible capability credential updates. This temporary feasible domain, relative to the previously reported flexible feasible domain, undergoes adaptive boundary shrinkage based on temperature changes, state of charge changes, cooling equipment status changes, network congestion levels, or task load changes. Furthermore, it calculates the flexible safety margin within the temporary feasible domain based on the adjustment instructions to be executed, and executes corresponding graded early warning actions according to the different levels of the safety margin. This solves the deficiency in traditional solutions where static feasible domains cannot be dynamically adjusted with changes in operating status. By inserting temporary feasible domain construction and margin assessment steps within the update interval, the security monitoring vacuum between two credential updates is filled, enabling the edge autonomous controller to promptly detect risks and take tiered responses when the state deteriorates. When the security margin is sufficient, it executes normally; when the margin drops to the warning range, it proactively shrinks non-critical services to restore security margin; and when the margin reaches the danger range, it immediately reports capability degradation and requests the coordination layer to reallocate. This tiered warning strategy avoids the problems of either over-conservatism or risk out of control caused by a uniform approach, enabling the computing center to maximize its flexible adjustment capabilities within the security boundary while ensuring self-protection capabilities under sudden disturbances.

[0062] Specifically, such as Figure 4 As shown, the edge autonomous controller constructs a temporary flexible feasible region based on the real-time state vector between two adjacent flexible capability credential updates. Current adjustment instructions The formula for calculating the flexible safety margin relative to the temporary feasible region is as follows: ; in, Let be the multidimensional flexible adjustment command vector issued from the upper layer, and j be the index number of the j-th linear constraint in the temporary feasible region. Let j be the row vector of the temporary feasible region constraint matrix. Let j be the row vector of the temporary feasible region boundary vector. The absolute value of the constraint limit. It is a very small positive number.

[0063] based on Calculate adjustment instructions The flexible safety margin, among which for The OK, for The One element, To prevent positive numbers with a denominator of zero; when Continue execution if the value exceeds the safety threshold. A yellow alert is triggered and non-critical business operations are scaled back when the threshold is less than or equal to the safety threshold and greater than the warning threshold. When the threshold is less than or equal to the warning threshold, a red alarm is triggered, the reporting capability is downgraded, and a request is made to the regional coordination aggregation layer to reallocate tasks.

[0064] In this embodiment, the edge autonomous controller is further configured to: if the flexible safety margin is greater than the safety threshold, not generate an alarm and continue to execute the adjustment command normally; if the flexible safety margin is less than or equal to the safety threshold but greater than the warning threshold, generate a first-level alarm and actively adjust the operating strategy to increase the flexible safety margin; if the flexible safety margin is less than or equal to the warning threshold, generate a second-level alarm, report the flexible capability degradation to the computing power center coordination controller, and request the computing power center coordination controller to regenerate the adjustment target.

[0065] It is understood that in this embodiment, the edge autonomous controller executes differentiated hierarchical response strategies based on different levels of the flexible safety margin. When the flexible safety margin is greater than the safety threshold, no alarm is generated and normal execution continues. When the flexible safety margin is between the safety threshold and the warning threshold, a first-level alarm is generated and the operating strategy is actively adjusted to increase the margin. When the flexible safety margin is lower than the warning threshold, a second-level alarm is generated and a capability degradation request is reported to the coordination layer for reallocation. This hierarchical response mechanism changes the simple processing method of single threshold triggering in traditional schemes. By setting two levels of warning lines, the safety threshold and the warning threshold, the execution... The system categorizes security risks into three levels: normal, warning, and danger. Within the warning range, security margins can be restored by proactively adjusting operational strategies locally, preventing premature capability degradation and task interruption. Within the danger range, timely reporting of degradation and requests for redistribution prevent potential out-of-bounds risks from continued execution. This allows the edge autonomous controller to maximize the completion of adjustment tasks while ensuring safety, and proactively seeks support from the coordination layer when risks become uncontrollable. This reduces unnecessary task interruptions and prevents the accumulation and expansion of risks, achieving an optimal balance between security and task completion rate during execution.

[0066] In this embodiment, the edge autonomous controller is further configured to: calculate the virtual inertial response power based on the energy storage state of charge, task queue length, current room temperature and grid frequency measurements in the current operating state; generate virtual inertial response power instructions based on the virtual inertial response power, and decompose and execute corresponding actions according to the priority order of the energy storage system, server load and cooling system.

[0067] In this embodiment, the edge autonomous controller is further configured to: calculate the virtual inertial response power based on the energy storage state of charge, task queue length, current room temperature and grid frequency measurements; generate virtual inertial response power instructions based on the virtual inertial response power, and decompose them into corresponding actions for the energy storage system, server load and cooling system in priority order.

[0068] It is understood that in this embodiment, the edge autonomous controller calculates the virtual inertial response power based on the energy storage state of charge, the length of the deferred task queue, the data center temperature, and the measured grid frequency. This power includes an inertial response component determined by the rate of frequency change and a damping response component determined by the frequency deviation. The virtual inertial response power command is decomposed step-by-step according to the priority order of the energy storage system, server load, and cooling system. This allows the energy storage system to provide rapid power support within milliseconds to seconds via the energy storage converter, the server load to provide secondary support within seconds through dynamic voltage and frequency adjustments and deferred task pauses, and the cooling system to provide slow damping support within minutes without exceeding temperature safety boundaries. This expands the adjustment dimensions of the virtual power plant in the computing center, enabling… The computing center has expanded its services beyond simply participating in slow adjustments such as peak shaving and reserve to include rapid frequency response and virtual inertia support for the power grid, thus enhancing its value realization potential in the power ancillary services market. Simultaneously, power is decomposed according to a priority order of response speed from fastest to slowest, allowing each type of adjustable resource to function optimally based on its physical characteristics. Energy storage systems leverage their millisecond-level response advantage to suppress rapid fluctuations, server loads utilize their second-level adjustability to provide short-term power support, and cooling systems leverage their thermal inertia to handle continuous, slow adjustments. Furthermore, by automatically reducing the virtual inertia constant or damping coefficient setting and resubmitting documentation when detecting insufficient energy storage state of charge, temperatures approaching upper limits, or service quality constraints, the computing center ensures that virtual inertia responses always operate within its safe and feasible domain.

[0069] Specifically, such as Figure 7 As shown, the edge autonomous controller is based on the energy storage state of charge. Delayable task queue length Computer room temperature and power grid frequency measurement value Calculate the virtual inertial response power: ; in, Set a value for the virtual inertia constant. Set the virtual damping coefficient value. This is the virtual inertia scaling factor. This is the virtual damping proportional coefficient. Where f is the rated frequency, and f is the real-time measured frequency of the power grid.

[0070] Decomposed according to priority: energy storage system, server load, and cooling system. This allows the energy storage system to prioritize providing rapid power support, while the server load provides secondary support through dynamic voltage and frequency adjustments and deferred task pauses, and the cooling system provides slow damping support without exceeding temperature safety limits.

[0071] and It should be noted that this must be within the current allowable range of the flexible feasible domain. The response priority is energy storage system, server load, and cooling system. The energy storage system undertakes rapid power injection or absorption at the millisecond to second level; the server load provides second-level support through DVFS, GPU power cap adjustment, and deferred task pause; the cooling system provides minute-level slow damping based on thermal inertia. If insufficient SOC, temperature approaching the upper limit, or service quality constraint risk is detected, the flexible capability certificate will be automatically downgraded and resubmitted.

[0072] The hierarchical control system for a computing center virtual power plant based on a flexible feasible domain, proposed in this application, utilizes a four-layer architecture consisting of a cloud server, a computing center coordination controller, an edge autonomous controller, and a terminal device controller. This architecture achieves a hierarchical decomposition and reliable closed-loop process from receiving grid dispatch requirements to execution by terminal devices. The edge autonomous controller establishes a target model based on its local operating state, uniformly representing adjustable power, adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient as a flexible feasible domain. This domain is then compressed into a flexible capability certificate containing constraint representation, state summary, and validity information, which is reported to the coordination controller. This solves the problems of limited flexibility capability description dimensions and difficulty in accurately assessing the true flexibility boundary in existing technologies. The controller performs global optimization based on credentials from each computing center before issuing adjustment targets. The edge autonomous controller does not execute mechanically, but combines real-time operating status and target model for local optimization and security verification. Before execution, it verifies the feasibility of the adjustment target under the current actual state, and achieves hierarchical response by calculating flexible safety margins in real time during execution. This allows it to adaptively adjust when encountering sudden disturbances such as equipment failure, sudden increase in tasks, or environmental changes, avoiding the risk of exceeding boundaries caused by state changes due to time delay between credential reporting and target arrival. Ultimately, without exceeding the boundaries of server temperature, data center thermal safety, energy storage charge status, power supply reliability, and computing service quality, the computing center can reliably respond to various types of power grid demands such as peak shaving, backup, frequency regulation, and virtual inertial support.

[0073] Figure 8 This is a flowchart illustrating a hierarchical control method for a computing center virtual power plant based on a flexible feasible domain, provided as an embodiment of this application.

[0074] like Figure 8 As shown, the hierarchical control method for a virtual power plant based on a flexible feasible domain is implemented using a hierarchical control system for a virtual power plant based on a flexible feasible domain, and includes the following steps: In step S101, the dispatching requirements issued by the power grid dispatching center are obtained.

[0075] It is understood that in this embodiment, the cloud server serves as the data interface between the system and the power grid dispatch center, receiving dispatch requests from the power grid dispatch center in real time. These dispatch requests include at least information such as target power, reserve capacity requirements, virtual inertia requirements, and auxiliary service types, and are forwarded to the computing center coordination controller as input for optimization calculations. Through the unified access of the cloud server, the power grid dispatch requests are transformed from external instructions to structured data that can be parsed and processed by the internal system, ensuring the integrity, accuracy, and real-time nature of the upper-level dispatch information. This provides an accurate source of tasks and a data foundation for subsequent coordination layer optimization and the generation of adjustment targets for each computing center, solving the problems of information distortion and response delay caused by manual interpretation of dispatch instructions or inconsistent interfaces in existing solutions.

[0076] In step S102, a target model describing various adjustable resources of the computing center is established based on the operating status. A multi-dimensional flexible capability vector consisting of up-adjustable power, down-adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient is constructed. Based on the target model, the flexible feasible region and flexible capability certificate at the current moment are generated. The flexible capability certificate includes at least the constraint representation of the flexible feasible region, state summary, validity information, and integrity verification information.

[0077] Specifically, the flexible capability certificate update cycle can be less than or equal to the scheduling rolling optimization cycle; when the edge autonomous controller detects that the rate of change of the data center temperature, the rate of change of the energy storage charge state, the rate of change of the deferred task queue, or the change in the available capacity of the cooling equipment exceeds the corresponding threshold, even if the preset update cycle has not yet been reached, it will trigger the early update of the certificate and report the capability degradation or capability recovery status.

[0078] It is understood that the embodiments of this application can establish a target model describing the dynamic characteristics of multiple adjustable resources in a computing center based on the real-time collected operating status by the edge autonomous controller. This target model integrates the operating characteristics of heterogeneous subsystems such as computing load, cooling system, energy storage system, power supply system, and data center thermal environment into a unified mathematical description framework, providing a model foundation for subsequent quantitative evaluation of flexibility capabilities. Based on this target model and the current operating status, a flexible feasible domain is generated, realizing a unified characterization of the multi-dimensional flexibility capabilities of the computing center under a given state, including adjustable power, adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient. This solves the problem of relying on a single power capacity or... Fixed reduction curves describe flexible capabilities but fail to reflect the coupling constraints between multidimensional capabilities. Further, the flexible feasible region is compressed into a flexible capability credential containing constraint representation, state summary, validity information, and integrity verification information, and reported to the computing center coordination controller. This encapsulates the complex nonlinear constraints within the computing center into a standardized data format that the coordination layer can directly access. The constraint representation provides a boundary description in the form of a system of linear inequalities for upper-level optimization solutions. The state summary and validity information enable the coordination layer to verify the timeliness and state consistency of the credential. The integrity verification information ensures that the credential has not been tampered with during transmission, thus providing a reliable and verifiable capability data foundation for upper-level scheduling decisions.

[0079] In step S103, based on the flexible capability credentials and scheduling requirements, the adjustment targets of each computing center in the future scheduling time domain are calculated under the constraint that the adjustment targets of each computing center are all located within the corresponding flexible feasible domain. Terminal equipment control instructions are generated according to the adjustment targets, operating status and target model.

[0080] It is understood that the embodiments of this application can, based on the flexible capability credentials reported by each edge autonomous controller and the power grid dispatching requirements forwarded by the cloud server, construct a rolling optimization time domain with multiple consecutive future moments, and use the flexible feasible domain corresponding to the constraint matrix and boundary vector in the flexible capability credentials of each computing center as the hard constraint of the capability boundary. This solves for the adjustment target that minimizes the total power tracking deviation, ancillary service costs, and the impact of computing power service quality on losses while maximizing market revenue. This fundamentally ensures that the adjustment targets of each computing center optimally allocated by the coordination layer will not exceed their actual safety boundaries, solving the defect in the prior art where the lack of feasibility boundary verification leads to dispatching instructions easily exceeding the actual adjustment capabilities of the computing center. After receiving the adjustment target, the edge autonomous controller does not execute it mechanically, but rather combines the real-time operating status and the target model. This approach performs local optimization, transforming abstract adjustment targets into specific terminal device control commands such as server cluster frequency settings, task migration amounts, cooling equipment speed settings, energy storage charging and discharging power, and backup power supply start-stop strategies. This achieves a precise mapping from global optimization targets to local device actions. Furthermore, since local optimization is based on the latest operating state, it can automatically correct deviations caused by time delays or state fluctuations between the targets issued by the coordination layer and the actual local state, ensuring that the generated control commands always match the current operating conditions. The global optimization of the coordination layer and the local optimization of the edge layer form a layered collaboration. The coordination layer ensures global economy and feasibility, while the edge layer ensures local security and executability, solving the deficiency in existing solutions where unified scheduling commands cannot adapt to the differentiated operating states of various computing centers.

[0081] When solving for the adjustment target, the coordinating controller sets flexible capability credentials that have expired, failed integrity verification, or whose status summary is inconsistent with the most recent telemetry status to an unavailable state, and requests the corresponding edge autonomous controller to regenerate the credentials. Before regenerating the credentials, the computing center does not participate in the allocation of new adjustment targets or only participates in the allocation according to the conservative default capability, thereby avoiding invalid credentials from entering the optimization calculation.

[0082] In step S104, after executing the terminal device control command, the target model and the next cycle flexible capability certificate are updated based on the actual response power, service quality impact, and device status.

[0083] It is understood that, after executing the terminal device control command, the target model can be calibrated and updated online using the actual response power, the service quality impact during execution, and the device status after execution as feedback data. Based on the updated model, a flexible capability certificate for the next cycle is regenerated and reported to the computing center coordination controller. This feedback update mechanism enables the target model to continuously correct its parameters using the measured data from each actual execution, allowing the model to continuously approximate the dynamic characteristics of the real system. This solves the model mismatch problem caused by equipment aging, environmental changes, or operating condition drift. At the same time, regenerating the flexible capability certificate based on the updated model realizes a closed-loop iteration between the reported capability and the actual executable capability. This transforms the experience of each execution into an improvement in the accuracy of the next reported capability, eliminating the gap between the reported capability and the actual executable capability in the prior art. This ensures that the capability boundary described by the certificate is always consistent with the current real adjustment capability of the computing center, providing a continuously accurate capability data foundation for the coordination layer's optimization decisions.

[0084] According to the hierarchical control method for virtual power plants in computing centers based on flexible feasible domains proposed in this application, the scheduling requirements issued by the power grid dispatch center are obtained and forwarded to the coordination controller. Simultaneously, the edge autonomous controllers of each computing center establish target models describing the dynamic characteristics of various adjustable resources of the computing center based on real-time operating status. Based on these target models, a flexible feasible domain containing a unified representation of multi-dimensional flexible capabilities is generated. This flexible feasible domain is then compressed into flexible capability credentials containing constraint representations, state summaries, validity information, and integrity verification information, and reported to the coordination controller. This achieves the encapsulation of complex nonlinear constraints of the computing center into a standardized data format that can be directly invoked and verified by the coordination layer. Based on the flexible capability credentials of each computing center and the power grid dispatch requirements, the coordination controller uses the constraint representations in each credential as hard constraints on the capability boundary. It calculates and issues adjustment targets for each computing center that do not exceed its actual safety boundary in the future scheduling time domain, fundamentally ensuring the feasibility of the coordination layer's allocation scheme. After receiving the adjustment targets, the edge autonomous controllers do not mechanically execute them but combine real-time operating status and the target model. Local optimization is performed to transform abstract regulation targets into specific, executable terminal device control commands, such as server cluster frequency settings, task migration amounts, cooling equipment speed settings, energy storage charging and discharging power, and backup power supply start-up and shutdown strategies. This ensures that the generated control commands always match the current operating conditions. After executing the terminal device control commands, the edge autonomous controller performs online calibration and updates to the target model based on the actual response power, service quality impact, and equipment status. Based on the updated model, it regenerates the flexible capability certificate for the next cycle and reports it to the coordination controller. This achieves a closed-loop iteration between reported capabilities and actual executable capabilities, with each execution experience contributing to the improvement of the accuracy of the next reported capability. Through the complete closed-loop process of cloud-based demand acquisition, edge modeling and certificate generation, coordination layer optimization and allocation, local command generation, and post-execution feedback updates, a data-driven, hierarchical, and continuously iterative collaborative working mechanism is formed between each link. This solves the defects of existing technologies, such as inaccurate capability descriptions and infeasible scheduling commands caused by the fragmentation of each link from grid demand to equipment execution and the lack of closed-loop feedback.

[0085] The hierarchical control system for a virtual power plant based on a flexible feasible domain, as described in this application, will now be described in detail with reference to specific embodiments: Example 1: A 20MW computing center is equipped with a 5MW / 10MWh energy storage system, and 30% of batch processing tasks are deferred. The initial state is SOC=60%, data center temperature 22℃, and a deferred task queue of 50TB. The edge autonomous controller calculates the maximum equivalent upscaling power to be approximately 8.5MW, of which energy storage discharge contributes 5MW, server frequency reduction and task delay contribute approximately 3MW, and cooling load optimization contributes approximately 0.5MW. The system generates a flexible capability certificate and participates in frequency response. When df / dt=-0.5Hz / s is detected, if H...set =4.0s, D set =2.0MW / Hz, then the controller calculates the virtual inertial response power and prioritizes the output from the energy storage, the server dynamically reduces the frequency to provide the remaining support, and the cooling system supplements the damping power according to the temperature margin.

[0086] Example 2: Three computing centers form a virtual power plant with a total load of approximately 50MW. The regional coordination aggregation layer commits to peak shaving of 15MW from 18:00 to 19:00 in the day-ahead market. At 17:45, rolling optimization yields peak shaving of 8MW for DC1, 5MW for DC2, and 2MW for DC3. At 18:10, one of the main chillers in DC1 trips. The edge autonomous controller recalculates the temporary feasible region and finds that the maximum peak shaving capacity has decreased from 10MW to 6MW with a negative flexibility safety margin, thus triggering a red alarm and reporting a capacity degradation. The regional coordination aggregation layer re-optimizes, allocating the 2MW shortfall to DC2, ultimately still meeting the total peak shaving commitment of 15MW.

[0087] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0088] When processor 902 executes the program, it implements the hierarchical control method for virtual power plants in computing centers based on flexible feasible domains provided in the above embodiments.

[0089] Furthermore, electronic devices also include: Communication interface 903 is used for communication between memory 901 and processor 902.

[0090] The memory 901 is used to store computer programs that can run on the processor 902.

[0091] The memory 901 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0092] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0093] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0094] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0095] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implements the above-described hierarchical control method for a virtual power plant in a computing center based on a flexible feasible domain.

[0096] This application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described hierarchical control method for a virtual power plant in a computing center based on a flexible feasible domain.

[0097] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0098] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0099] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0100] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0101] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A hierarchical control system for a virtual power plant based on a flexible feasible domain, characterized in that, include: Cloud servers, computing center coordination controllers, edge autonomous controllers, and terminal device controllers; The cloud server is used to receive scheduling requests issued by the power grid dispatch center, and convert the scheduling requests into structured scheduling tasks containing target power, response start time, duration, ramp constraints and auxiliary service types before sending them to the computing power center coordination controller. The computing center coordination controller communicates with the cloud server and at least one edge autonomous controller of the computing center. It is used to receive the flexible capability credentials reported by each edge autonomous controller, and based on the constraint matrix, boundary vector, state summary and validity information in the flexible capability credentials, it performs rolling optimization with feasible domain boundary constraints on the structured scheduling task, generates the adjustment target of each computing center in the future scheduling time domain, and sends the adjustment target to the edge autonomous controller of the corresponding computing center. Edge autonomous controllers are deployed in each computing center to establish a target model describing the coupling relationship between computing load, cooling system, energy storage system, and data center thermal environment based on the operating status. They construct a multi-dimensional flexible capability vector consisting of up-adjustable power, down-adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient. Based on the target model and the multi-dimensional flexible capability vector, they generate the flexible feasible region at the current moment and compress the flexible feasible region into a flexible capability certificate, which is then reported to the computing center coordination controller. The flexible capability certificate includes the constraint representation, state summary, validity information, and integrity verification information of the flexible feasible region. It also generates terminal device control commands based on the adjustment target, operating status, and the target model. Terminal device controllers are deployed at the server clusters, cooling equipment, energy storage converters, and power distribution equipment of each computing center. They are used to execute terminal device control commands issued by the edge autonomous controller and to feed back the corresponding actual response power, device status, and computing power service quality indicators to the edge autonomous controller. This allows the edge autonomous controller to perform online correction of the target model based on the actual response power, device status, and computing power service quality indicators, and to update the flexible feasible domain and flexible capability credentials for the next cycle.

2. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 1, characterized in that, The edge autonomous controller is further used for: The current operating status information and safe operating constraints of the edge autonomous controller are obtained. The current operating status information includes at least the server load rate, the deferred task queue, the data center temperature, the server chip temperature, the operating status of the cooling equipment, the energy storage state of charge, and the power grid frequency measurement. The safe operating constraints include at least the upper limit of temperature, the upper and lower limits of the energy storage state of charge, the power supply capacity constraint, the network latency constraint, and the task completion amount constraint. Using the operational status information as the starting reference for boundary search, and aiming to maximize the value of the flexibility dimension, the limit boundary points of each dimension that satisfy the safe operation constraints are searched in the space formed by the multi-dimensional flexibility vector. Using the limit boundary points of each dimension as initial vertices, a combined traversal is performed on multiple flexible capability dimensions to eliminate candidate points that do not meet the safe operation constraints, thereby obtaining combined boundary vertices that meet the safe operation constraints. Based on the limit boundary points of each dimension and the combined boundary vertices, a polyhedral fitting process is performed using all boundary points obtained through traversal search as fitting data to obtain the target multidimensional flexible feasible region, wherein the target multidimensional flexible feasible region is jointly defined by the constraint matrix and the boundary vector.

3. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 2, characterized in that, The edge autonomous controller also includes a verification module. The verification module is used to input the current operating state into the target model before the edge autonomous controller generates terminal device control instructions according to the adjustment target. The target model outputs the predicted value corresponding to the execution of the adjustment target. The predicted value includes: data center temperature, server chip temperature, energy storage state of charge, network latency, and task completion amount. The target model includes: server thermal model, cooling system model, energy storage model, and task queuing model. If any predicted value reaches or exceeds the corresponding preset safety boundary, the flexible capability vector corresponding to the adjustment target is projected to the nearest feasible point in the current flexible feasible domain according to the preset distance metric, and the terminal device control command is regenerated based on the nearest feasible point. If, after multiple iterations of projection, the predicted value still exceeds the safety boundary, or the deviation between the nearest feasible point and the adjustment target exceeds the allowable deviation threshold, then the adjustment target is rejected, and the corrected flexible feasible domain is re-reported to the computing center coordination controller.

4. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 1, characterized in that, The computing center coordination controller is further used for: The scheduling requirements at each moment within the rolling optimization time domain are obtained by constructing a rolling optimization time domain from multiple consecutive future moments. The flexible feasible region corresponding to the constraint matrix and boundary vector in the flexible capability certificate of each computing center is used as the capability boundary constraint of each computing center. Using the equivalent regulation power of each computing center at each time point and the corresponding response time, duration and ramp rate as decision variables, and under the premise of satisfying the capability boundary constraints of each computing center, the total value of the regulation target covers the scheduling requirements, and each individual regulation target is located within the corresponding flexible feasible domain, the optimization objective is to minimize the total power tracking deviation, the cost of regulation auxiliary services and the loss of computing power business service quality, and maximize the benefits of power market regulation. The obtained regulation targets of each computing center are then distributed to the corresponding edge autonomous controller.

5. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 1, characterized in that, The edge autonomous controller is further used for: Between two consecutive flexible capability credential updates, a temporary feasible domain is constructed based on the operational status at the current moment; The flexible safety margin is calculated based on the minimum normalized distance between the flexible capability vector corresponding to the adjustment instruction to be executed and each constraint boundary of the temporary feasible domain. Execute corresponding early warning actions based on the flexibility of the safety margin.

6. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 5, characterized in that, The edge autonomous controller is further used for: If the flexible safety margin is greater than the safety threshold, no alarm will be generated, and the adjustment command will continue to be executed normally. If the flexible safety margin is less than or equal to the safety threshold and greater than the warning threshold, a first-level alarm is generated, and the operating strategy is proactively adjusted to increase the flexible safety margin. If the flexible safety margin is less than or equal to the warning threshold, a second-level alarm is generated, and a flexible capability degradation is reported to the computing power center coordination controller, requesting the computing power center coordination controller to regenerate the adjustment target.

7. The hierarchical control system for a virtual power plant based on a flexible feasible domain according to claim 5, characterized in that, The edge autonomous controller is further used for: The virtual inertial response power is calculated based on the energy storage state of charge, the length of the delayed task queue, the current room temperature, and the measured grid frequency. The virtual inertia response power command is generated based on the virtual inertia response power, and the corresponding actions are decomposed and executed according to the priority order of the energy storage system, server load, and cooling system.

8. A hierarchical control method for a computing center virtual power plant based on a flexible feasible domain, characterized in that, The method is applied to the implementation of the hierarchical control system for a virtual power plant based on a flexible feasible domain, as described in claims 1-7, wherein the method includes the following steps: Obtain dispatch requests issued by the power grid dispatch center; Based on the operating status, a target model describing various adjustable resources of the computing center is established. A multi-dimensional flexible capability vector is constructed, consisting of up-adjustable power, down-adjustable power, response time, duration, ramp rate, virtual inertia constant, and virtual damping coefficient. Based on the target model and the multi-dimensional flexible capability vector, the flexible feasible region at the current moment is generated. The flexible capability certificate at least includes the constraint representation, state summary, validity information, and integrity verification information of the flexible feasible region. Based on the flexible capability credentials and the scheduling requirements, under the constraint that the adjustment targets of each computing center are all located within the corresponding flexible feasible domain, the adjustment targets of each computing center in the future scheduling time domain are calculated, and terminal device control instructions are generated according to the adjustment targets, operating status and target model. After executing the terminal device control command, the target model and the next cycle's flexible capability certificate are updated based on the actual response power, service quality impact, and device status.

9. An electronic device, comprising: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to perform the hierarchical control method for a virtual power plant based on a flexible feasible domain as described in claim 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they are used to implement the hierarchical control method for a virtual power plant based on a flexible feasible domain as described in claim 8.

Images