A computing power-electricity collaborative scheduling method based on a dynamic barrier option model

Abstract

The application discloses a kind of computing power-power collaborative scheduling method based on dynamic barrier option model, including the following steps: step S1. computing power task preliminary screening and gradient product pricing;Step S2. establish and utilize dynamic barrier option model, calculate instantaneous expected net income difference;Step S3. double finite state machine smooth delivery: FSM-B is executed based on instantaneous expected net income difference and the state transition control of time sequence physical constraint parameter.This application discloses the traditional hardware depreciation concept, focuses on quantifying the consistency repair cost in distributed computing and exponential default fine, so that computing power center can accurately calculate profit and loss balance point when participating in peak shaving, to avoid inverted loss.The application introduces dynamic barrier option model, and complex algorithm network interaction is converted into scientific financial decision, and the risk premium generated by computing power task over time can be accurately described by the dynamically floating barrier point, to maximize the arbitrage space of data center.

Description

Technical Field

[0001] This invention relates to the field of power peak-shaving and dispatching technology, specifically a computing power-power collaborative dispatching method. Background Technology

[0002] With the advancement of the construction of new power systems, data centers, as high-quality, flexible, and high-energy-consuming loads, have become an important way to improve the flexibility of the power grid by participating in ancillary services such as grid peak shaving.

[0003] Current research on data centers' participation in power grid dispatch has made some progress. For example, the paper "Collaborative Optimization of Data Centers and Power Systems under the Background of Energy Internet (I): Data Center Energy Consumption Model" analyzes the scheduling flexibility of workloads in the spatiotemporal dimensions; the paper "A Two-Level Economic Dispatch Model Considering the Potential for Adjusting Data Center Power Load" constructs a two-way collaborative economic dispatch model.

[0004] Existing technologies still have the following limitations and scheduling logic oscillation issues:

[0005] First, the physical cost of computing power load participating in peak shaving is not accurately quantified, and even ignores the micro-physical cost of computing power interruption: existing models mostly regard computing power as an ideal elastic load, and do not fully consider the consistency repair cost (such as memory dump, state rollback calculation overhead) and overdue default risk caused by forced suspension of tasks.

[0006] Second, the lack of an arbitrage mechanism: When computing power participates in peak shaving, there is a lack of a financial arbitrage mechanism to adapt to grid fluctuations. The failure to scientifically quantify the marginal risk premium of computing power centers during the scheduling process leads to blind arbitrage decisions.

[0007] Third, the lack of an effective logical isolation mechanism makes it impossible to accurately respond to peak shaving signals: When computing load participates in peak shaving, the underlying scheduling lacks a smooth delivery strategy at the business level and lacks an effective logical isolation mechanism, making it difficult to ensure the SLA of high-priority tasks while achieving an accurate response of elastic tasks to peak shaving signals. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, the present invention aims to provide a computing power-power coordinated scheduling method based on a dynamic barrier option model. This method, by introducing a dynamic barrier option model, transforms complex computing network interactions into scientific financial decisions. The dynamically rising barrier points can accurately describe the risk premium generated by computing power tasks over time, maximizing the arbitrage space of data centers.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0010] A computing power-electricity coordinated dispatch method based on a dynamic barrier option model includes the following steps:

[0011] Step S1. Initial Screening of Computing Power Tasks and Tiered Product Pricing: Receive computing power task requests submitted by users, extract the spatiotemporal feature parameters of the tasks, including maximum latency tolerance, deadline, data volume, and estimated remaining execution time; calculate the consistency repair cost when the task is suspended based on the spatiotemporal feature parameters, the consistency repair cost consists of the physical cost of memory dumping and the computing power overhead of state rollback; introduce an exponential penalty model based on the overdue ratio to calculate the expected penalty; extract the interruptible surplus time of the tasks, construct an envelope function, and based on the comparison result of the envelope function and the dynamic adaptive system security threshold, identify and encapsulate the tasks into either Class A rigid operation products or Class B interruptible products, and generate tiered pricing discounts for Class B interruptible products;

[0012] Step S2. Establish and utilize a dynamic barrier option model to calculate the instantaneous expected net return difference: map the real-time peak-shaving subsidy price of the power grid to the underlying asset price of the option, and map the sum of the real-time consistency repair cost and the default penalty risk to the strike price; construct a dynamic barrier point that increases over time, which is generated based on the dynamic time penalty coefficient calculated by the surplus time ratio; construct an arbitrage trigger absolute threshold by superimposing a basic target net profit threshold; the underlying scheduler polls the real-time peak-shaving subsidy price of the power grid at high frequency, and the difference between the real-time peak-shaving subsidy price of the power grid and the arbitrage trigger absolute threshold is the instantaneous expected net return difference;

[0013] Step S3. Smooth Delivery with Dual Finite State Machines: Dual finite state machines FSM-A and FSM-B are instantiated in parallel at the underlying computing node. For Class A rigid operation products, FSM-A performs interruption shielding calculations through a state lock-in mechanism to shield the network-side peak-shaving signal. For Class B interruptible products, FSM-B performs state transition control based on the instantaneous expected net revenue difference and time-series physical constraint parameters obtained in Step S2, and achieves smooth scheduling through business logic isolation.

[0014] Furthermore, in step S1, the formula for calculating the consistency repair cost is:

[0015] ,

[0016] in, For the cost of consistency restoration, The physical cost of memory dumping The computational cost for state rollback;

[0017] The calculations for the physical cost of memory dumping and the computational cost of state rollback are as follows:

[0018] ,

[0019] In the formula, This refers to the amount of memory-resident data when the task is suspended. Write the energy consumption constant per unit of data volume. To preset the average nodal marginal electricity price within the real-time scheduling period, This is the physical impact penalty coefficient. This refers to the cost of a single hardware depreciation. To recalculate time for breakpoint recovery, Rated power for computing nodes.

[0020] Further, in step S1, the formula for calculating the expected penalty is:

[0021] ,

[0022] In the formula, For the user's basic contract amount, The actual delivery time of the task. For the deadline, This is the penalty sensitivity coefficient.

[0023] Furthermore, in step S1, the method for constructing the envelope surface function is as follows: extracting the interruptible slack time of the task. Construct the envelope surface function :

[0024] ,

[0025] In the formula, These are the normalized weighting coefficients. It is a very small constant, used to prevent issues arising as the task approaches its deadline. When the value approaches or equals 0, a mathematical singularity with a denominator of zero appears in the calculation of the envelope surface function; a safety threshold is set for the dynamic adaptive system. ,in Based on historical average repair costs, For power grid peak shaving fluctuations, and This is the risk preference coefficient;

[0026] The judgment rule is: if Packaged as a Class A rigid operation product; if It is packaged as a Class B interruptible product.

[0027] Furthermore, in step S1, the method for generating tiered pricing discounts for Class B interruptible products includes:

[0028] Get the maximum latency tolerance Combining the probability density function of the duration of a single peak-shaving interruption in the power grid This leads to the probability that a peak-shaving interruption can be successfully absorbed without triggering a default. ;

[0029] According to the preset profit sharing ratio Generate price discounts for users ; The total subsidy that the system expects to obtain from the grid peak shaving cycle within the maximum latency tolerance authorized by the user;

[0030] Finally, a user-perceived dynamic quote is generated. , This is the standard billing benchmark price for the system's regular Class A rigid computing power service.

[0031] Furthermore, in step S2, the method for constructing the dynamic obstacle point and the absolute threshold for arbitrage triggering includes:

[0032] The real-time consistency repair cost established in step S1 Risk of default penalty The sum is mapped to the absolute physical defense baseline. ;

[0033] Calculate the current time Interruptible surplus time ,in To estimate the remaining execution time, Deadline;

[0034] Construct a dynamic time penalty coefficient based on the computation-save time ratio.

[0035] ,

[0036] in, The minimum constant is the penalty base. A preset constant greater than 0 serves as a scaling factor for the penalty intensity, used to adjust the overall magnitude of the penalty term; nonlinear compression index. The preset exponential sensitivity parameter is significant because it uses an exponential form to make the change in the computation-sufficiency time ratio produce a non-linear amplified penalty effect, causing the decision threshold to rise sharply in the later stages.

[0037] Generate dynamic obstacle points that float in real time. ;

[0038] Overlaying a preset basic target net profit threshold To ensure that data centers can obtain substantial economic benefits from participating in peak shaving, a minimum profit target is set for each arbitrage activity, thus establishing an absolute threshold for arbitrage triggering. ;

[0039] The underlying scheduler frequently polls the real-time peak-shaving subsidy quotes from the power grid. Calculate the instantaneous expected net income difference .

[0040] Furthermore, in step S3, the FSM-A performs the interrupt masking calculation using the state lock-up mechanism as follows:

[0041] ,

[0042] In the formula, This is the current scheduling time; This is the start time for the computing task; The actual (or estimated) completion time of the computing task; This means that all grid peak-shaving signals from the grid side are shielded at the kernel bus level, achieving absolute physical isolation and smooth delivery.

[0043] Further, in step S3, the timing physical constraint parameters include:

[0044] a. Minimum running time constraint : Set the basic runtime constant Minutes; to ensure the mission can be completed between two shutdowns. A valid breakpoint retention period is used to dynamically calculate the minimum actual running time required:

[0045] ,

[0046] In the formula, The average interval between generating a complete snapshot of in-memory data. The fixed time consumed for process context recovery and warm-up;

[0047] b. Minimum downtime constraint Set the basic downtime constant. Minutes, using Newton's law of cooling to calculate the minimum downtime required for safe cooling of the chip:

[0048] ,

[0049] In the formula, The thermal resistance-capacitance time constant of the server cooling system. To trigger the CPU / memory core temperature at the moment of suspension, The low temperature threshold required to allow safe re-energization. This refers to the ambient temperature of the computer room.

[0050] Further, in step S3, the state transition control includes:

[0051] Suspension transition logic: Assume the task's execution duration is... If and only if and At this time, the system enters a suspended state;

[0052] Restore transition logic: Assume the task's downtime is [time value]. If and only if and At that time, the system switches to a resumed running state.

[0053] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described computing power-electricity collaborative scheduling method based on a dynamic obstacle option model.

[0054] The beneficial effects of this invention are as follows:

[0055] This invention abandons the traditional concept of "hardware depreciation" and focuses on quantifying the "consistency repair cost" and "exponential default penalty" in distributed computing, enabling computing centers to accurately calculate the break-even point when participating in peak shaving and avoid inverted losses.

[0056] By introducing a dynamic barrier option model, complex computing network interactions are transformed into scientific financial decisions. The dynamically rising barrier points can accurately describe the risk premium generated by computing power tasks over time, maximizing the arbitrage space of data centers.

[0057] By utilizing a dual finite state machine smooth delivery pattern, complete isolation of A / B class business logic is achieved. By replacing dangerous kernel-level interception with minimum start and stop constraints at the business level, the stability of high-priority tasks is guaranteed, and execution oscillations caused by frequent switching of scheduling algorithms are effectively avoided. Attached Figure Description

[0058] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0059] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0060] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper surface," "lower surface," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "forward," "reverse," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0061] The core of this invention lies in achieving secure coupling between computing load and power grid commands through a three-stage closed loop of "cost calculation -> financial pricing -> physical defense".

[0062] like Figure 1 As shown, a computing power-electricity collaborative scheduling method based on a dynamic barrier option model includes the following steps:

[0063] Step S1. Initial Screening of Computing Power Tasks and Tiered Product Pricing: Receive computing power task requests submitted by users, extract the spatiotemporal feature parameters of the tasks, including maximum latency tolerance, deadline, data volume, and estimated remaining execution time; calculate the consistency repair cost when the task is suspended based on the spatiotemporal feature parameters, the consistency repair cost consisting of the physical cost of memory dumping and the computing power overhead of state rollback; introduce an exponential penalty model based on the overdue ratio to calculate the expected default penalty; extract the interruptible surplus time of the tasks, construct an envelope surface function, and based on the comparison result of the envelope surface function and the dynamic adaptive system security threshold, identify and encapsulate the tasks into either Class A rigid operation products or Class B interruptible products, and generate tiered pricing discounts for Class B interruptible products.

[0064] In this embodiment, the data center receives computation requests submitted by users. Taking a "distributed training of a certain model" task as an example, the system extracts its feature parameters: deadline. The amount of task state data in memory per hour Estimated remaining execution time Hour.

[0065] In step S1, the formula for calculating the consistency repair cost is as follows:

[0066] ,

[0067] in, For the cost of consistency restoration, The physical cost of memory dumping The computational cost for state rollback;

[0068] The calculations for the physical cost of memory dumping and the computational cost of state rollback are as follows:

[0069] ,

[0070] In the formula, This refers to the amount of memory-resident data when a task is suspended. Write the energy consumption constant per unit of data volume. To preset the average nodal marginal electricity price within the real-time scheduling period, This is the physical impact penalty coefficient. This refers to the cost of a single hardware depreciation. To recalculate time for breakpoint recovery, Rated power for computing nodes.

[0071] Set the energy consumption constant for writing per unit of data Let the average electricity price within the current peak-shaving window be... Yuan / kWh (This value is derived from a weighted average of real-time quotes over the past 30 minutes, rather than using the instantaneous peak quote of 1.2 yuan / kWh). Assuming a recovery recalculation time... Hourly rated power of computing nodes Substituting the data, the calculation yields... Yuan.

[0072] In step S1, the formula for calculating the expected penalty is:

[0073] ,

[0074] In the formula, For the user's basic contract amount, The actual delivery time of the task. For the deadline, This is the penalty sensitivity coefficient.

[0075] Set the base contract amount Yuan, Punishment Sensitivity If the actual delivery time is expected... Hours (overdue by 2 hours), substituting into the formula to calculate... Yuan.

[0076] Extract interruptible surplus time for the task The method for constructing the envelope surface function is as follows: extract the interruptible slack time of the task. Construct the envelope surface function :

[0077]

[0078] In the formula, These are the normalized weighting coefficients. It is a very small constant, used to prevent issues arising as the task approaches its deadline. When the value approaches or equals 0, a mathematical singularity with a denominator of zero appears in the calculation of the envelope surface function; a safety threshold is set for the dynamic adaptive system. ,in Based on historical average repair costs, For power grid peak shaving fluctuations, and This represents the risk preference coefficient.

[0079] In step S1, the method for generating tiered pricing discounts for Category B interruptible products includes:

[0080] Get the maximum latency tolerance Combining the probability density function of the duration of a single peak-shaving interruption in the power grid This leads to the probability that a peak-shaving interruption can be successfully absorbed without triggering a default. ;

[0081] According to the preset profit sharing ratio Generate price discounts for users ;

[0082] Finally, a user-perceived dynamic quote is generated. .

[0083] In the method for generating price discounts for users: setting The pre-defined arbitrage revenue sharing ratio between the data center and the user (set as follows) This is used to quantify the distribution rules of peak-shaving benefits between service providers and demanders; Generate and assign price discounts to users based on the system; The total subsidy the system expects to obtain from the power grid peak-shaving cycle, within the maximum latency tolerance authorized by the user, is then distributed according to a preset sharing ratio. Generate price discounts for users Ultimately, a user-perceived dynamic quote is generated. .in, This is for the final user-perceived dynamic quote (i.e., the final contract price that locks in the SLA). This is the standard billing benchmark price for the system's regular Class A rigid computing power service.

[0084] The judgment rule is as follows:

[0085] like It is packaged as a Class A rigid operation product;

[0086] like The system determines that the task falls outside the envelope and encapsulates it as a Class B interruptible product. The system then generates a price discount for the user. and output dynamic pricing based on user perception. .

[0087] Step S2. Establish and utilize a dynamic barrier option model to calculate the instantaneous expected net return difference: map the real-time peak-shaving subsidy price of the power grid to the underlying asset price of the option, and map the sum of the real-time consistency repair cost and the default penalty risk to the strike price; construct a dynamic barrier point that increases over time, which is generated based on the dynamic time penalty coefficient calculated by the surplus time ratio; construct an arbitrage trigger absolute threshold by superimposing a basic target net profit threshold; the underlying scheduler polls the real-time peak-shaving subsidy price of the power grid at high frequency, and the difference between the real-time peak-shaving subsidy price of the power grid and the arbitrage trigger absolute threshold is the instantaneous expected net return difference.

[0088] In step S2, the method for constructing the dynamic obstacle point and the absolute threshold for arbitrage triggering includes:

[0089] The real-time consistency repair cost established in step S1 Risk of default penalty The sum is mapped to the absolute physical defense baseline. ;

[0090] Calculate the current time Interruptible surplus time ,in To estimate the remaining execution time, Deadline;

[0091] Construct a dynamic time penalty coefficient based on the computation-save time ratio.

[0092] ,

[0093] in, The minimum constant is the penalty base. A preset constant greater than 0 serves as a scaling factor for the penalty intensity, used to adjust the overall magnitude of the penalty term; nonlinear compression index. The preset exponential sensitivity parameter is significant because it uses an exponential form to make the change in the computation-sufficiency time ratio produce a non-linear amplified penalty effect, causing the decision threshold to rise sharply in the later stages.

[0094] Generate dynamic obstacle points that float in real time. ;

[0095] Overlaying a preset basic target net profit threshold To ensure that data centers can obtain substantial economic benefits from participating in peak shaving, a minimum profit target is set for each arbitrage activity, thus establishing an absolute threshold for arbitrage triggering. ;

[0096] The underlying scheduler frequently polls the real-time peak-shaving subsidy quotes from the power grid. Calculate the instantaneous expected net income difference .

[0097] Initial stage of the mission ( (Hours): There's still a long way to go before the deadline. Larger Smaller, calculated The price is relatively low. At this time, if the real-time peak-shaving subsidy quote from the power grid is low... Exceeding the absolute threshold for arbitrage triggering The system determines the instantaneous expected net return difference. If it is determined that exercising the option is profitable, it will prepare to trigger an interruption.

[0098] Later stages of the mission ( (Hours): The task is about to expire. Approaching 0 leads to a barrier point. A surge. Even with grid subsidies. Maintaining a high level, due to The system determines that an interruption at this point would result in a serious breach of contract, thus protecting the continuous execution of the task.

[0099] Step S3. Smooth Delivery with Dual Finite State Machines: Dual finite state machines FSM-A and FSM-B are instantiated in parallel at the underlying computing node. For Class A rigid operation products, FSM-A performs interruption shielding calculations through a state lock-in mechanism to shield the network-side peak-shaving signal. For Class B interruptible products, FSM-B performs state transition control based on the instantaneous expected net revenue difference and time-series physical constraint parameters obtained in Step S2, and achieves smooth scheduling through business logic isolation.

[0100] In step S3, the FSM-A performs the interrupt masking calculation using the state lock-up mechanism, as follows:

[0101] ,

[0102] In the formula, This is the current scheduling time; This is the start time for the computing task; The actual (or estimated) completion time of the computing task; This means that all grid peak-shaving signals from the grid side are shielded at the kernel bus level, achieving absolute physical isolation and smooth delivery.

[0103] In step S3, the timing physical constraint parameters include: a. minimum running time constraint : Set the basic runtime constant Minutes; to ensure the mission can be completed between two shutdowns. A valid breakpoint retention period is used to dynamically calculate the minimum actual running time required:

[0104] ,

[0105] In the formula, The average interval between generating a complete snapshot of in-memory data. The fixed time consumed for process context recovery and warm-up;

[0106] b. Minimum downtime constraint Set the basic downtime constant. Minutes, using Newton's law of cooling to calculate the minimum downtime required for safe cooling of the chip:

[0107] ,

[0108] In the formula, The thermal resistance-capacitance time constant of the server cooling system. To trigger the CPU / memory core temperature at the moment of suspension, The low temperature threshold required to allow safe re-energization. This refers to the ambient temperature of the computer room.

[0109] In step S3, the state transition control includes:

[0110] Suspension transition logic: Assume the task's execution duration is... If and only if and At this time, the system enters a suspended state;

[0111] Restore transition logic: Assume the task's downtime is [time value]. If and only if and At that time, the system switches to a resumed running state.

[0112] To avoid frequent server startups and shutdowns caused by power grid signal quality or high-frequency price fluctuations, this step instantiates two parallel finite state machines (Dual-FSM) at the underlying layer of the scheduling system.

[0113] For rigid tasks of type A, FSM-A enters... After the state is reached, the interrupt masking calculation is performed. Shield all peak-shaving signals from the grid side to ensure physical isolation.

[0114] For the aforementioned Type B training task, FSM-B must simultaneously meet the economic criterion when receiving the peak-shaving signal. ) and timing constraint criteria ( Among them, the minimum running time constraint Ensure that breakpoints are saved upon task completion; minimum downtime constraint. Ensure the chip is safely cooled to the low temperature threshold the following.

[0115] A computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the aforementioned computational power-electricity coordinated scheduling method based on a dynamic obstacle option model. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0116] The working principle of this invention: This invention abandons the traditional concept of "hardware depreciation" and focuses on quantifying the "consistency repair cost" and "exponential default penalty" in distributed computing. This allows computing centers to accurately calculate the break-even point when participating in peak shaving, avoiding inverted losses. By introducing a dynamic barrier option model, complex network interactions are transformed into scientific financial decisions. The dynamically rising barrier point can accurately describe the risk premium generated by computing tasks over time, maximizing the arbitrage space of data centers. This application utilizes a dual finite state machine smooth delivery mode to achieve complete isolation of A / B class business logic. By replacing dangerous kernel-level interception with minimum start / stop constraints at the business level, it ensures the stability of high-priority tasks and effectively avoids execution oscillations caused by frequent switching of scheduling algorithms.

[0117] Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those embodiments or examples, without contradiction. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A computing power-electricity collaborative scheduling method based on a dynamic obstacle option model, characterized in that, Includes the following steps: Step S1. Initial Screening of Computing Power Tasks and Tiered Product Pricing: Receive computing power task requests submitted by users, extract the spatiotemporal feature parameters of the tasks, including maximum latency tolerance, deadline, data volume, and estimated remaining execution time; calculate the consistency repair cost when the task is suspended based on the spatiotemporal feature parameters, the consistency repair cost consists of the physical cost of memory dumping and the computing power overhead of state rollback; introduce an exponential penalty model based on the overdue ratio to calculate the expected penalty; extract the interruptible surplus time of the tasks, construct an envelope function, and based on the comparison result of the envelope function and the dynamic adaptive system security threshold, identify and encapsulate the tasks into either Class A rigid operation products or Class B interruptible products, and generate tiered pricing discounts for Class B interruptible products; Step S2. Establish and utilize a dynamic barrier option model to calculate the instantaneous expected net return difference: map the real-time peak-shaving subsidy price of the power grid to the underlying asset price of the option, and map the sum of the real-time consistency repair cost and the default penalty risk to the strike price; construct a dynamic barrier point that increases over time, which is generated based on the dynamic time penalty coefficient calculated by the surplus time ratio; construct an arbitrage trigger absolute threshold by superimposing a basic target net profit threshold; the underlying scheduler polls the real-time peak-shaving subsidy price of the power grid at high frequency, and the difference between the real-time peak-shaving subsidy price of the power grid and the arbitrage trigger absolute threshold is the instantaneous expected net return difference; Step S3. Smooth Delivery with Dual Finite State Machines: Dual finite state machines FSM-A and FSM-B are instantiated in parallel at the underlying computing node. For Class A rigid operation products, FSM-A performs interruption shielding calculations through a state lock-in mechanism to shield the network-side peak-shaving signal. For Class B interruptible products, FSM-B performs state transition control based on the instantaneous expected net revenue difference and time-series physical constraint parameters obtained in Step S2, and achieves smooth scheduling through business logic isolation.

2. The computing power-electricity coordinated scheduling method based on the dynamic obstacle option model according to claim 1, characterized in that, In step S1, the formula for calculating the consistency repair cost is as follows: ， in, For the cost of consistency restoration, The physical cost of memory dumping The computational cost for state rollback; The calculations for the physical cost of memory dumping and the computational cost of state rollback are as follows: ， In the formula, This refers to the amount of memory-resident data when a task is suspended. Write the energy consumption constant per unit of data volume. To preset the average nodal marginal electricity price within the real-time scheduling period, This is the physical impact penalty coefficient. This refers to the cost of a single hardware depreciation. To recalculate time for breakpoint recovery, Rated power for computing nodes.

3. The computing power-electricity coordinated scheduling method based on the dynamic obstacle option model according to claim 2, characterized in that, In step S1, the formula for calculating the expected penalty is: ， In the formula, For the user's basic contract amount, The actual delivery time of the task. For the deadline, This is the penalty sensitivity coefficient.

4. The computing power-electricity coordinated scheduling method based on the dynamic obstacle option model according to claim 3, characterized in that, In step S1, the method for constructing the envelope surface function is as follows: extract the interruptible slack time of the task. Construct the envelope surface function : ， In the formula, These are the normalized weighting coefficients. It is a very small constant, used to prevent issues arising as the task approaches its deadline. When the value approaches or equals 0, mathematical singularities with zero denominators appear in the calculation of the envelope surface function; Setting a safety threshold for a dynamic adaptive system ,in Based on historical average repair costs, For power grid peak shaving fluctuations, and This is the risk preference coefficient; The judgment rule is: if Packaged as a Class A rigid operation product; if It is packaged as a Class B interruptible product.

5. The computing power-electricity coordinated scheduling method based on the dynamic obstacle option model according to claim 4, characterized in that, In step S1, the method for generating tiered pricing discounts for Category B interruptible products includes: Get the maximum latency tolerance Combining the probability density function of the duration of a single peak-shaving interruption in the power grid This leads to the probability that a peak-shaving interruption can be successfully absorbed without triggering a default. ; According to the preset profit sharing ratio Generate price discounts for users ; The total subsidy that the system expects to obtain from the grid peak shaving cycle within the maximum latency tolerance authorized by the user; Finally, a user-perceived dynamic quote is generated. , This is the standard billing benchmark price for the system's regular Class A rigid computing power service.

6. The computing power-electricity coordinated scheduling method based on the dynamic barrier option model according to claim 5, characterized in that, In step S2, the method for constructing the dynamic obstacle point and the absolute threshold for arbitrage triggering includes: The real-time consistency repair cost established in step S1 Risk of default penalty The sum is mapped to the absolute physical defense baseline. ; Calculate the current time Interruptible rich time ,in To estimate the remaining execution time, Deadline; Construct a dynamic time penalty coefficient based on the computation-save time ratio. , in, The minimum constant is the penalty base. A preset constant greater than 0 serves as a scaling factor for the penalty intensity, used to adjust the overall magnitude of the penalty term; nonlinear compression index. The preset exponential sensitivity parameter is significant because it uses an exponential form to make the change in the computation-sufficiency time ratio produce a non-linear amplified penalty effect, causing the decision threshold to rise sharply in the later stages. Generate dynamic obstacle points that float in real time. ; Overlaying a preset basic target net profit threshold To ensure that data centers can obtain substantial economic benefits from participating in peak shaving, a minimum profit target is set for each arbitrage activity, thus establishing an absolute threshold for arbitrage triggering. ; The underlying scheduler frequently polls the real-time peak-shaving subsidy quotes from the power grid. Calculate the instantaneous expected net income difference .

7. The computing power-electricity collaborative scheduling method based on a dynamic obstacle option model according to claim 6, characterized in that, In step S3, the FSM-A performs the interrupt masking calculation using the state lock-up mechanism, as follows: ， In the formula, This is the current scheduling time; This is the start time for the computing task; The actual (or estimated) completion time of the computing task; This means that all grid peak-shaving signals from the grid side are shielded at the kernel bus level, achieving absolute physical isolation and smooth delivery.

8. The computing power-electricity collaborative scheduling method based on a dynamic obstacle option model according to claim 7, characterized in that, In step S3, the timing physical constraint parameters include: a. Minimum running time constraint : Set the basic runtime constant Minutes; to ensure the mission can be completed between two shutdowns. A valid breakpoint retention period is used to dynamically calculate the minimum actual running time required: ， In the formula, The average interval between generating a complete snapshot of in-memory data. The fixed time consumed for process context recovery and warm-up; b. Minimum downtime constraint Set the basic downtime constant. Minutes, using Newton's law of cooling to calculate the minimum downtime required for safe cooling of the chip: ， In the formula, The thermal resistance-capacitance time constant of the server cooling system. To trigger the CPU / memory core temperature at the moment of suspension, The low temperature threshold required to allow safe re-energization. This refers to the ambient temperature of the computer room.

9. The computing power-electricity collaborative scheduling method based on a dynamic obstacle option model according to claim 8, characterized in that, In step S3, the state transition control includes: suspending transition logic: assuming the task's execution duration is... If and only if and At this time, the system enters a suspended state; Restore transition logic: Assume the task's downtime is [time value]. If and only if and At that time, the system switches to a resumed running state.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the computing power-electricity collaborative scheduling method based on the dynamic obstacle option model as described in any one of claims 1 to 9.

Images