Method and device for collaborative optimization scheduling of computing and power consumption considering timing reconstruction of computing tasks

Abstract

The application provides a kind of computing power task timing reconstruction's electric carbon collaborative optimization scheduling method and device, belong to power system optimization operation and dispatching technical field.The method includes: obtaining multi-source system operation data;According to the multi-source system operation data, a multi-source system model is constructed, and the constraint condition of multi-source system model is determined;Based on the multi-source system model, the network security constraint considering dynamic reactive power support is constructed, and the network security default penalty function is determined;Based on the multi-source system operation data, multi-source system model and network security default penalty function, a multi-objective optimization function that economic, low carbon and new energy consumption capacity are mutually coordinated is constructed;Improved grey wolf optimization algorithm is used for solving.The application can improve the global exploration ability, convergence precision and feasible solution acquisition rate of electric carbon collaborative optimization scheduling, which is conducive to ensuring that the main network voltage is stable in the safe range while realizing the significant reduction of system comprehensive operation cost and carbon emissions.

Description

Technical Field

[0001] This application relates to the field of power system optimization operation and dispatching technology, and in particular to a computing-power-carbon collaborative optimization dispatching method and apparatus that takes into account the timing reconstruction of computing power tasks. Background Technology

[0002] With the advancement of the digital economy, the continuous load growth of large-scale computing centers and the power output fluctuations of high-proportion wind and solar renewable energy sources have profoundly changed the supply-demand balance and dispatch structure of the power system. Pumped storage is a key energy storage resource that combines peak shaving, reserve, and renewable energy consumption support. At the same time, computing centers, due to the time-limit flexibility of some of their computing tasks, can provide dispatchable potential through time-series reconfiguration. Together, these two constitute an important flexible resource for the coordinated dispatch of power generation, grid, load, and storage in the new power system.

[0003] In related technologies, scheduling methods for interaction between computing centers, pumped storage power stations and the power grid typically treat computing centers as equivalent to conventional, movable loads and incorporate them into the power grid scheduling model. With the goal of optimizing system economy, heuristic solution algorithms often employ simplified networks or "post-verification" mechanisms. For example, candidate scheduling schemes are first searched using heuristic algorithms, and then node voltages, branch power, etc. are verified. Schemes that exceed limits are eliminated or additional penalty terms are added.

[0004] However, the above-mentioned technical solutions have the following shortcomings: First, equating the computing center with a conventional movable load makes it difficult to accurately define the temporal reconstruction boundary of the computing task between different time periods, and it is difficult to fully quantify the complementary and coordinated relationship between pumped storage energy transfer across time periods and computing center task reconstruction to achieve deep synergy between source, storage, and load. Second, existing heuristic solution algorithms may lead to a large number of candidate solutions being eliminated due to exceeding the limits, resulting in a low rate of obtaining feasible solutions. Third, existing methods are mostly dominated by single economic considerations in the construction of objective functions, lacking coordination between multiple objectives. Moreover, the solution algorithms have inherent defects such as uneven initial distribution, limited convergence ability, and lack of active physical constraint repair mechanisms under high-dimensional multi-constraint conditions, which limits their engineering application in actual power grid dispatching scenarios. Summary of the Invention

[0005] This application provides a computing-power-carbon collaborative optimization scheduling method and apparatus that takes into account the timing reconstruction of computing tasks. This method can improve the global exploration capability, convergence accuracy, and feasible solution acquisition rate of the computing-power-carbon collaborative optimization scheduling, which is beneficial for significantly reducing the overall system operating cost and carbon emissions while ensuring the main grid voltage remains stable within a safe range. The technical solution includes at least the following: Firstly, a computational-power-carbon co-optimization scheduling method considering computational task time-series reconfiguration is provided, comprising: acquiring multi-source system operation data; constructing a multi-source system model based on the multi-source system operation data and determining the constraints of the multi-source system model, wherein the multi-source system model is used to characterize the boundary of computational task time-series reconfiguration based on latency characteristics of the computing center, the operation characteristics of pumped storage power stations, and the output characteristics of thermal power and new energy sources; constructing network security constraints considering dynamic reactive power support based on the multi-source system model and determining a network security violation penalty function; constructing a multi-objective optimization function that coordinates economic efficiency, low carbon emissions, and new energy absorption capacity based on the multi-source system operation data, the multi-source system model, and the network security violation penalty function; solving the optimal scheduling scheme using an improved gray wolf optimization algorithm, wherein the improved gray wolf optimization algorithm performs physical constraint repair on the position vector of gray wolf individuals based on the constraints, and performs a comprehensive fitness evaluation on the repaired position vector of gray wolf individuals based on the network security constraints, the network security violation penalty function, and the multi-objective optimization function.

[0006] Optionally, the multi-source system model includes a computing center task time-series reconstruction model, a pumped storage power station reservoir energy evolution model, and a thermal power and new energy output model; the constraints include operating rules constraints for the computing center, operating rules constraints for the pumped storage power station, and operating rules constraints for thermal power and new energy.

[0007] Optionally, the expression for the network security constraint is:

[0008]

[0009]

[0010]

[0011] in, and They are nodes During the period The net active power injection and net reactive power injection, For the set of power grid nodes, and They are nodes and nodes During the period The node voltage amplitude, and They are nodes ,node The conductivity and susceptance between them For nodes With nodes During the period The phase angle difference, and They represent the lines respectively. In the nodes To node , by node To node Apparent power in both directions, For the line Maximum permissible apparent power For the collection of transmission lines, For the scheduling period, and They are nodes The lower and upper limits of the voltage. For thermal power units During the period Unproductive efforts and thermal power units The lower and upper limits of reactive power output, For the collection of thermal power units, For nodes Dynamic reactive power compensation device during time period The actual reactive power output, and These represent the lower and upper limits of reactive power output allowed by the dynamic reactive power compensation device, respectively. For nodes Dynamic reactive power compensation device during time period The target reactive power output is obtained based on the current voltage deviation. For the limited projection operator, This is the voltage deviation proportional control coefficient. The nodes obtained for AC power flow calculation During the period The node voltage amplitude, This is the reference value for node voltage. For nodes Dynamic reactive power compensation device during time period Reactive power output after smoothing To ensure smooth coefficient updates, the expression for the network security breach penalty function is as follows:

[0012]

[0013] in, To indicate the line During the period Apparent power Indicates the degree to which the node voltage exceeds the limit. Indicates the degree to which the line capacity has exceeded its limit. This indicates the degree to which the reactive power boundary of a thermal power unit is exceeded. For cybersecurity breach of contract penalties, , , and These are the penalty weights for exceeding node voltage limits, exceeding line capacity limits, exceeding reactive power output limits, and being infeasible for power flow. This represents the number of time periods during which a feasible solution for the current trend was not obtained.

[0014] Optionally, the multi-objective optimization function is expressed as:

[0015] in, Indicates the overall optimization objective. For the comprehensive operating costs of multi-source systems, For carbon dioxide emissions from multi-source systems, This refers to unused renewable energy power.

[0016] Optionally, the step of using an improved gray wolf optimization algorithm to solve the problem includes: setting initial parameters for the improved gray wolf optimization algorithm; encoding decision variables based on the multi-source system operating data and the multi-source system model, and introducing Sobol low-discrepancy sequences to generate initial gray wolf population position vectors; performing physical constraint repair on the position vectors of individual gray wolves based on the constraints; calculating the comprehensive fitness of the repaired position vectors of individual gray wolves based on the network security constraints, the network security breach penalty function, and the multi-objective optimization function; and selecting the three gray wolf individuals with the lowest comprehensive fitness and marking them as... Wolf, wolves and Wolves, and mark the remaining gray wolf individuals as Wolf; based on Wolf, wolves and The wolf's position vector is updated using a nonlinear cosine adaptive convergence factor. The wolf's position vector; determine if the maximum number of iterations has been reached. If not, proceed with a new round of physical constraint repair, comprehensive fitness evaluation, and nonlinear adaptive position update; otherwise, output the current position vector. The wolf's position vector is used as the optimal scheduling scheme.

[0017] Optionally, the formula for calculating the overall fitness is:

[0018] in, The position vector of the individual gray wolf Overall adaptability , , The weights for economic efficiency, low carbon emissions, and renewable energy consumption targets are respectively. , , These are the position vectors of individual gray wolves. The corresponding comprehensive operating costs of multi-source systems, carbon dioxide emissions, and unused renewable energy power. The position vector of the individual gray wolf Corresponding cybersecurity breach penalties , , These are the comprehensive operating costs, carbon dioxide emissions, and unconsumed renewable energy power of the preset reference scheme.

[0019] Optionally, the expression for the nonlinear cosine adaptive convergence factor is:

[0020] in, Indicates the current iteration number. The maximum number of iterations, For the first Nonlinear cosine adaptive convergence factor in the next iteration.

[0021] Secondly, a computing-power-carbon collaborative optimization scheduling device considering computing power task time-series reconfiguration is provided, comprising: an acquisition module for acquiring multi-source system operation data; a first construction module for constructing a multi-source system model based on the multi-source system operation data and determining the constraints of the multi-source system model, wherein the multi-source system model is used to characterize the computing power center's computing power task time-series reconfiguration boundary based on latency characteristics, the operation characteristics of pumped storage power stations, and the output characteristics of thermal power and new energy sources; and a second construction module for constructing network security constraints considering dynamic reactive power support based on the multi-source system model and determining network security breach penalties. The system includes a penalty function; a third construction module, which constructs a multi-objective optimization function that coordinates economic efficiency, low carbon emissions, and new energy absorption capacity based on the multi-source system operation data, the multi-source system model, and the network security breach penalty function; and a solution module, which uses an improved gray wolf optimization algorithm to solve the problem and obtain the optimal scheduling scheme. The improved gray wolf optimization algorithm performs physical constraint repair on the position vectors of individual gray wolves based on the constraints, and performs a comprehensive fitness evaluation on the repaired position vectors of individual gray wolves based on the network security constraints, the network security breach penalty function, and the multi-objective optimization function.

[0022] Thirdly, a computer device is provided, the computer device comprising: a memory and a processor, wherein the memory stores at least one computer program, the at least one computer program being loaded and executed by the processor to realize the aforementioned computational-electricity-carbon collaborative optimization scheduling method considering computational power task timing reconstruction.

[0023] Fourthly, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to realize the aforementioned computational-electricity-carbon collaborative optimization scheduling method that takes into account the timing reconstruction of computing power tasks.

[0024] The beneficial effects of the technical solution provided in this application include at least the following: In this embodiment, a multi-source system model is constructed to accurately define the feasible domain of computing power task time-series reconstruction, providing a decision boundary basis for the effective release of flexible resources in the computing power center. This model can precisely characterize the cross-time period energy transfer operation characteristics of pumped storage power stations and the computing power task time-series reconstruction boundary, achieving deep spatiotemporal synergy between high-energy-consuming computing loads and energy storage resources. This is beneficial for improving the absorption rate of new energy and smoothing system net load fluctuations. By constructing network security constraints that consider dynamic reactive power support and determining corresponding network security violation penalty functions, a multi-objective optimization function is constructed that coordinates economic efficiency, low carbon emissions, and new energy absorption capacity. An improved gray wolf optimization algorithm is used for solving this function. Based on the constraints, the algorithm actively repairs the physical constraints of individual gray wolves and performs a comprehensive fitness evaluation. This overcomes the shortcomings of the standard algorithm, such as uneven initial distribution, insufficient convergence ability, and lack of a physical constraint active repair mechanism. This significantly improves the global exploration capability, convergence accuracy, and feasible solution acquisition rate in the high-dimensional strongly constrained solution space. Ultimately, it can facilitate the time-series migration of computing tasks, the cross-period energy transfer of pumped storage, and the effective synergy between thermal power and new energy output. This is conducive to ensuring that the main grid voltage remains stable within a safe range while significantly reducing the overall system operating cost and carbon emissions, and has good engineering applicability. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a computing, power, and carbon collaborative optimization scheduling method that takes into account the timing reconstruction of computing tasks, according to an embodiment of this application.

[0027] Figure 2This is a flowchart illustrating the solution process of an improved gray wolf optimization algorithm provided in an embodiment of this application.

[0028] Figure 3 This is a schematic diagram of the improved IEEE 118-node system provided in the embodiments of this application.

[0029] Figure 4 This is a schematic diagram of the workload of various computing centers provided in the embodiments of this application.

[0030] Figure 5 This is a schematic diagram of the timing reconstruction of computing power tasks in the computing center provided in the embodiments of this application.

[0031] Figure 6 This is a schematic diagram of the coordinated scheduling of pumped storage and computing center provided in the embodiments of this application.

[0032] Figure 7 This is a schematic diagram of the overall source-load-storage scheduling provided in the embodiments of this application.

[0033] Figure 8 This is a structural block diagram of a computing power-carbon collaborative optimization scheduling device that takes into account the timing reconstruction of computing tasks, as provided in an embodiment of this application.

[0034] Figure 9 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0035] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The terms “connected” or “linked” and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0037] Figure 1 This is a flowchart of a computing-power-carbon collaborative optimization scheduling method that considers computing power task timing reconstruction, as provided in an embodiment of this application. Figure 1As shown, the method includes: Step S1: Obtain multi-source system operation data.

[0038] Step S2: Construct a multi-source system model based on the multi-source system operation data, and determine the constraints of the multi-source system model.

[0039] The multi-source system model is used to characterize the time-series reconstruction boundary of computing tasks based on latency characteristics of computing centers, the operation characteristics of pumped storage power stations, and the output characteristics of thermal power and new energy.

[0040] Step S3: Construct network security constraints based on the multi-source system model and dynamic reactive power support, and determine the network security breach penalty function.

[0041] Step S4: Based on multi-source system operation data, multi-source system model and network security breach penalty function, construct a multi-objective optimization function that coordinates economic efficiency, low carbon emissions and new energy absorption capacity.

[0042] Step S5: Solve using the improved Grey Wolf optimization algorithm to obtain the optimal scheduling scheme.

[0043] The improved gray wolf optimization algorithm performs physical constraint repair on the position vectors of individual gray wolves based on constraint conditions, and performs comprehensive fitness evaluation on the repaired position vectors of individual gray wolves based on network security constraints, network security breach penalty functions, and multi-objective optimization functions.

[0044] In this embodiment, a multi-source system model is constructed to accurately define the feasible domain of computing power task time-series reconstruction, providing a decision boundary basis for the effective release of flexible resources in the computing power center. This model can precisely characterize the cross-time period energy transfer operation characteristics of pumped storage power stations and the computing power task time-series reconstruction boundary, achieving deep spatiotemporal synergy between high-energy-consuming computing loads and energy storage resources. This is beneficial for improving the absorption rate of new energy and smoothing system net load fluctuations. By constructing network security constraints that consider dynamic reactive power support and determining corresponding network security violation penalty functions, a multi-objective optimization function is constructed that coordinates economic efficiency, low carbon emissions, and new energy absorption capacity. An improved gray wolf optimization algorithm is used for solving this function. Based on the constraints, the algorithm actively repairs the physical constraints of individual gray wolves and performs a comprehensive fitness evaluation. This overcomes the shortcomings of the standard algorithm, such as uneven initial distribution, insufficient convergence ability, and lack of a physical constraint active repair mechanism. This significantly improves the global exploration capability, convergence accuracy, and feasible solution acquisition rate in the high-dimensional strongly constrained solution space. Ultimately, it can facilitate the time-series migration of computing tasks, the cross-period energy transfer of pumped storage, and the effective synergy between thermal power and new energy output. This is conducive to ensuring that the main grid voltage remains stable within a safe range while significantly reducing the overall system operating cost and carbon emissions, and has good engineering applicability.

[0045] Optionally, in step S1, the multi-source system operation data includes the network topology of the multi-source system, node load parameters, dynamic reactive power compensation device parameters, computing power task load and facility parameters of the computing center, initial state and physical parameters of the pumped storage power station, and operating parameters of thermal power and new energy, etc.

[0046] For example, in step S1, the multi-source system can also be initialized and the scheduling period can be set. Time step In this embodiment, the scheduling cycle is 24 hours, and the duration of a single scheduling period is 1 hour.

[0047] Optionally, in step S2, the multi-source system model includes a computing center task time-series reconstruction model, a pumped storage power station reservoir energy evolution model, and a thermal power and new energy output model. Constraints include operating rules for the computing center, operating rules for the pumped storage power station, and operating rules for thermal power and new energy.

[0048] For example, the computing center task time-series reconfiguration model is used to rigorously characterize the boundaries of computing task time-series reconfiguration based on latency characteristics in the computing center in both physical and temporal dimensions. The computing center task time-series reconfiguration model is described below.

[0049] Computing center during time period The total computing power task arrival volume consists of three parts: rigid tasks, elastic tasks, and training tasks, expressed as: (1) in, This represents the total number of computing power tasks reached. For rigid task arrival quantity, For the arrival quantity of elastic tasks, This represents the number of training tasks achieved.

[0050] Time period The actual task execution volume consists of the immediate execution volume of rigid tasks, the volume of elastic tasks that are migrated to the execution period within the allowed delay window, and the training task volume, expressed as: (2) in, For server clusters in time periods The actual equivalent calculation workload undertaken For flexible tasks arriving at the time period And after the computing power task time sequence reconstruction, it is arranged in the time period The amount of tasks performed For the training task, the time period reached And after the computing power task time sequence reconstruction, it is arranged in the time period The amount of tasks performed.

[0051] The set of allowed execution periods for transferable tasks is represented as follows: (3) in, To arrive at the time period The The set of allowed execution periods for class-transferable tasks. For the first The maximum allowable delay window for transferable tasks. During subsequent scheduling optimization, only task arrangements falling within this set of allowable execution periods (legal time windows) are considered feasible solutions.

[0052] Migrative tasks cannot be executed ahead of schedule or exceed their service period, and the total number of tasks is conserved, represented as: (4) in, For the first In a class of transferable tasks, arrival at a time period And after the computing power task time sequence reconstruction, it is arranged in the time period The amount of tasks performed For time period The Number of transferable tasks reached by class For rigid tasks in time periods The actual execution volume. The first two constraints in equation (4) ensure that the transferable task is neither lost nor goes out of bounds, while the third constraint ensures that the rigid task runs according to the rule of execution upon arrival.

[0053] The server capacity constraint of the computing center can be expressed as: (5) in, For server cluster size, This represents the maximum equivalent computational workload that a single server can undertake under given operational constraints. This constraint (5) can prevent excessive tasks from being backlogged and scheduled to certain low-price periods during subsequent optimization and scheduling processes.

[0054] Given the actual execution volume, the time period The equivalent number of servers in operation is represented as follows: (6) in, For time period The number of servers that are running reflects the direct correlation between the actual load and the server deployment.

[0055] The power of information equipment consists of the server's maintenance power and the incremental power caused by the load, expressed as: (7) in, For information equipment power, and These represent the idle maintenance power and the peak power under full load, respectively. Equation (7) shows that the power of the computing center is not simply linearly amplified by the amount of tasks, but is affected by both the number of online servers and the load intensity.

[0056] After considering the energy consumption of cooling, power supply and distribution, and auxiliary facilities, the total power consumption of the computing center can be expressed as: (8) in, For time period The total power consumption of the computing center For time period The power utilization efficiency coefficient. Equation (8) can further map the power of information equipment (IT equipment) to the total power of the facility, which is closer to the actual energy consumption structure of the computing center.

[0057] Task latency can be calculated by multiplying the number of migrated tasks by the migration duration, and expressed as: (9) in, For the first Total task latency for class-transferable tasks.

[0058] When task migration is not allowed, elastic tasks and training tasks can be executed upon arrival, as follows: (10) in, This is an indicator function. Equation (10) can provide a unified benchmark for scenarios where task migration is not enabled.

[0059] For example, a pumped storage power station reservoir energy evolution model is used to characterize the operating characteristics of a pumped storage power station. The following is an explanation of the pumped storage power station reservoir energy evolution model.

[0060] Pumped storage power stations can be in generating, pumping, or shut-down states during their operating cycle, and their power boundary can be expressed as: (11) in, Pumped storage power station During the period Power generation capacity, and These are the lower and upper limits of power generation, respectively. Pumped storage power station During the period Pumping power, and These are the lower and upper limits of pumping power, respectively. and These represent the power generation and pumping states of a pumped storage power station, respectively.

[0061] Pumping and power generation cannot occur simultaneously within the same time period. This constraint can be expressed as: (12) Equation (12) can ensure that a pumped storage power station will not charge or release energy to the grid at the same time.

[0062] The reservoir capacity of a pumped-storage power station is recursively calculated as pumping increases and power generation decreases, expressed as follows: (13) in, Pumped storage power station During the period Storage capacity energy, This represents the reservoir capacity from the previous period. For pumping efficiency, For power generation efficiency.

[0063] The reservoir capacity and carryover capacity of pumped storage power stations must both meet the allowable range, expressed as follows: (14) in, Pumped storage power station At the end of the scheduling cycle Storage capacity energy, and These are the lower and upper limits of the reservoir's energy capacity during operation. and These are the lower and upper limits of the allowed reservoir capacity energy carryover at the end of the scheduling cycle, respectively. Equation (14) can preserve intraday flexibility without disrupting the continuous operation conditions for the next cycle.

[0064] The net power injection into the power grid from pumped storage is expressed as: (15) in, Pumped storage power station During the period The net power injected into the power grid, when positive, indicates that electricity is being supplied to the grid, and when negative, it indicates that electrical energy is being absorbed from the grid for pumping water.

[0065] To conform to the typical operating mode of fixed-speed pumped storage, the number of start-ups and shutdowns of pumping and power generation can be constrained within the scheduling cycle, as follows: (16) in, and These represent the maximum number of start-stop cycles allowed in pumping and power generation modes, respectively. This constraint (16) can make the energy storage scheduling of pumped storage more in line with engineering operation habits and avoid frequent switching of operating conditions.

[0066] For example, the output models for thermal power and new energy sources are used to characterize the output characteristics of thermal power and new energy sources. The output models for thermal power and new energy sources are explained below.

[0067] The power output boundaries of wind and solar power in new energy sources can be expressed as: (17) in, For wind farm collection, A collection of photovoltaic power plants, and Wind farm With photovoltaic power station During the period Actual grid-connected power, and These represent the predicted available power for wind farms and photovoltaic power plants, respectively.

[0068] To quantify the extent of wind and solar curtailment, the amount of renewable energy not absorbed during the dispatch cycle is represented as follows: (18) in, This refers to unused renewable energy power.

[0069] The active power output of thermal power units must always be within their technically permissible range, expressed as: (19) in, For the collection of thermal power units, For thermal power units During the period Those who have made contributions and These are the minimum stable output and maximum technical output of the thermal power unit, respectively. The constraint of this formula (19) can ensure that the thermal power unit will not deviate from its safe and economical operating range when undertaking peak shaving and supply guarantee tasks.

[0070] The planned total output of conventional power sources such as thermal power units in each time period is determined by the system base load, the power load of the computing center, the net injected power of pumped storage, and the actual grid-connected power of new energy sources, as expressed as: (20) in, For time period The required total planned output of thermal power units, For the system's basic load, This is a collection of pumped-storage power stations. When pumped-storage power is generated, The value is positive, thus reducing the demand for conventional power sources such as thermal power units; during pumping, A negative value is equivalent to increasing the amount of power supply that conventional power sources such as thermal power units need to provide. Equation (20) connects the priority consumption of new energy sources, the cross-time regulation of pumped storage, and the backup power supply of conventional power sources such as thermal power units into a unified power distribution relationship.

[0071] For example, in step S3, the expression for the network security constraint is: (twenty one) (twenty two) (twenty three) (twenty four) in, and They are nodes During the period The net active power injection and net reactive power injection, For the set of power grid nodes, and They are nodes and nodes During the period The node voltage amplitude, and They are nodes ,node The conductivity and susceptance between them For nodes With nodes During the period The phase angle difference, and They represent the lines respectively. In the nodes To node , by node To node Apparent power in both directions, For the line Maximum permissible apparent power For the collection of transmission lines, For the scheduling period, and They are nodes The lower and upper limits of the voltage. For thermal power units During the period Unproductive efforts and thermal power units The lower and upper limits of reactive power output, For the collection of thermal power units, For nodes Dynamic reactive power compensation device during time period The actual reactive power output, and These represent the lower and upper limits of reactive power output allowed by the dynamic reactive power compensation device, respectively. For nodes Dynamic reactive power compensation device during time period The target reactive power output is obtained based on the current voltage deviation. For the limited projection operator, This is the voltage deviation proportional control coefficient. The nodes obtained for AC power flow calculation During the period The node voltage amplitude, This is the reference value for node voltage. For nodes Dynamic reactive power compensation device during time period Reactive power output after smoothing To smoothly update the coefficients, .

[0072] In network security constraints that take into account dynamic reactive power support, each node in the time period The net active and reactive power injections satisfy the AC power flow balance relationship and satisfy equation (21). Equation (21) is used to convert the power injection of the source, load and storage into the network power flow state, so as to determine whether the candidate solution will cause voltage abnormality or power flow imbalance. Equation (22) adopts a bidirectional constraint method, which can simultaneously cover the forward and reverse power flow of the line, so as to avoid the situation that one direction seems feasible but the other direction has exceeded the limit. The constraint of equation (23) can facilitate the inclusion of voltage safety and generator excitation capability into the evaluation, so that the candidate solution is not only economical, but also feasible.

[0073] To enhance voltage support under conditions of high proportion of new energy and large-scale computing load, dynamic reactive power compensation devices are configured at key nodes. In this embodiment, the dynamic reactive power compensation device can be a static var generator (SVG). Equation (24) above represents the reactive power output of the dynamic reactive power compensation device adjusted by voltage deviation closed loop. The limiting projection operator is used to limit the target reactive power output calculated according to the voltage deviation within the physical adjustment range of the dynamic reactive power compensation device, so as to avoid the compensation command exceeding the equipment capacity. When the node voltage is lower than the reference value, the device increases reactive power injection; when the node voltage is higher than the reference value, the device decreases reactive power injection or absorbs reactive power.

[0074] For candidate schemes that exceed limits or are not feasible for power flow, the penalties for exceeding node voltage limits, line capacity limits, and reactive power output limits of conventional units can be calculated separately.

[0075] For example, the expression for the network security breach penalty function is: (25) (26) in, To indicate the line During the period Apparent power Indicates the degree to which the node voltage exceeds the limit. Indicates the degree to which the line capacity has exceeded its limit. This indicates the degree to which the reactive power boundary of a thermal power unit is exceeded. For cybersecurity breach of contract penalties, , , and These are the penalty weights for exceeding node voltage limits, exceeding line capacity limits, exceeding reactive power output limits, and being infeasible for power flow. This represents the number of time periods during which a feasible solution for the current trend was not obtained.

[0076] For example, in step S4, the multi-objective optimization function is expressed as: (27) in, Indicates the overall optimization objective. For the comprehensive operating costs of multi-source systems, For carbon dioxide emissions from multi-source systems, This refers to unconsumed renewable energy. The three objectives correspond to economic efficiency, low carbon emissions, and renewable energy utilization levels, respectively. There are inherent conflicts among these three objectives, therefore, multi-objective optimization is required rather than single-objective minimization.

[0077] Specifically, the comprehensive operating cost consists of system-side costs, computing center costs, deductions from pumped storage revenue, and cybersecurity penalties, expressed as follows: (28) in, For the operating costs on the power system side, Costs for electricity purchases for the computing center and delays for portable tasks. For the benefits generated from pumped storage participating in dispatch, penalties for cybersecurity breaches. Used to exclude infeasible network operation solutions during the optimization search process, and not as actual operating expenses in the economic benefit analysis.

[0078] System-side costs can be further divided into thermal power unit fuel costs, pumped storage operation and maintenance costs, renewable energy curtailment costs, off-load costs, and carbon trading costs, expressed as follows: (29) in, For the operating costs of thermal power units, For the operation and maintenance costs of pumped storage equipment, Losses due to wind and solar power curtailment As punishment for insufficient power supply, Transaction costs resulting from excess carbon emissions.

[0079] The operating cost expression for a thermal power unit is: (30) in, , and These are the fuel cost coefficients. Equation (30) uses a quadratic function, which can reflect the characteristics of the rising marginal cost of thermal power units and is also compatible with the scheduling optimization process.

[0080] The expressions for electricity purchase and migration-enabled task delay costs for computing centers are as follows: (31) in, For time period Electricity price, and These are the cumulative latency for the elastic task and the training task, respectively. and These are the unit delay cost coefficients for elastic tasks and training tasks, respectively. Equation (31) incorporates both electricity consumption and delay costs into the cost, avoiding the indefinite postponement of tasks simply by shifting peak loads to valleys.

[0081] The expressions for the operation and maintenance costs and operating revenue of pumped storage equipment are as follows: (32) in, and These represent the unit operation and maintenance costs of pumped storage hydroelectric power generation and pumping operation, respectively. As a backup compensation price, This refers to the available backup capacity.

[0082] The expressions for renewable energy curtailment loss and load shedding loss are: (33) in, The unit of power curtailment penalty coefficient, For time period The power of the load shedding The unit loss of load penalty coefficient.

[0083] The expressions for carbon emissions from thermal power units, system carbon quotas, and carbon trading fees are as follows: (34) in, For thermal power units The carbon emission coefficient per unit of electricity generated, For the system's carbon quota, Carbon quota coefficient per unit of electricity consumption. The price is the unit price for carbon emissions trading.

[0084] Figure 2 This is a flowchart illustrating the solution process of an improved gray wolf optimization algorithm provided in an embodiment of this application. Figure 2 As shown, exemplarily, in step S5, an improved gray wolf optimization algorithm is used to solve the problem, including the following steps: Step S51: Set the initial parameters for the improved Grey Wolf optimization algorithm.

[0085] For example, the multi-source system operation data obtained in step S1 can be loaded, and the initial parameters of the improved gray wolf optimization algorithm can be set, including the gray wolf population size and the maximum number of iterations. And the penalty coefficient.

[0086] Step S52: Encode decision variables based on multi-source system operation data and multi-source system model, and introduce Sobol low-difference sequences to generate the initial gray wolf population location vector.

[0087] For example, the aforementioned steps S1 and S2 can encode the decision variables as position vectors of individual gray wolves. The position vector of a single gray wolf can encode the output preference of a thermal power unit, the power generation and pumping power of a pumped storage power station, the migration ratio between the flexible task and the training task, and the execution preference between the flexible task and the training task. Specifically, its encoding structure can be represented as follows: (35) in, This represents the position vector of an individual gray wolf. Based on the power output preference of thermal power units, and These refer to the power generation and pumping capacity of a pumped storage power station, respectively. and These represent the transfer rates for the flexible task and the training task, respectively. and These represent the execution preferences for the elastic task and the training task, respectively. Generating the initial gray wolf population position vector using Sobol low-discrepancy sequences can make the initial solution more evenly distributed within the feasible region.

[0088] Step S53: Perform physical constraint repair on the position vector of the individual gray wolf based on the constraint conditions.

[0089] For example, step S53 is used to repair the position vectors of the individual gray wolves generated in step S52 into candidate scheduling schemes that satisfy the various constraints in step S2. In step S53, the candidate solutions can be sequentially repaired, state recursively calculated, and derived quantities recalculated according to the operating rules constraints of the computing center, the operating rules constraints of the pumped storage power station, and the operating rules constraints of thermal power and new energy.

[0090] Specifically, for the computing power task variable, the execution time of the elastic task and training task generated in step S52 is first restricted based on the set of allowed execution time periods for the transferable tasks in step S2. For example, the set of allowed execution time periods for various types of transferable tasks can be determined according to equation (3), prohibiting tasks from being executed in advance or scheduled outside the service period; according to equation (4), the total execution volume of various types of transferable tasks is kept consistent with the number of tasks arriving, and rigid tasks are processed according to the rule of execution upon arrival; according to equation (5), the maximum carrying capacity of the server cluster in each time period is checked.

[0091] When the number of tasks to be scheduled in a certain period exceeds the server's capacity limit, the excess portion is allocated sequentially to time periods within the allowed execution window that still have remaining capacity, according to task execution preferences. The number of servers to be started in equation (6) and the power of information equipment and the total power consumption of the computing center in equations (7) and (8) are recalculated. If a task cannot be allocated within all legal time periods, the candidate solution is marked as infeasible and an infeasibility penalty is imposed in the subsequent fitness evaluation. The constraint violation cannot be concealed by executing ahead of time or exceeding the server capacity.

[0092] For the completed repair task sequence, the cumulative delay of the elastic task and the training task is recalculated according to Equation (9); when the task migration function is turned off, the elastic task and the training task are restored to the state of execution upon arrival according to Equation (10).

[0093] Specifically, for pumped storage variables, the operating conditions of each time period are first determined based on the pumped storage operating state variables in step S2. According to equation (11), the power generation and pumping power are projected to their respective allowable upper and lower limits; when pumping and power generation occur simultaneously in the same time period, one of the operating conditions is retained according to the current operating state, and the other power is set to zero, thereby satisfying the mutual exclusion constraint of pumping and generation in equation (12).

[0094] For pumping or power generation states below the equipment's minimum operating power, the corresponding operating state is set to shutdown, and the power for that period is set to zero. Subsequently, the pumped storage reservoir energy state is recursively calculated for each period according to equation (13): the reservoir energy is increased during pumping, and the reservoir energy is reduced according to the power generation efficiency during power generation. If the reservoir energy exceeds the upper or lower limit specified in equation (14) during the recursion process, the pumping power or power generation power for the current period is adjusted in the opposite direction, and subsequent periods are recursively calculated until the reservoir energy for the entire cycle meets the safety capacity constraint.

[0095] After completing the full cycle recursion, further check whether the reservoir energy at the end of the scheduling cycle meets the carry-over interval requirements; if not, prioritize adjusting the pumping or power generation periods with adjustment margins at the end of the period. After the repair is completed, recalculate the net power injection of pumped storage to the power grid according to formula (15), and recount the number of start-ups and shutdowns of pumping and power generation conditions according to formula (16) to ensure that the number of start-ups and shutdowns does not exceed the allowable upper limit.

[0096] Specifically, for wind power and photovoltaic variables, based on the new energy available power prediction results in step S2, the actual grid-connected power for each time period is projected into the range from zero to the predicted available power specified in equation (17); the portion exceeding the available power is considered invalid output and is cut off. After the correction, the amount of wind and solar power curtailment is recalculated according to equation (18).

[0097] For the active power output of thermal power units, the active power output of each unit at each time period is projected to the range between the minimum stable output and the maximum technical output according to equation (19). The total planned output of thermal power units required by the system is calculated according to equation (20), and the total output is allocated according to the remaining adjustable capacity of each thermal power unit to ensure that each unit meets the output boundary of equation (19).

[0098] When the planned total output of thermal power units required by the system exceeds the maximum output that all thermal power units can provide, or is lower than the sum of the minimum stable output of all units, it indicates that there is no feasible conventional unit allocation scheme that satisfies power balance in the current task reconfiguration, pumped storage operation and new energy output combination, and the candidate scheme can be marked as active power balance infeasible.

[0099] After completing the above three types of physical constraint repairs, the position vector of the individual gray wolf is obtained after satisfying the constraints of task timing, pumped storage operation status, reservoir energy evolution, thermal power unit output, available power of new energy sources, and system active power balance.

[0100] Step S54: Calculate the comprehensive fitness of the position vector of the repaired gray wolf individual based on network security constraints, network security breach penalty function and multi-objective optimization function.

[0101] For example, the candidate scheduling scheme repaired in step S53 can be substituted into the AC power flow calculation, and the net active and reactive power injection and node voltage status of each node in each time period can be calculated according to equation (21). For nodes configured with SVG, the actual reactive power output of SVG in the previous time period is used as the initial value, and the reactive power output of SVG is updated according to the node voltage deviation according to equation (24). After each update of the reactive power output of SVG, the AC power flow calculation is re-executed until the change in reactive power output of SVG is less than the preset threshold or the maximum number of closed-loop iterations is reached.

[0102] Based on the converged power flow results, the bidirectional apparent power capacity of the line is checked according to Equation (22), the node voltage and the reactive power output boundary of the thermal power unit are checked according to Equation (23), and the node voltage exceeding the limit, the line capacity exceeding the limit, and the reactive power output exceeding the limit of the thermal power unit are calculated according to Equation (25); the amount of non-feasible power flow is calculated according to Equation (26) and the network security violation penalty is determined.

[0103] Furthermore, the comprehensive operating cost, carbon dioxide emissions, and unconsumed renewable energy power of the candidate scheduling scheme can be calculated based on equations (27) to (34). The above three normalized objectives are combined with network security breach penalties to form the comprehensive fitness of the position vector of the individual gray wolf.

[0104] In this embodiment, the formula for calculating the overall fitness is: (36) in, The position vector of the individual gray wolf Overall adaptability , , The weights for economic efficiency, low carbon emissions, and renewable energy consumption targets are respectively. ,and , , All are non-negative numbers. , , These are the position vectors of individual gray wolves. The corresponding comprehensive operating costs of multi-source systems, carbon dioxide emissions, and unused renewable energy power. The position vector of the individual gray wolf Corresponding cybersecurity breach penalties , , These are the comprehensive operating costs, carbon dioxide emissions, and unconsumed renewable energy power of the preset reference scheme, respectively. The lower the comprehensive fitness value, the better the candidate scheme's economy, low carbon emissions, renewable energy consumption capacity, and network security.

[0105] Step S55: Select the three gray wolf individuals with the lowest overall fitness and mark them as follows: Wolf, wolves and Wolves, and mark the remaining gray wolf individuals as Wolf.

[0106] For example, all gray wolf individuals can be sorted in ascending order based on the overall fitness value calculated in step S54, and the three gray wolf individuals with the lowest overall fitness values ​​can be selected and marked as follows: Wolf, wolves and The three alpha wolves will guide the pack's subsequent hunting and searching, and mark the remaining gray wolves as... Wolf.

[0107] Step S56: Based on Wolf, wolves and The wolf's position vector is updated using a nonlinear cosine adaptive convergence factor. The wolf's position vector.

[0108] For example, the expression for the nonlinear cosine adaptive convergence factor is: (37) in, Indicates the current iteration number. The maximum number of iterations, For the first The nonlinear cosine adaptive convergence factor is used in the next iteration. This factor balances the algorithm's global exploration capability in the early stages with its local optimization accuracy in the later stages.

[0109] For example, the th... can be calculated based on the nonlinear cosine adaptive convergence factor. The first gray wolf individual in the... In the next iteration, the distance vector between the three wolves is updated, and the distance vector between the three wolves is updated. The first gray wolf individual in the... The final position vector in the next iteration is represented as: (38) (39) in, , , The first Individual gray wolves and Wolf, wolves and The distance vector between wolves , , They are respectively Wolf, wolves and The wolf's current position vector This indicates element-wise multiplication. This indicates taking the absolute value of each element. , , , , , They are respectively A dynamically changing vector of synergy coefficients. For the first indivual wolves in The position vector in the next iteration. For the first indivual wolves in The position vector in the next iteration.

[0110] Step S57: Determine if the maximum number of iterations has been reached. If not, return to step S53 for a new round of physical constraint repair, comprehensive fitness evaluation, and nonlinear adaptive position update; otherwise, output the current... The wolf's position vector is used as the optimal scheduling scheme.

[0111] For example, the iteration count can be updated to And determine whether the maximum number of iterations has been reached. If the condition is not met, return to step S53 for a new round of physical constraint repair, comprehensive fitness evaluation, and nonlinear adaptive position update; if the maximum number of iterations is reached (the termination condition is met), output the current iteration. The position vector corresponding to the wolf is used as the optimal scheduling scheme. Finally, a rigorous network security check can be performed on this representative solution, and the corresponding wind-solar energy absorption curve, pumped storage operation strategy, computing power reconstruction sequence, and comprehensive target cost can be output.

[0112] The improved Grey Wolf optimization algorithm, implemented through steps S51 to S57, constructs a complete closed-loop solution system encompassing multidimensional variable mapping, space exploration, and rigorous boundary verification. This improved algorithm eliminates invalid solutions that violate pumped storage operation conditions and computational task timing patterns by pre-emptively fixing physical constraints, significantly compressing the invalid search space and reducing computational overhead. Furthermore, it combines multi-objective linear weighting with a high-value dynamic penalty function, and through the synergistic design of a Sobol low-discrepancy sequence and a nonlinear cosine adaptive convergence factor, guides the Grey Wolf population to converge rapidly towards the feasible region, effectively overcoming the hysteresis phenomenon in complex non-convex optimization. The final optimal representative solution not only achieves global optimization in the economic and low-carbon dimensions of load transfer, energy storage synergy, and new energy consumption, but also ensures the absolute physical feasibility of the solution in a real power distribution network environment, providing highly robust underlying algorithmic support for the deep integration and scheduling of computing centers and new power systems.

[0113] The effects of this application will be further illustrated below through computer simulation.

[0114] Figure 3 This is a schematic diagram of the improved IEEE 118-node system provided in an embodiment of this application. Figure 3 As shown, an improved IEEE 118-node mainnet computational example can be constructed, with a scheduling period of 24 hours and a time step of 1 hour. The system is configured with wind power, photovoltaic, thermal power units, pumped storage power stations, computing centers, and dynamic reactive power compensation devices. The dynamic reactive power compensation devices, SVG equivalent dynamic reactive power compensation, are deployed on Bus118, Bus59, Bus83, Bus65, Bus93, Bus101, Bus95, Bus21, and Bus38. The aforementioned computational-power-carbon co-optimization scheduling method, which considers the time-series reconstruction of computing tasks, is used for optimization scheduling. In the computational example scenario, pumped storage is allowed to participate in cross-time period regulation, and computing tasks are allowed to be reconstructed within their time constraints. To examine the applicability of different computing business structures, three types of computing centers are set up: enterprise business type, AI training type, and university research type.

[0115] Figure 4 This is a schematic diagram of the workload of various computing centers provided in the embodiments of this application. Figure 5 This is a schematic diagram of the timing reconstruction of computing power tasks in a computing center provided in an embodiment of this application. See also... Figure 4 and Figure 5 Enterprise-oriented computing centers have a high proportion of rigid tasks, and the adjustment range of the execution time of transferable tasks after scheduling is relatively limited. AI training-oriented computing centers have a high proportion of training tasks, and training tasks have greater room for reconstruction within the scheduling cycle. University research-oriented computing centers have a high proportion of both flexible tasks and training tasks, and their task execution curves show more obvious time-adjustable characteristics. Therefore, this application can form differentiated task execution plans based on the rigidity ratio, delay window, and delay cost of different business tasks, rather than equating the computing load to a single fixed load.

[0116] Figure 6 This is a schematic diagram of the coordinated scheduling of pumped storage and computing center provided in the embodiments of this application. Figure 6 The evolution of pumped storage power and reservoir capacity is illustrated. For example... Figure 6 As shown, in the three types of computing center scenarios, pumped storage enters the pumping state during low-load periods and transitions to the power generation state in subsequent periods. The reservoir energy curve is consistent with the changes in pumping and generating power, indicating that the proposed reservoir energy evolution, pumping-generating mutual exclusion, and end-of-period reservoir energy constraints can be effectively satisfied. Pumped storage and computing task reconfiguration jointly participate in the system source-load balance, providing cross-period adjustment capabilities for different types of computing loads.

[0117] Figure 7 This is a schematic diagram of the overall source-load-storage scheduling provided in the embodiments of this application. Figure 7 The overall dispatch results for thermal power, wind power, photovoltaic power, pumped storage, and load side are shown. For example... Figure 7 As shown, with the increase in daytime photovoltaic output, the output of thermal power decreases accordingly; pumped storage achieves energy transfer across time periods through switching between pumping and power generation states. The daily electricity consumption of computing centers in the three scenarios are 4955.4 MWh, 5378.9 MWh, and 5803.9 MWh, respectively. The corresponding computing task time-series reconstruction scale and carbon emission levels differ, indicating that the method provided in this application can adapt to computing center access scenarios with different task compositions and energy efficiency characteristics.

[0118] The following are device embodiments of this application. For details not described in detail in the device embodiments, please refer to the above method embodiments.

[0119] Figure 8 This is a structural block diagram of a computing power-carbon collaborative optimization scheduling device that considers computing power task timing reconstruction, as provided in an embodiment of this application. Figure 8As shown, the computing power-carbon collaborative optimization scheduling device 100, which takes into account the timing reconstruction of computing power tasks, includes an acquisition module 101, a first construction module 102, a second construction module 103, a third construction module 104, and a solution module 105.

[0120] The acquisition module 101 is used to acquire multi-source system operation data.

[0121] The first construction module 102 is used to construct a multi-source system model based on the multi-source system operation data and determine the constraints of the multi-source system model. The multi-source system model is used to characterize the time sequence reconstruction boundary of the computing power center based on the latency characteristics, the operation characteristics of the pumped storage power station, and the output characteristics of thermal power and new energy.

[0122] The second construction module 103 is used to construct network security constraints based on a multi-source system model and dynamic reactive power support, and to determine the network security breach penalty function.

[0123] The third construction module 104 constructs a multi-objective optimization function that coordinates economic efficiency, low carbon emissions, and new energy consumption capacity based on multi-source system operation data, multi-source system models, and network security breach penalty functions.

[0124] The solution module 105 is used to solve the optimal scheduling scheme using the improved gray wolf optimization algorithm. The improved gray wolf optimization algorithm performs physical constraint repair on the position vector of gray wolf individuals based on constraint conditions, and performs comprehensive fitness evaluation on the position vector of the repaired gray wolf individuals based on network security constraints, network security breach penalty function and multi-objective optimization function.

[0125] It should be noted that the above embodiments of the computing power-carbon collaborative optimization scheduling device considering computing power task timing reconstruction are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the computing power-carbon collaborative optimization scheduling device considering computing power task timing reconstruction and the computing power-carbon collaborative optimization scheduling method embodiments considering computing power task timing reconstruction belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0126] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in each embodiment of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0127] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a terminal device (which may be a personal computer, mobile phone, or communication device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] Figure 9 This is a schematic diagram of the structure of the computer device provided in an embodiment of this application. Figure 9 As shown, the computer device 200 includes a processor 201 and a memory 202.

[0129] Processor 201 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 201 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 201 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 201 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, processor 201 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0130] The memory 202 may include one or more computer-readable storage media, which may be non-transitory. The memory 202 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 202 is used to store at least one instruction, which is executed by the processor 201 to implement the computational-electrical-carbon co-optimization scheduling method considering computational power task timing reconstruction provided in the embodiments of this application.

[0131] Those skilled in the art will understand that Figure 9 The structure shown does not constitute a limitation on the computer device 200, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0132] This application also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of a computer device, enables the computer device to execute the computing power-electricity-carbon collaborative optimization scheduling method considering computing power task timing reconstruction provided in this application.

[0133] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the computational, power, and carbon collaborative optimization scheduling method considering computational power task timing reconstruction provided in this application.

[0134] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A computational-power-carbon collaborative optimization scheduling method considering computational power task time-series reconstruction, characterized in that, include: Acquire multi-source system operation data; A multi-source system model is constructed based on the multi-source system operation data, and the constraints of the multi-source system model are determined. The multi-source system model is used to characterize the time sequence reconstruction boundary of computing power tasks based on latency characteristics of the computing power center, the operation characteristics of pumped storage power stations, and the output characteristics of thermal power and new energy. Based on the multi-source system model, network security constraints considering dynamic reactive power support are constructed, and a network security breach penalty function is determined. Based on the multi-source system operation data, the multi-source system model, and the network security breach penalty function, a multi-objective optimization function that coordinates economic efficiency, low carbon emissions, and new energy consumption capacity is constructed. An improved gray wolf optimization algorithm is used to solve the problem and obtain the optimal scheduling scheme. The improved gray wolf optimization algorithm performs physical constraint repair on the position vector of gray wolf individuals based on the constraints, and performs comprehensive fitness evaluation on the repaired position vector of gray wolf individuals based on the network security constraints, the network security violation penalty function and the multi-objective optimization function.

2. The computing power-carbon collaborative optimization scheduling method considering computing power task time sequence reconstruction according to claim 1, characterized in that, The multi-source system model includes a computing center task time-series reconstruction model, a pumped storage power station reservoir energy evolution model, and a thermal power and new energy output model. The constraints include the operating rules of the computing center, the operating rules of the pumped storage power station, and the operating rules of thermal power and new energy.

3. The computational-power-carbon collaborative optimization scheduling method considering computational power task time sequence reconstruction according to claim 1, characterized in that, The expression for the network security constraint is: in, and They are nodes During the period The net active power injection and net reactive power injection, For the set of power grid nodes, and They are nodes and nodes During the period The node voltage amplitude, and They are nodes ,node The conductivity and susceptance between them For nodes With nodes During the period The phase angle difference, and They represent the lines respectively. In the nodes To node , by node To node Apparent power in both directions, For the line Maximum permissible apparent power For the collection of transmission lines, For the scheduling period, and They are nodes The lower and upper limits of the voltage. For thermal power units During the period Unproductive efforts and thermal power units The lower and upper limits of reactive power output, For the collection of thermal power units, For nodes Dynamic reactive power compensation device during time period The actual reactive power output, and These represent the lower and upper limits of reactive power output allowed by the dynamic reactive power compensation device, respectively. For nodes Dynamic reactive power compensation device during time period The target reactive power output is obtained based on the current voltage deviation. For the limited projection operator, This is the voltage deviation proportional control coefficient. The nodes obtained for AC power flow calculation During the period The node voltage amplitude, This is the reference value for node voltage. For nodes Dynamic reactive power compensation device during time period Reactive power output after smoothing To smoothly update the coefficients; The expression for the network security breach penalty function is: in, To indicate the line During the period Apparent power Indicates the degree to which the node voltage exceeds the limit. Indicates the degree to which the line capacity has exceeded its limit. This indicates the degree to which the reactive power boundary of a thermal power unit is exceeded. For cybersecurity breach of contract penalties, , , and These are the penalty weights for exceeding node voltage limits, exceeding line capacity limits, exceeding reactive power output limits, and being infeasible for power flow. This represents the number of time periods during which a feasible solution for the current trend was not obtained.

4. The computational power-carbon collaborative optimization scheduling method considering computational power task time sequence reconstruction according to claim 3, characterized in that, The multi-objective optimization function is expressed as: in, Indicates the overall optimization objective. For the comprehensive operating costs of multi-source systems, For carbon dioxide emissions from multi-source systems, This refers to unused renewable energy power.

5. The computational power-carbon collaborative optimization scheduling method considering computational power task time sequence reconstruction according to claim 4, characterized in that, The method of solving the problem using the improved gray wolf optimization algorithm includes: Set the initial parameters for the improved gray wolf optimization algorithm; Decision variables are encoded based on the multi-source system operation data and the multi-source system model, and Sobol low-difference sequences are introduced to generate the initial gray wolf population location vector. Physical constraint repair is performed on the position vector of the individual gray wolf based on the aforementioned constraints; The comprehensive fitness of the position vector of the repaired gray wolf individual is calculated based on the network security constraints, the network security breach penalty function, and the multi-objective optimization function. The three gray wolf individuals with the lowest overall fitness were selected and labeled as follows: Wolf, wolves and Wolves, and mark the remaining gray wolf individuals as Wolf; based on Wolf, wolves and The wolf's position vector is updated using a nonlinear cosine adaptive convergence factor. The wolf's position vector; Determine if the maximum number of iterations has been reached. If not, proceed with a new round of physical constraint repair, comprehensive fitness evaluation, and nonlinear adaptive position update; otherwise, output the current position. The wolf's position vector is used as the optimal scheduling scheme.

6. The computational power-carbon collaborative optimization scheduling method considering computational power task time sequence reconstruction according to claim 5, characterized in that, The formula for calculating the overall fitness is: in, The position vector of the individual gray wolf Overall adaptability , , The weights for economic efficiency, low carbon emissions, and renewable energy consumption targets are respectively. , , These are the position vectors of individual gray wolves. The corresponding comprehensive operating costs of multi-source systems, carbon dioxide emissions, and unused renewable energy power. The position vector of the individual gray wolf Corresponding cybersecurity breach penalties , , These are the comprehensive operating costs, carbon dioxide emissions, and unconsumed renewable energy power of the preset reference scheme.

7. The computational power-carbon collaborative optimization scheduling method considering computational power task time sequence reconstruction according to claim 5, characterized in that, The expression for the nonlinear cosine adaptive convergence factor is: in, Indicates the current iteration number. The maximum number of iterations, For the first Nonlinear cosine adaptive convergence factor in the next iteration.

8. A computing-power-carbon collaborative optimization scheduling device that considers computing power task time sequence reconstruction, characterized in that, include: The acquisition module is used to acquire operational data from multi-source systems. The first construction module is used to construct a multi-source system model based on the multi-source system operation data and determine the constraints of the multi-source system model. The multi-source system model is used to characterize the computing power center's computing power task time sequence reconstruction boundary based on latency characteristics, the operation characteristics of pumped storage power stations, and the output characteristics of thermal power and new energy. The second construction module is used to construct network security constraints based on the multi-source system model and dynamic reactive power support, and to determine the network security breach penalty function. The third construction module, based on the multi-source system operation data, the multi-source system model, and the network security breach penalty function, constructs a multi-objective optimization function that coordinates economic efficiency, low carbon emissions, and new energy consumption capacity. The solution module is used to solve the optimal scheduling scheme using the improved gray wolf optimization algorithm. The improved gray wolf optimization algorithm performs physical constraint repair on the position vector of gray wolf individuals based on the constraints, and performs a comprehensive fitness evaluation on the repaired position vector of gray wolf individuals based on the network security constraints, the network security breach penalty function, and the multi-objective optimization function.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the computational power-carbon collaborative optimization scheduling method for computing power task timing reconstruction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the computational, power, and carbon collaborative optimization scheduling method for computing power and task timing reconstruction as described in any one of claims 1 to 7.

Images