Method and system for regional power grid new energy consumption based on computing power-electric power space-time cooperation

Abstract

The application discloses a kind of based on computing power-power space-time coordination regional power grid new energy consumption method and system.The method comprises: obtaining the real-time operating state data of regional power grid and the data center cluster pending computing power task data;Based on the time sensitivity of task, the computing power task is divided into flexible load, and its energy conversion relationship is determined;Based on real-time operating state and energy conversion relationship, the cross-regional migration amount of flexible load and the output state of generator set are used as decision variable, and the cooperative scheduling instruction containing computing power task cross-regional migration path and generator set power adjustment is generated;In response to the instruction, the flexible load of load center area is migrated to data center across regions through communication network for execution, and the output of corresponding regional generator set is adjusted synchronously.The application is used to solve the problem of serious new energy curtailment due to limited physical power transmission channel and mismatch between power load and new energy resources in spatial distribution in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system operation control, and more particularly, to a regional power grid new energy consumption method and system based on algorithm-power space-time coordination. BACKGROUND

[0002] The existing regional power grid new energy consumption technology mainly relies on the transmission capacity of the physical power grid and the regulation capacity of the power supply side. Specifically, the commonly used technical solutions include: first, constructing an ultra-high voltage (UHV) transmission channel, trying to transport the surplus power of the energy-rich area to the load center through physical expansion; second, configuring large-scale energy storage power stations (such as pumped storage, electrochemical energy storage) or carrying out flexible transformation of thermal power units, and smoothing the random fluctuations of new energy output through "time shifting"; third, carrying out traditional demand side response, guiding local users to adapt to the power grid operation state through peak clipping and valley filling or interrupting load. On the data center side, the existing technology mainly focuses on reducing the power usage effectiveness (PUE) value, such as using liquid cooling technology or waste heat recovery, or carrying out static site selection optimization in the planning stage.

[0003] However, the above existing technical solutions have obvious limitations in dealing with large-scale new energy consumption. First, the construction of physical transmission channels has a long construction period, high investment cost, and physical constraints such as thermal stability limit. When the channel is congested, the sending end has to abandon wind and light, resulting in the physical jam problem of "sending electricity but not out"; second, the configuration cost of energy storage resources is high, and it is difficult to completely consume large-scale surplus power; third, traditional load regulation is mainly based on the logic of "local consumption" or "time transfer", lacking spatial flexibility, and the existing data center scheduling system and power grid scheduling system are mutually isolated, usually regarding the data center as a rigid load that cannot be moved, and failing to take advantage of the characteristics of low delay transmission of algorithm tasks across the network. This passive mode of "electricity following algorithm" leads to serious waste of regulation resources and cannot effectively solve the new energy consumption bottleneck caused by geographical spatial mismatch. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a regional power grid new energy consumption method based on algorithm-power space-time coordination, which identifies flexible algorithm load according to time sensitivity and migrates it from the load center to the data center for execution across the region using the communication network, while adjusting the corresponding generator output of the region, to solve the problem of difficult new energy consumption caused by limited physical transmission channels and spatial mismatch between energy supply and demand.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] The method for new energy consumption of regional power grid based on space-time coordination of computing power and power includes the following steps: obtaining real-time operation state data of the regional power grid and data center cluster pending computing power task data; based on the computing power task data, dividing the computing power task into flexible load according to the time sensitivity of the task, and determining the energy consumption conversion relationship required for computing power task execution; based on the real-time operation state data and the energy consumption conversion relationship, taking the cross-regional migration amount of flexible load and the output state of generator set as the collaborative decision variable, and generating a collaborative scheduling instruction containing the cross-regional migration path of computing power task and the power adjustment of generator set; in response to the collaborative scheduling instruction, migrating the flexible load located in the load center region to the data center for execution through the communication network across regions, and synchronously adjusting the output of the corresponding regional generator set.

[0007] In a preferred embodiment, the determining of the energy consumption conversion relationship required for computing power task execution includes: determining the basic no-load power consumption of the data center in the no-load state; determining the dynamic incremental power consumption required for processing a unit data volume according to the calculation type of the computing power task; multiplying the basic no-load power consumption and the dynamic incremental power consumption by the sum of the computing power task load amount to obtain the real-time power load value required for the data center to consume the computing power task.

[0008] In a preferred embodiment, the generating of the collaborative scheduling instruction containing the cross-regional migration path of computing power task and the power adjustment of generator set includes: constructing a collaborative scheduling model with the goal of minimizing the comprehensive operation cost and maximizing new energy consumption; the collaborative scheduling model includes: calculating the total cost of computing power and power scheduling, which includes the operation fuel cost of conventional generator set, the penalty cost of abandoned wind and light due to the inability to consume new energy, and the network bandwidth rental cost and delay penalty cost generated by the cross-regional transmission of computing power task; using a multivariate constraint optimization algorithm, the collaborative scheduling model is solved in the solution space that meets the power system operation constraints and computing power network constraints, to obtain the computing power task allocation scheme and unit output scheme that minimizes the total cost.

[0009] In a preferred embodiment, the power system operation constraints are used to ensure that the power network is safely operated within the physical transmission limit, including node power balance constraints, line transmission capacity constraints, and unit operation state constraints; the computing power network constraints are used to ensure that the amount of tasks migrated to the data center does not exceed the upper limit of the export network bandwidth of the data center, and the sum of the total processing time and network transmission time of the task does not exceed the maximum delay time allowed for the task.

[0010] In a preferred embodiment, the dividing of the computing power tasks into flexible loads according to the time sensitivity of the tasks comprises: obtaining the deadline requirement and the estimated processing time of the computing power task; calculating the delay tolerance of the task, the delay tolerance representing the ratio between the longest allowed waiting time of the task and the estimated processing time; and marking the task with a delay tolerance higher than a preset threshold as a flexible load that can be migrated across regions.

[0011] In a preferred embodiment, the generating of the cooperative scheduling instruction comprising the computing power task cross-region migration path and the generator power adjustment comprises: when detecting that there is new energy curtailment in the first region and there is flexible load demand in the second region, comparing the residual capacity of the physical power transmission channel with the available bandwidth of the computing power transmission channel; and if the residual capacity of the physical power transmission channel is insufficient to transmit the required power for absorbing new energy, generating routing instruction for sending the flexible load data of the second region to the data center in the first region for calculation through the optical fiber network in priority.

[0012] In a preferred embodiment, the method for absorbing new energy of the regional power grid is executed in two stages of day-ahead planning and real-time within a day, comprising: in the day-ahead planning stage, formulating the start-stop plan of the conventional unit based on the weather forecast and historical task data of the next day, and reserving the computing redundancy resources of the data center; and in the real-time within a day stage, based on the ultra-short-term new energy output prediction, rolling updating the cooperative scheduling instruction every preset time period, and dynamically adjusting the migration amount of the flexible load according to the instantaneous deviation of the new energy output.

[0013] In a preferred embodiment, the migrating of the flexible load located in the load center region to the data center for execution through the communication network across regions comprises: before the cross-region migration of the computing power task, detecting the congestion state and transmission delay of the communication network between regions; and during the task migration process, if it is monitored that the transmission delay of the communication network exceeds a safety threshold, triggering a fallback mechanism to stop the cross-region migration of subsequent tasks, switching the tasks that have not been migrated back to the local data center for processing, and synchronously triggering the standby adjustment resources of the local power grid.

[0014] The application provides a regional power grid new energy consumption system based on space-time cooperation of computing power and electric power, comprising: a data acquisition module, configured to acquire real-time operation state data of a regional power grid and to-be-processed computing power task data of a data center cluster; a task analysis module, configured to divide computing power tasks into flexible loads according to time sensitivity of the tasks based on the computing power task data, and to determine energy consumption conversion relations required for execution of the computing power tasks; a strategy decision module, configured to take a cross-regional migration amount of the flexible loads and an output state of a generator set as cooperative decision variables based on the real-time operation state data and the energy consumption conversion relations, and to generate a cooperative scheduling instruction containing a computing power task cross-regional migration path and generator set power adjustment; and a cooperative execution module, configured to respond to the cooperative scheduling instruction, and to migrate the flexible loads located in a load center region to a data center for execution through a communication network in a cross-regional manner, and to synchronously adjust the output of the corresponding regional generator set.

[0015] A regional power grid new energy consumption device based on space-time cooperation of computing power and electric power, comprising a memory and a processor: the memory is configured to store a program; the processor is configured to execute the program to realize each step of the method.

[0016] The method has the following technical effects and advantages:

[0017] The method fuses real-time power grid operation states and computing power task data, accurately identifies flexible computing power loads that can be migrated across regions based on time sensitivity of the tasks, and establishes a quantitative relation between the flexible computing power loads and energy consumption; the flexible loads are dynamically migrated from a load center to a data center for execution through a communication network as a virtual power transmission channel, and the output of a generator set is synchronously adjusted. The method has the following advantages: the method breaks the limitation of a fixed geographical position of a traditional electric power load, uses cross-domain transmission of data streams to replace transmission of physical electric currents, and realizes flexible transfer of electric power loads in space; thus, in the case that a physical power transmission channel is limited, electric power demand can still be accurately scheduled to a new energy rich area for local consumption, the spatial mismatch between new energy production and consumption is effectively solved, and the new energy consumption capacity and operation flexibility of the whole network are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A method flowchart of the method is shown in the following figure.

[0019] Figure 2 A cooperative response diagram of physical channel blockage triggering computing power cross-region migration is shown in the following figure.

[0020] Figure 3A comparison chart of day-ahead planning and intraday real-time rolling scheduling curves provided for embodiments of the present invention;

[0021] Figure 4 A schematic diagram of the fallback mechanism for network congestion-triggered task migration circuit breaking provided in an embodiment of the present invention;

[0022] Figure 5 This is a block diagram of a regional power grid renewable energy consumption system based on computing power-electricity spatiotemporal coordination, provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1, Figure 1 This invention presents a method for regional power grid renewable energy consumption based on computing power-electricity spatiotemporal coordination, comprising the following steps:

[0025] S1 acquires real-time operating status data of the regional power grid and pending computing power task data of the data center cluster.

[0026] It should be noted that this step aims to establish a panoramic data perception foundation for the power grid physical system and the computing power information system, specifically through the construction of a two-layer heterogeneous data acquisition interface. On the regional power grid side, the system communicates with the Energy Management System (EMS) and Wide Area Measurement System (WAMS) via the power dispatch data network, utilizing the IEC 60870-5-104 or C37.118 communication protocol to frequently acquire data on the power grid topology, real-time power flow at key transmission sections, real-time output status of conventional thermal power units, and real-time output and ultra-short-term forecast values ​​of wind and solar power plants. This data is then integrated into a set of power grid operating statuses. On the data center side, the system connects to the cluster resource schedulers (such as Kubernetes API Server or the Slurm scheduling system) of each regional data center via a fiber optic broadband network, retrieving metadata from the queue of tasks to be processed in real time. This metadata includes task submission time, the amount of data required for the task, task priority, and the expected completion time set by the task publisher. This data is then integrated into a set of computing power task statuses. To eliminate data latency discrepancies between the two systems, the system performs timestamp alignment and outlier cleanup on the collected raw data, forming a unified time-section state matrix. The system at time... The obtained comprehensive operational status set It is expressed as follows:

[0027] (1)

[0028] In the formula, is The system comprehensive operation state set at the moment; is the real-time active power output vector of each conventional generator in the regional power grid, including the current power generation and climbing state of each unit; is the real-time output vector of the new energy station, including the actual power generation of wind power and photovoltaic and the predicted power in the future one dispatching period; is the real-time active power flow vector of each transmission line and tie line, which is used for subsequent evaluation of the congestion of the physical channel; is the to-be-processed computing task queue vector of the data center cluster, wherein each element includes a unique identification ID, an input data volume and a deadline timestamp of a single computing task.

[0029] In order to intuitively illustrate the data structure obtained in this step, Table 1 shows the real-time state snapshot of the regional power grid nodes and data center nodes at a moment. Through the state matrix, the system can simultaneously master the physical surplus and deficiency of the power side and the load demand of the computing power side.

[0030] Table 1

[0031]

[0032] S2, based on the computing task data, dividing the computing task into flexible load according to the time sensitivity of the task, and determining the energy consumption conversion relationship required for the execution of the computing task.

[0033] It should be noted that the traditional power dispatching scheme usually regards the data center as a whole, rigid load node, only focuses on the fluctuation of the total power, and ignores the fact that the tasks running in the data center actually have significant differences and adjustment potential. This "black box" management method cannot distinguish which energy consumption must be met immediately and which energy consumption can be transferred in time and space, resulting in a great waste of adjustment resources. This step breaks through the limitations of the traditional perspective by deeply analyzing the time attributes of the computing task, accurately identifies the "flexible" part that can be dispatched across regions with the power grid, and thus changes the data center from a passive power consumer to a flexible resource that can actively participate in power grid interaction, greatly improving the adjustment freedom of the system.

[0034] The computing task is divided into flexible load according to the time sensitivity of the task, including:

[0035] Get the deadline requirement and estimated processing time of the computing task. The system first reads the service level agreement (SLA) parameters in the task metadata, extracts the first absolute deadline of the computing task and an estimated processing duration derived from historical statistics of similar tasks;

[0036] Then, the delay tolerance of the task is calculated, which represents the ratio of the maximum waiting time allowed by the task to the estimated processing duration, specifically, the ratio of the maximum floating time window of the task under the premise of not breaching the contract to the actual execution time, and the delay tolerance The calculation formula is as follows:

[0037] (2)

[0038] In the formula, is the delay tolerance of the i-th computing task; is the deadline of the task; is the current system decision time; is the estimated processing duration required by the task; represents the total remaining time of the task from the deadline.

[0039] Subsequently, tasks with a delay tolerance higher than a preset threshold are marked as flexible loads that can be migrated across regions, and a time sensitivity threshold is set. When , it indicates that the task has sufficient waiting time to cover the additional delay caused by cross-regional network transmission, so it is marked as a “flexible load” and added to the migration queue; otherwise, it is marked as a “rigid load” and must be processed immediately.

[0040] The conversion relationship of the energy consumption required by the computing task includes:

[0041] The basic no-load power consumption of the data center in the no-load state is determined, which does not fluctuate instantaneously with the change of the computing task quantity, mainly including the minimum power to maintain the server in the on state, the basic operation power of the refrigeration system, and the energy consumption of lighting and security;

[0042] Then, the dynamic incremental power consumption required to process a unit of data volume is determined according to the calculation type of the computing task. Different types of tasks (such as AI training with intensive floating point operations and database query with intensive I / O) have different energy consumption requirements for CPU / GPU. The system matches the corresponding energy efficiency coefficient according to the task type;

[0043] Finally, the sum of the basic no-load power consumption and the dynamic incremental power consumption multiplied by the computing task load is taken as the real-time power load value required by the data center to absorb the computing task. This conversion relationship realizes the quantitative mapping from “bit stream” to “watt stream”, and the real-time power load value of the i-th data center in​ Real-time power load at a moment The calculation formula is as follows:

[0044] (3)

[0045] In the formula, is the total power of the data center; is the basic no-load power consumption of the data center; is the power utilization efficiency coefficient of the data center; is the current task set allocated in the data center executed; is the dynamic energy consumption coefficient of a specific type of task processing unit data volume (unit: kWh / GB or kWh / TFLOPS); is the data load of the i th task. This step realizes the conversion of the originally abstract computer task into a virtual power load that can be perceived, quantified and controlled by the power grid dispatching system by establishing an accurate task elasticity evaluation model and an energy consumption conversion model, thereby laying a physical foundation for subsequent "computing with electricity" collaborative scheduling.

[0046] S3, based on the real-time running state data and the energy consumption conversion relationship, taking the cross-regional migration amount of the flexible load and the output state of the generator set as the collaborative decision variables, generating a collaborative scheduling instruction containing the cross-regional migration path of the computing power task and the power adjustment of the generator set.

[0047] It should be noted that when facing the blockage of the inter-regional power transmission channel, the traditional power dispatching scheme is often limited by the topological constraints of the physical power grid and can only be forced to cut off the surplus new energy (abandoning wind and light) of the sending end power grid or limit the load demand (orderly power consumption) of the receiving end power grid. This "grid-centered" passive adjustment method not only causes a huge waste of clean energy, but also increases the overall operation cost of the system. The method of this step introduces "computing power flow" as a virtual energy carrier that can cross the boundary of the physical power grid, uses the extremely high transmission bandwidth and extremely low transmission energy consumption of the optical fiber network to "transport" the power demand that must be consumed at the load center to the new energy-rich area for local consumption, thereby effectively breaking through the bottleneck of the power transmission capacity without increasing the investment in physical power transmission lines, and realizing the optimal allocation of the whole network resources in a larger time and space range.

[0048] The generation of the collaborative scheduling instruction containing the cross-regional migration path of the computing power task and the power adjustment of the generator set includes:

[0049] The generation of the collaborative scheduling instruction containing the cross-regional migration path of the computing power task and the power adjustment of the generator set includes:

[0050] ​A collaborative scheduling model is constructed to minimize the comprehensive operation cost and maximize the new energy consumption, and the collaborative scheduling model comprises: a total cost of computing power and power scheduling, the total cost covering an operation fuel cost of a conventional generating unit, a penalty cost of abandoned wind and light due to the inability to consume new energy, and a network bandwidth rental cost and a delay penalty cost generated by the cross-regional transmission of computing power tasks, and a target function of the total cost The calculation formula of the target function of the total cost is as follows:

[0051] (4)

[0052] In the formula, is a scheduling period; is a number of conventional generating units; is an active power output of the unit at the time t; is a start-stop state variable (1 represents start, and 0 represents stop) of the unit at the time t; is a consumption characteristic coefficient of the unit; is a single start cost of the unit, is used to identify whether the start action of the unit occurs; is a number of new energy stations; is a unit penalty coefficient of abandoned electricity; is abandoned electricity power of the new energy station; is a number of data center nodes; is computing power task data volume migrated from a source data center to a target data center at the time t; is a network bandwidth transmission fee of unit data volume; is a delay time penalty coefficient; is a network transmission delay of the node to the node. A multivariate constraint optimization algorithm is used to solve the collaborative scheduling model in a solution space meeting power system operation constraints and computing power network constraints, to obtain a computing power task allocation scheme and a unit output scheme that minimize the total cost, and specifically, a mixed integer linear programming (MILP) algorithm or a particle swarm optimization (PSO) algorithm is used to perform iterative optimization on the computing power migration volume and the unit output as joint optimization variables.

[0053]

[0054] ​​​​​​​​​​​​The power system operation constraints are used to ensure the safe operation of the power network within physical transmission limits, specifically covering node power balance constraints, line transmission capacity constraints, and unit operating status constraints, namely:

[0055] Establish nodal power balance equations based on a DC power flow model. For any node in the power grid... The injected power (generation power and virtual increment) and outflow power (load and line power flow) must be balanced in real time. The formula for the node power balance constraint is as follows:

[0056] (5)

[0057] In the formula, To connect at node A collection of generating units or stations; Contribute to new energy forecasting; This refers to the amount of abandoned electricity. For nodes The basic rigid load; This is the equivalent injected power (negative for migration in, positive for migration out) caused by the migration of computing power tasks to the data center power of this node. and They are nodes and adjacent nodes The voltage phase angle; This refers to the line reactance.

[0058] It should be noted that, This is the equivalent value of the change in net injected power of the node caused by the cross-region migration of computing power tasks, and it is related to the computing power migration variable. The relationship is defined as follows:

[0059] (6)

[0060] That is, when computing power tasks are generated from nodes When migrating out, the local load decreases, which is equivalent to injecting positive power into the power grid (migrating out is positive); when computing tasks migrate into the node... When the local load increases, it is equivalent to injecting negative power into the power grid (migration is negative). This is the energy consumption conversion factor.

[0061] The formula for the line transmission capacity constraint is as follows:

[0062] (7)

[0063] In the formula, For the line Thermal stability limit of transmission capacity.

[0064] The unit operation state constraint formula is as follows:

[0065] (8)

[0066] (9)

[0067] In the formula, is a unit start-stop state variable (0 or 1), which ensures that when (shut down), the unit output must be 0; when (start up), the output is between the minimum technical output and the maximum output . is the unit ramp rate limit.

[0068] The computing power network constraint is used to ensure that the amount of tasks migrated to the data center does not exceed the upper limit of the export network bandwidth of the data center, and the sum of the total processing time of the tasks and the network transmission time does not exceed the maximum delay time allowed for the task. The specific constraint formula is as follows:

[0069] (10)

[0070] (11)

[0071] In the formula, is the maximum receiving bandwidth of the target data center ; is the scheduling period length; is the data processing rate of the target data center; is the link transmission rate; is the maximum allowed delay of the task .

[0072] Based on the solving logic of the above optimization model, the embodiment is not a simple conditional judgment on the specific space scheduling strategy, but a routing optimization mechanism based on shadow price (Shadow Price) guidance. The specific implementation details are as follows:

[0073] During the solving process, the system calculates the dual variable (Dual Variable) or Lagrange multiplier of each transmission line capacity constraint in the power system operation constraint condition in real time. When the physical power transmission channel from the first region (new energy rich area) to the second region (load center) does not reach the limit, the dual variable is zero, and the power transmission cost is only the network loss; when the physical power transmission channel reaches the thermal stability limit (i.e. congestion occurs), the dual variable of the constraint rapidly rises, meaning that the marginal cost of transmitting an additional unit of power through the physical line tends to infinity.

[0074] At this point, the optimization algorithm will automatically compare the computing power transmission cost (i.e. ) and the marginal cost of congestion. As long as the generalized cost of migrating computing power tasks is lower than the cost of power-wasting penalties caused by physical congestion ( The algorithm will then transfer computing power. (From the load center) To the rich area () is a non-zero preferred solution.

[0075] In practice, the system generates routing instructions based on the following logic:

[0076] 1) Read the physical line at the current time Trend value and limit value ;

[0077] 2) If Prioritize increasing physical delivery and maintain local execution of computing tasks. );

[0078] 3) If Lock the physical transmission power and start the computing power flow calculation;

[0079] 4) Calculate the unit load from the area Migrate to Comprehensive cost and the area's power curtailment penalty contrast;

[0080] 5) If Then let Its value is determined by the objective function. The gradient descent direction determines the path until the bandwidth limit is reached. Or it can absorb all the surplus electricity.

[0081] Among them, the comprehensive cost This refers to the specific quantification of the aforementioned computing power transmission cost, representing the migration cost per unit of data volume, and its calculation formula is as follows: ;

[0082] The benefits of the power abandonment penalty The formula for calculating the marginal loss incurred when the corresponding amount of electricity cannot be consumed locally and is therefore forcibly abandoned is as follows: .

[0083] To verify the effectiveness of the aforementioned shadow price-based routing mechanism, this embodiment simulates a typical congestion mitigation process, and the results are as follows: Figure 2 As shown in the figure, the upper curve represents the real-time power flow changes of the physical power transmission channel (West-to-East Power Transmission), and the lower curve represents the cross-regional migration rate of the generated computing power tasks.

[0084] From Figure 2 it can be seen that at moment, the physical channel flow reaches the thermal stability limit, at which the congestion shadow price increases sharply; the optimization algorithm responds quickly, and the migration rate of the lower power increases from 0 to 50 Gbps, indicating that the system has "bypassed" part of the energy demand through the fiber network. With the migration of the power load, the power demand of the load center decreases, and the flow pressure of the physical channel is relieved at moment, proving the decoupling effect of collaborative scheduling.

[0085] In addition, in order to cope with the randomness of new energy output fluctuation, the generation of the above collaborative scheduling instruction is not a one-time static process, but is executed in stages relying on a multi-time scale rolling coordination mechanism. The objective function of the above collaborative scheduling model is decomposed into day-ahead planning and intra-day scheduling two stages for progressive solving, and the specific implementation steps are as follows:

[0086] 1) In the day-ahead planning stage, the system is based on the next 24-hour new energy power prediction curve and historical power task data to minimize the total cost of the whole day as the goal, and solve the collaborative scheduling model for a long period. The core purpose of this solving is to determine the start-stop state variable (0 / 1 binary variable) of the conventional generator, that is, to decide which units to keep on in the next day to provide inertia support. At the same time, the system reserves the amount of computing resources and network bandwidth of the data center in the new energy rich area according to the predicted new energy surplus distribution, and under the constraint of the objective function , it locks the amount of computing resources and network bandwidth reserved for the next day's "power migration" in advance, and forms the basic scheduling plan for the next day;

[0087] 2) In the intra-day scheduling stage, the system introduces model predictive control (MPC) technology, based on higher time resolution (such as 15-minute level) ultra-short-term new energy prediction data, under the premise that the unit start-stop state is fixed in the day-ahead stage, the collaborative scheduling model is solved in a high-frequency rolling manner. Specifically, the system opens a backward extending finite prediction time window (for example, the next 4 hours) every preset time period (for example, 15 minutes), and minimizes the objective function again in this window, focusing on the real-time active power output of the generator and the cross-regional migration amount The two continuous decision variables are finely optimized. After the solution is completed, the system only issues the current first time section coordination scheduling instruction to the execution mechanism, and the optimization results of the subsequent period are input as the initial state for the next round of calculation. Through this rolling update mechanism, when the new energy output deviates from the predicted value, the optimization algorithm can dynamically increase or decrease the data flow to the new energy-rich area by adjusting the values of the variables in real time, thereby suppressing the impact of prediction errors on the system in the time dimension.

[0088] The modification effect of the multi-time scale rolling mechanism on the scheduling instruction is shown in Figure 3 . The dashed line in the figure represents the benchmark power consumption plan of the western data center formulated in the day-ahead planning stage, and the solid line represents the actual execution curve in the real-time rolling stage. It can be seen that at about 14:00, due to the sudden increase of the actual output of the western wind power (positive deviation) higher than the predicted value, the real-time rolling mechanism dynamically increases the amount of flexible computing power load migrated to this area (the solid line is higher than the dashed line), thereby realizing the "instantaneous consumption" of the sudden surplus wind power and avoiding the occurrence of wind curtailment.

[0089] This step expands the traditional single-dimensional power dispatch to "computing power-power" multi-dimensional coordinated dispatch by constructing an optimization dispatch model deeply coupling computing power and power, and combining dynamic routing strategies in space and rolling correction mechanisms in time. Under the premise of ensuring the safety of the power grid and the quality of computing power services, the transmission channel resources and new energy generation resources of the entire network are maximally utilized.

[0090] S4, in response to the coordinated scheduling instruction, migrating the flexible load located in the load center region to the data center across regions through the communication network and synchronously adjusting the output of the corresponding regional generator set.

[0091] It should be noted that this step is the closed-loop execution link of "computing power-power" coordinated dispatch. The system first decouples the coordinated scheduling instruction generated by S3 into computing power dispatch sub-instructions and power dispatch sub-instructions. For the computing power side, the system issues routing strategies through a software-defined network (SDN) controller and a cloud resource management platform, establishes data transmission tunnels from the load center (such as the eastern node) to the new energy-rich area (such as the western node), and starts virtual machine migration or container image pulling operations; for the power side, the system issues new active power base points to the relevant generator sets through an automatic generation control (AGC) system. Specifically, at the sending end (new energy-rich area), due to the reception of additional computing power load, the system will order to increase the output upper limit of the wind and light station in this area or reduce the pressure drop amplitude of the thermal power generator set, thereby realizing the local consumption of new energy; at the receiving end (load center), due to the migration of part of the computing power load, the system will order to reduce the output of the thermal power generator set in this area, thereby reducing the consumption of fossil energy.

[0092] The flexible load located in the load center region is migrated to the data center across regions through a communication network for execution, including:

[0093] Before performing the cross-region migration of computing power tasks, the congestion state and transmission delay of the inter-region communication network are detected, specifically, a network probe is used to send detection data packets in real time, the round-trip time (RTT) and packet loss rate of the link are measured, and the port bandwidth utilization of the backbone network router is queried. Only when the link state meets the quality of service (QoS) requirements, the data transmission channel is opened.

[0094] During the task migration process, if it is monitored that the transmission delay of the communication network exceeds the safety threshold, a fallback mechanism is triggered to stop the cross-region migration of subsequent tasks, switch the un-migrated tasks back to the local data center for processing, and synchronously trigger the standby regulation resources of the local power grid. Specifically, the system sets a network delay safety threshold , and continuously monitors the real-time delay during transmission . When the delay exceeds the safety threshold for a certain period of time, it is determined that the network link is congested or faulty, at which time the system immediately performs a "fusing" operation: first, interrupt the sending of new data packets to the remote location, and re-press the remaining to-be-processed tasks into the job queue of the local data center, to ensure that the tasks are not lost; second, due to the return of the tasks, the local power load rises, and the power dispatching system quickly calls the local spinning reserve capacity (Spinning Reserve) or fast start-stop unit to fill the power gap, to ensure the stability of the local power grid frequency.

[0095] The triggering process of the communication link state awareness and fallback mechanism is shown in Figure 4 . The figure shows the real-time relationship between network transmission delay and task migration throughput. At time, the network delay suddenly increases and exceeds the safety threshold (50 ms), triggering the fusing protection logic of the system, and the task migration throughput automatically drops to zero within milliseconds, effectively preventing the accumulation and timeout of tasks in the congested network.

[0096] This step solves the risk problem of high dependence of computing power business on network communication quality by establishing a double protection mechanism of "pre-detection" and "in-process fusing", ensuring that the system can smoothly return to the "local autonomous mode" from the "cross-domain collaborative mode" when the communication network fluctuates, and guaranteeing the continuity of data business and the safe and stable operation of the power system.

[0097] Embodiment 2, Figure 5 a new energy consumption system of regional power grid based on computing power-electricity space-time collaboration is given, including:

[0098] a data acquisition module configured to acquire real-time operation state data of a regional power grid and to-be-processed computing power task data of a data center cluster;

[0099] a task analysis module configured to divide computing power tasks into flexible loads according to time sensitivity of the tasks based on the computing power task data, and to determine an energy consumption conversion relationship required for execution of the computing power tasks;

[0100] a strategy decision module configured to take a cross-regional migration amount of the flexible loads and an output state of a generator unit as cooperative decision variables based on the real-time operation state data and the energy consumption conversion relationship, and to generate a cooperative scheduling instruction including a computing power task cross-regional migration path and a generator unit power adjustment;

[0101] a cooperative execution module configured to migrate the flexible loads located in a load center region to a data center for execution across regions through a communication network and to synchronously adjust the output of the generator units in corresponding regions in response to the cooperative scheduling instruction.

[0102] Embodiment 3: A regional power grid new energy consumption device based on space-time cooperation of computing power and power, comprising a memory and a processor: the memory is configured to store a program; the processor is configured to execute the program to implement any of the embodiments of Embodiment 1.

[0103] Since the regional power grid new energy consumption device based on space-time cooperation of computing power and power introduced in the embodiment is a device used to implement the method in Embodiment 1 of the present application, the specific implementation of the electronic device of the present embodiment and its various forms can be understood by those skilled in the art based on the method introduced in Embodiment 1 of the present application, so the method of the electronic device in the present application will not be described in detail. As long as the device used to implement the method in the present application is implemented by those skilled in the art, it belongs to the scope of the present application.

[0104] The above formulas are dimensionless values calculated, and the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0105] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0106] Those skilled in the art can understand that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0107] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0108] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0109] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for regional power grid renewable energy consumption based on computing power-electricity spatiotemporal coordination, characterized in that, Includes the following steps: Acquire real-time operational status data of the regional power grid and pending computing power task data of the data center cluster; Based on the time sensitivity of computing task data, computing tasks are divided into flexible loads, and the energy consumption conversion relationship required for the execution of computing tasks is determined. Using the cross-regional migration of flexible loads and the output status of generator units as collaborative decision variables, and based on real-time operating status data and energy consumption conversion relationships, a collaborative scheduling instruction is generated that includes the cross-regional migration path of computing tasks and the adjustment of generator unit power. This includes: calculating the dual variables of the capacity constraints of each transmission line in the power system operation constraints; when it is detected that there is renewable energy curtailment in the first region and flexible load demand in the second region, and the physical transmission channel between the first region and the second region reaches its limit, causing the corresponding dual variable to be greater than zero, comparing the remaining capacity of the physical transmission channel with the available bandwidth of the computing power transmission channel; if the remaining capacity of the physical transmission channel is insufficient to transmit and absorb the electricity required by renewable energy, and the generalized cost of computing task migration is lower than the curtailment penalty cost caused by physical congestion, then a routing instruction is preferentially generated to migrate the flexible load data of the second region to the data center of the first region for calculation through the communication network. The system executes coordinated scheduling commands to migrate flexible loads in the load center area to the data center for processing across regions via the communication network, and simultaneously adjusts the output of generator units in the corresponding areas.

2. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 1, characterized in that, The determination of the energy consumption conversion relationship required for the execution of the computing task includes: Determine the baseline no-load power consumption of the data center; Determine the dynamic incremental power consumption required to process a unit of data based on the type of computation of the computing task. The sum of the base idle power consumption and the dynamic incremental power consumption multiplied by the computing power task load is used as the real-time power load value required by the data center to absorb computing power tasks.

3. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 1, characterized in that, The generation of collaborative scheduling instructions, which includes cross-regional migration paths for computing tasks and adjustments to generator power, includes: A collaborative scheduling model is constructed with the goal of minimizing the overall operating cost and maximizing the absorption of new energy sources. The overall operating cost includes the fuel cost of the power system, the penalty cost for abandoning new energy sources, and the cost of cross-domain migration of computing tasks. Under the constraints of power system operation and computing network, the collaborative scheduling model is solved to obtain the optimal cross-regional allocation scheme of computing tasks and generator output scheme.

4. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 3, characterized in that, The power system operation constraints are used to ensure the safe operation of the power network within the physical transmission limits, including power balance constraints, line transmission capacity constraints, and unit operating status constraints. The computing power network constraints are used to ensure that the amount of tasks migrated into the data center does not exceed the upper limit of the data center's outbound network bandwidth, and that the sum of the processing latency and network transmission latency of each computing power task does not exceed the maximum allowed latency time for that task.

5. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 1, characterized in that, The time sensitivity of computing power task data is used to classify computing power tasks into flexible loads, including: Obtain the deadline requirements and estimated processing time for computing power tasks; Calculate the latency tolerance of the task, which represents the ratio of the longest allowed waiting time to the estimated processing time; Tasks with latency tolerance exceeding a preset threshold are marked as flexible loads that can be migrated across regions.

6. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 1, characterized in that, The regional power grid renewable energy consumption method also includes two phases: day-ahead planning and intraday real-time implementation. Specific implementation methods include: During the current planning phase, based on future cycles of new energy output forecasts and computing power demand forecasts, generator start-up and shutdown plans and data center resource reservation plans are formulated. During the intraday scheduling phase, based on ultra-short-term renewable energy output forecasts, the coordinated scheduling instructions are periodically updated and the cross-regional migration of flexible loads is dynamically adjusted according to the real-time deviation of renewable energy output.

7. The regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination according to claim 1, characterized in that, The process of migrating flexible loads from the load center area to the data center for processing across regions via a communication network includes: Before and during the cross-regional migration of computing power tasks, monitor the transmission latency of the inter-regional communication network; When the transmission delay is detected to exceed a preset safety threshold, a rollback mechanism is triggered. The rollback mechanism includes: suspending the cross-regional migration of subsequent tasks, switching unmigrated tasks to the local data center for processing, and triggering the deployment of backup regulation resources of the local power grid.

8. A system using the regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire real-time operating status data of the regional power grid and pending computing power task data of the data center cluster; The task analysis module is used to divide computing tasks into flexible loads based on the time sensitivity of computing task data, and determine the energy consumption conversion relationship required for the execution of computing tasks. The strategy decision module is used to generate collaborative scheduling instructions that include the cross-regional migration path of computing tasks and the power output status of generator sets, based on real-time operating status data and energy consumption conversion relationships, using the cross-regional migration amount of flexible loads and the output status of generator sets as collaborative decision variables. The collaborative execution module is used to execute collaborative scheduling instructions, which migrate flexible loads in the load center area to the data center for processing across regions through the communication network, and synchronously adjust the output of generator sets in the corresponding areas.

9. A computer device, characterized in that, The device includes a processor and a memory: The memory is used to store computer programs and send the instructions of the computer programs to the processor; The processor executes, according to the instructions of the computer program, a regional power grid renewable energy consumption method based on computing power-electricity spatiotemporal coordination as described in any one of claims 1-7.

Images