Electronics co-optimization method and system based on digital twinning, electronic device and storage medium

Abstract

The application provides an algorithm and electricity collaborative optimization method and system based on digital twinning, electronic equipment and storage medium, belongs to the technical field of power resource scheduling optimization, including obtaining historical and real-time power system parameter information and calculation task information, and constructing a dynamic recursive state prediction model based on digital twinning; using the prediction model, the power demand and calculation task load at the next time are predicted, and according to the prediction result, the current power demand and calculation task load are preliminarily scheduled; on the basis of preliminary scheduling, the scheduling of the current power demand and calculation task load is collaboratively optimized according to the Lagrange optimization algorithm, so as to generate a scheduling result that minimizes the scheduling deviation of the power demand and the calculation task load, and the scheduling result is dynamically adjusted through a stability correction factor. The application realizes accurate collaborative optimization between power demand and calculation task load.

Description

Technical Field

[0001] This invention relates to the field of power resource dispatch optimization technology, and in particular to a computational power collaborative optimization method, system, electronic device and storage medium based on digital twins. Background Technology

[0002] Computing power is a new type of productive force in the digital economy era and a key engine for building new power systems and energy systems. Data centers are the carriers of computing power, forming powerful computing capabilities and large-scale data processing capabilities by equipping them with various servers, graphics cards, and other hardware devices to meet the application scenarios of high-performance computing. With the continuous increase in global energy demand and the increasing complexity of computing tasks, the coordinated scheduling of power systems and data centers faces unprecedented challenges.

[0003] Traditional power dispatching methods often rely on historical data or static models, lacking the ability to adapt to dynamic changes and real-time loads. Meanwhile, data center computing loads are constantly increasing and fluctuating, making the rational allocation of computing resources particularly important. To improve system energy efficiency and ensure rational resource allocation, achieving precise co-optimization between power demand and computing load has become a key issue in modern power and computing task management. Summary of the Invention

[0004] To address the above problems, this invention provides a digital twin-based computing-power collaborative optimization method, system, electronic device, and storage medium to achieve resource scheduling optimization between power systems and data centers.

[0005] This invention provides a computing-electricity co-optimization method based on digital twins, comprising:

[0006] Acquire historical power system parameter information and historical calculation task information within the first historical period, historical power system parameter information and historical calculation task information within the second historical period, as well as real-time power system parameter information and real-time calculation task information;

[0007] By utilizing historical power system parameter information and historical calculation task information within the first historical period, a dynamic recursive state prediction model based on digital twins is constructed.

[0008] Using the constructed dynamic recursive state prediction model, the power demand and computing task load of the next moment are predicted based on the historical power system parameter information and historical computing task information in the second historical period. Based on the prediction results, the power demand and computing task load of the current moment are initially scheduled.

[0009] Based on the initial scheduling, the scheduling of power demand and computing task load at the current moment is coordinated and optimized according to the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load. The scheduling result is then dynamically adjusted by a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0010] Furthermore, the power system parameter information includes power load, current, voltage, power, and temperature, and the calculation task information includes the execution progress of the calculation task, load allocation, response time, and energy consumption.

[0011] Furthermore, the construction of a dynamic recursive state prediction model based on digital twins using historical power system parameter information and historical computational task information within the first historical period includes constructing a dynamic recursive state prediction model for power demand and a dynamic recursive state prediction model for computational task load. The constructed dynamic recursive state prediction model for power demand is as follows:

[0012]

[0013] In the formula, It is a forecast of electricity demand; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The electricity demand; It is a historic moment The weighting coefficients of the electricity demand data represent the degree of influence of past time points on electricity demand forecasts; It is a historic moment The dynamic adjustment coefficient of the power demand data is adjusted in real time as the load fluctuation rate changes;

[0014] The constructed dynamic recursive state prediction model for the computational task load is as follows:

[0015]

[0016] In the formula, It is a predicted value for calculating task load; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The computational workload; It is a historic moment The weighting coefficients of the computational task load data represent the degree of influence of past time points on the prediction of computational task load. It is a historic moment The dynamic adjustment coefficient of the computational task load data is used to adapt to the rate of fluctuation of the computational task load.

[0017] Furthermore, the preliminary scheduling of current power demand and computing task load based on the prediction results includes performing preliminary scheduling of current power demand and computing task load using the following formula:

[0018]

[0019]

[0020] In the formula, and At any moment Initial allocation of power demand and computational workload; and These are the predicted values ​​for electricity demand and the predicted values ​​for the computational task load; It is a real-time load adaptation factor, used to dynamically adjust power demand dispatch according to changes in real-time load in order to cope with load fluctuations; and They are time points Maximum power demand limits and maximum computing task load limits are set to ensure the system is not overloaded.

[0021] Furthermore, based on the initial scheduling, the process of collaboratively optimizing the scheduling of current power demand and computing task load using the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load includes determining the objective function of the scheduling error between power demand and computing task load, and obtaining the allocation strategy by minimizing the objective function, wherein the objective function is:

[0022]

[0023] In the formula, It is the objective function for scheduling optimization, used to measure electricity demand. and computational task load The scheduling error is identified and penalized, with the goal of minimizing the scheduling error. It represents the total length of the scheduling period, indicating the number of moments in the entire scheduling cycle; and It is the Lagrange multiplier, used to control the penalty for power demand scheduling errors and calculation task load scheduling errors.

[0024] Furthermore, the dynamic adjustment of the scheduling results based on the stability correction factor according to real-time power system parameter information and real-time calculation task information includes determining the stability correction factor, wherein the formula for the stability correction factor is:

[0025]

[0026] In the formula, It is a stability correction factor, representing the value at time t. The scheduling stability adjustment factor; It is a stability correction coefficient, used to control the degree of influence of errors on the stability correction factor. These are weighting coefficients used to adjust the strength of the impact of the difference between the current actual load and the future predicted load on stability correction. and The time is obtained based on real-time power system parameter information and real-time calculation task information. The power demand and computing workload.

[0027] Furthermore, the dynamic adjustment of the scheduling results by using a stability correction factor based on real-time power system parameter information and real-time computing task information includes determining the final power demand and computing task load scheduling results using the following formula:

[0028]

[0029]

[0030] In the formula, and It minimizes the scheduling optimization objective function, resulting in the optimized power demand scheduling quantity and computational task load scheduling quantity. It is a stability correction factor.

[0031] Furthermore, the scheduling of power demand and computing task load at the current moment is coordinated and optimized according to the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load. This includes constraining the optimization process with constraints, which include power demand constraints, computing task timeliness constraints, and system stability constraints.

[0032] Furthermore, the power demand constraint, the computational task timeliness constraint, and the system stability constraint are as follows:

[0033] Electricity demand constraints:

[0034]

[0035] Time constraints for computation tasks:

[0036]

[0037] System stability constraints:

[0038]

[0039]

[0040] In the formula, It is a moment electricity demand, For a moment The computational workload; It is a moment Maximum electricity demand limit; It is a moment The execution time of the corresponding computational task; It is the maximum timeliness requirement for computing tasks, that is, the maximum tolerable execution time for each task; It is the tolerance for power load error, representing the time interval. electricity demand Compared with the predicted value The maximum error between; It calculates the tolerance for prediction errors in the task, representing the time interval. Computational workload Compared with the predicted value The maximum error between them.

[0041] The present invention provides a computing and power collaborative optimization system based on digital twins, including a data acquisition module, a prediction model construction module, a preliminary scheduling module, a resource scheduling optimization module, and a scheduling scheme output module;

[0042] The data acquisition module is used to acquire historical power system parameter information and historical calculation task information in the first historical period, historical power system parameter information and historical calculation task information in the second historical period, as well as real-time power system parameter information and real-time calculation task information.

[0043] The prediction model construction module is used to construct a dynamic recursive state prediction model based on digital twins by utilizing the historical power system parameter information and historical calculation task information within the first historical period.

[0044] The preliminary scheduling module is used to use the constructed dynamic recursive state prediction model to predict the power demand and computing task load of the next moment based on the historical power system parameter information and historical computing task information in the second historical period, and to perform preliminary scheduling of the power demand and computing task load of the current moment based on the prediction results.

[0045] The resource scheduling optimization module is used to perform coordinated optimization of the scheduling of power demand and computing task load at the current moment based on the initial scheduling and the Lagrange optimization algorithm, so as to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load, and dynamically adjust the scheduling result through a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0046] The scheduling scheme output module is used to output the adjusted scheduling result.

[0047] Furthermore, the data acquisition module includes a first historical data acquisition unit, a second historical data acquisition unit, and a real-time data acquisition unit, wherein,

[0048] The first historical data acquisition unit is used to acquire historical power system parameter information and historical calculation task information within the first historical time period;

[0049] The second historical data acquisition unit is used to acquire historical power system parameter information and historical calculation task information within the second historical period.

[0050] The real-time data acquisition unit is used to acquire real-time power system parameter information and real-time calculation task information.

[0051] Furthermore, the preliminary scheduling module includes a prediction unit and a scheduling unit, wherein,

[0052] The prediction unit is used to use the constructed dynamic recursive state prediction model to predict the power demand and computing task load of the next moment based on the historical power system parameter information and historical computing task information in the second historical period, and to transmit the prediction results to the scheduling unit.

[0053] The scheduling unit is used to perform preliminary scheduling of the current power demand and computing task load based on the prediction results of the prediction unit.

[0054] Furthermore, the resource scheduling optimization module includes an optimization unit and a constraint unit, wherein,

[0055] The optimization unit is used to perform coordinated optimization of the scheduling of power demand and computing task load at the current moment based on the initial scheduling and according to the Lagrange optimization algorithm, so as to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load, and dynamically adjust the scheduling result by means of a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0056] The constraint unit is used to constrain the scheduling result using constraint conditions.

[0057] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the following method:

[0058] Acquire historical power system parameter information and historical calculation task information within the first historical period, historical power system parameter information and historical calculation task information within the second historical period, as well as real-time power system parameter information and real-time calculation task information;

[0059] By utilizing historical power system parameter information and historical calculation task information within the first historical period, a dynamic recursive state prediction model based on digital twins is constructed.

[0060] Using the constructed dynamic recursive state prediction model, the power demand and computing task load of the next moment are predicted based on the historical power system parameter information and historical computing task information in the second historical period. Based on the prediction results, the power demand and computing task load of the current moment are initially scheduled.

[0061] Based on the initial scheduling, the scheduling of power demand and computing task load at the current moment is coordinated and optimized according to the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load. The scheduling result is then dynamically adjusted by a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0062] The present invention provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the following method:

[0063] Acquire historical power system parameter information and historical calculation task information within the first historical period, historical power system parameter information and historical calculation task information within the second historical period, as well as real-time power system parameter information and real-time calculation task information;

[0064] By utilizing historical power system parameter information and historical calculation task information within the first historical period, a dynamic recursive state prediction model based on digital twins is constructed.

[0065] Using the constructed dynamic recursive state prediction model, the power demand and computing task load of the next moment are predicted based on the historical power system parameter information and historical computing task information in the second historical period. Based on the prediction results, the power demand and computing task load of the current moment are initially scheduled.

[0066] Based on the initial scheduling, the scheduling of power demand and computing task load at the current moment is coordinated and optimized according to the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load. The scheduling result is then dynamically adjusted by a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0067] This invention provides a digital twin-based computing-power collaborative optimization method, system, electronic device, and storage medium. By accurately predicting power demand and computing load, and intelligently scheduling based on the prediction results, it can more efficiently allocate current power demand and computing load, avoiding energy waste and overuse of computing resources, thereby significantly improving system energy efficiency. By introducing a stability correction factor, the scheduling scheme can be dynamically adjusted to cope with fluctuations in power load and computing load. Real-time monitoring and feedback mechanisms enable the system to maintain stable operation when the load changes, ensuring the effective allocation of current power demand and computing load. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the computing and power collaborative optimization method based on digital twins according to Embodiment 1 of the present invention.

[0069] Figure 2 This is a schematic diagram of the structure of the computing and power collaborative optimization system based on digital twins according to Embodiment 2 of the present invention.

[0070] Figure labeling: 10, Data acquisition module; 20, Predictive model construction module; 30, Preliminary scheduling module; 40, Resource scheduling optimization module; 50, Scheduling scheme output module. Detailed Implementation

[0071] The following describes specific embodiments and appendices. Figures 1-2 The invention is described in detail so that those skilled in the art can more fully understand its purpose, features and effects.

[0072] Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. In the event of any discrepancy between the definitions of terms in this invention and their commonly understood meaning by one of ordinary skill in the art to which this invention pertains, the definitions provided in this invention shall prevail.

[0073] Existing power dispatch relies on historical data or static models, often ignoring the dynamic characteristics of load fluctuations. This results in the system being unable to adjust dispatch strategies in a timely manner when power load fluctuates significantly, leading to over-dispatch or under-dispatch.

[0074] Furthermore, the widespread application of renewable energy, especially the integration of fluctuating energy sources such as wind and solar power, has brought more uncertainty and challenges to power system dispatch. Traditional dispatch methods lack flexibility and cannot effectively integrate the fluctuating characteristics of renewable energy with the real-time dispatch needs of green energy, leading to uneven resource utilization and affecting the stability and sustainability of the system.

[0075] The rise of smart microgrids and green data centers demands that power systems more efficiently coordinate with computing task load scheduling to achieve optimal system operation. This invention provides a digital twin-based computing-power co-optimization method, system, electronic device, and storage medium. Based on digital twin technology, and through real-time data monitoring and prediction models, it enhances the intelligence level of power demand and computing task load scheduling, thereby achieving more efficient and stable system operation.

[0076] Example 1

[0077] As a specific embodiment of the present invention, this embodiment provides a computing-power collaborative optimization method based on digital twins, referring to... Figure 1 The specific steps are as follows:

[0078] S100: Obtain historical power system parameter information and historical calculation task information within the first historical period, historical power system parameter information and historical calculation task information within the second historical period, as well as real-time power system parameter information and real-time calculation task information.

[0079] S200. Using historical power system parameter information and historical calculation task information within the first historical period, a dynamic recursive state prediction model based on digital twins is constructed.

[0080] S300. Using the constructed dynamic recursive state prediction model, the power demand and computing task load of the next moment are predicted based on the historical power system parameter information and historical computing task information in the second historical period. Based on the prediction results, the power demand and computing task load of the current moment are initially scheduled.

[0081] S400. Based on the initial scheduling, the scheduling of power demand and computing task load at the current moment is coordinated and optimized according to the Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load. The scheduling result is dynamically adjusted by a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0082] The computing-power co-optimization scheme based on digital twins in this embodiment can more efficiently allocate the current power demand and computing task load by accurately predicting power demand and computing task load and performing intelligent scheduling based on the prediction results. This avoids energy waste and overuse of computing resources, thereby significantly improving the system's energy efficiency. By introducing a stability correction factor, the scheduling scheme can be dynamically adjusted to cope with fluctuations in power load and computing task load. The real-time monitoring and feedback mechanism enables the system to maintain stable operation when the load changes, ensuring the effective allocation of the current power demand and computing task load.

[0083] Furthermore, in S100, the power system parameter information includes data such as power load, current, voltage, power, and temperature; the calculation task information includes data such as the execution progress of the calculation task, load allocation, response time, and energy consumption.

[0084] By deploying various sensors in power systems and data centers, real-time information on power system parameters such as power load, current, voltage, power, and temperature, as well as computational task information such as execution progress, load allocation, response time, and energy consumption, is acquired. In one feasible embodiment, sensors deployed in the power system primarily collect operational data from the power grid, recording power load fluctuations, changes in current and voltage, and changes in renewable energy generation for each time period; sensors deployed in the data center primarily collect information on computing resource usage, recording the execution progress, load allocation, response time, and energy consumption of computational tasks.

[0085] In one feasible embodiment, the power system parameter information can be collected through smart meters, current transformers, voltage sensors, frequency collectors, temperature sensors, and energy storage system monitoring modules, including data such as current, voltage, active power, reactive power, frequency, equipment temperature, energy storage capacity, and load type identification; the computing task information is collected through server management chips (such as BMC), operating system monitoring interfaces (such as SNMP), high-performance counters, and task scheduling monitoring modules, including data such as task arrival rate, task priority, computing node CPU / GPU utilization, memory occupancy, node temperature, execution latency, resource waiting time, and task completion rate.

[0086] Historical power system parameter information and historical calculation task information record past operating data of the power system and data center, which can be obtained from databases, log systems, and equipment collection records.

[0087] The construction of a dynamic recursive state prediction model based on digital twins, utilizing historical power system parameter information and historical computing task information within the first historical period, includes constructing a dynamic recursive state prediction model for power demand and a dynamic recursive state prediction model for computing task load.

[0088] In this embodiment, the digital twin model uses the power system and data center in the physical world as mapping objects to construct a high-fidelity dynamic equivalent model. It continuously and synchronously acquires the current operating status through multiple types of sensors deployed on-site. The physical quantity data collected by these sensors, such as power load, current, voltage, power, temperature, execution progress, load allocation, response time, and energy consumption, are injected into the virtual model in real time, driving the digital twin model to maintain consistency with the actual system in the time dimension.

[0089] In this embodiment, power demand refers to electrical load, which reflects the power required by the system at a certain moment; computing task load refers to the comprehensive reflection of the number, complexity, and resource consumption intensity of computing tasks that the data center needs to process within a certain period of time, including the number of concurrent tasks, task complexity, resource usage, and task execution time.

[0090] This embodiment employs a digital twin-assisted dynamic recursive state prediction modeling mechanism. Based on the original historical data, the digital twin model normalizes, compresses, and performs nonlinear transformations on each variable to obtain a set of state variable sequences that can be used for modeling. Weighted influence functions and disturbance enhancement functions are then constructed on these variables to build a prediction expression with state recursive memory characteristics.

[0091] The constructed dynamic recursive state prediction model for the aforementioned electricity demand is as follows:

[0092]

[0093] In the formula, It is a forecast of electricity demand; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The electricity demand reflects the electricity load at past moments; It is a historic moment The weighting coefficients of the electricity demand data represent the degree of influence of past time on electricity demand forecasts, and are obtained by fitting historical data or by machine learning algorithms; It is a historic moment The dynamic adjustment coefficient of the electricity demand data is adjusted in real time as the load fluctuation rate changes. By monitoring the fluctuation of electricity load and the changes in renewable energy generation, the coefficient is dynamically adjusted so that the model can respond to load fluctuations at different rates of change.

[0094] In this embodiment, the modeling of the power load is not a simple superposition of historical load values, but rather utilizes a digital twin model to represent the state at each moment. After encoding and weighting, a dynamic perturbation term reflecting the changing trend is added. .

[0095] By incorporating changes in renewable energy generation, electricity resources can be allocated more rationally, green energy can be fully utilized, dependence on traditional energy sources can be reduced, and the development of green electricity and smart microgrids can be promoted.

[0096] The constructed dynamic recursive state prediction model for the computational task load is as follows:

[0097]

[0098] In the formula, It is a predicted value for calculating task load; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The computational workload is obtained based on actual system operation records; It is a historic moment The weighting coefficients of the computational task load data represent the degree of influence of past time points on the prediction of computational task load. It is a historic moment The dynamic adjustment coefficient of the computational task load data is used to adapt to the rate of fluctuation of the computational task load.

[0099] In this embodiment, the computational task load is modeled, and a state sequence is constructed. And introduce a dynamic nonlinear adjustment coefficient. This is to simulate non-stationary factors such as sudden tasks and resource congestion in real systems.

[0100] By combining historical computing task load and fluctuation data, the weighting coefficients and dynamic adjustment coefficients in the dynamic recursive state prediction model of the computing task load are adjusted.

[0101] All inputs and Instead of directly sampling values, the results are multi-parameter combined inferences output by the digital twin model—that is, equivalent "system state variables" formed by combining factors such as power fluctuation trends, task migration patterns, and resource temperature coupling characteristics based on real-time simulation of the real system. These state variables, after being embedded, transformed, and modeled with dynamic response, become the core driver of the recursive model.

[0102] The model also provides a dynamic calibration mechanism. Whenever the prediction error deviates from the system's set stability threshold, the digital twin automatically reverts to historical simulation states and readjusts the weighting coefficients in the prediction parameters. , With dynamic adjustment coefficient , This ensures that the model's response capability adapts to changes over time. This is achieved through multi-period residual detection and sliding window regression iteration, a process that does not rely on neural networks but is based on state residual field modeling and parameter perturbation response analysis.

[0103] Although the dynamic recursive state prediction model is trained on historical data, it needs to continuously introduce real-time data streams during operation to compare the error between the predicted and actual values ​​and dynamically adjust the prediction coefficients. Timely capture of non-periodic events (such as sudden computing tasks or local power grid jumps) is necessary to enable model adaptation; otherwise, the model will always be based on the past and will be unable to cope with sudden changes in the system.

[0104] The dynamic recursive state prediction model driven by digital twins combines physical consistency, temporal correlation and adaptive adjustment capabilities. It can achieve high-precision prediction of future power demand and computing task load, and has strong adaptability and good stability. It is particularly suitable for collaborative optimization scheduling scenarios with strong time variation and complex coupling behavior in green data centers and smart microgrids.

[0105] In this embodiment, a dynamic recursive state prediction model is constructed using digital twins to jointly predict power demand and computing load. To better predict future power demand and computing load, a dynamic recursive state prediction model is adopted, taking into account the impact of historical data on the current moment and adapting to the volatility of power and computing load through dynamic adjustment coefficients. Power demand prediction is achieved by weighting power demand data over a period of time, while introducing a dynamic adjustment coefficient to allow the prediction to be adjusted accordingly based on the rate of change in past load.

[0106] Furthermore, in S200, the preliminary scheduling of the current power demand and computing task load based on the prediction results includes performing preliminary scheduling of the current power demand and computing task load using the following formula:

[0107]

[0108]

[0109] In the formula, and At the current moment Initial allocation of power demand and computational workload; and These are the predicted values ​​for electricity demand and the predicted values ​​for the computational task load;

[0110] It is a real-time load adaptation factor used to dynamically adjust power demand dispatch according to changes in real-time load in order to cope with load fluctuations. It is obtained by normalizing the growth rate of the current load compared to the previous moment, and combining the degree of load deviation from the average level (i.e., standard deviation) over a period of time. It is a dynamically calculated adjustment coefficient after being weighted by preset weights and used to flexibly adjust the dispatch amplification ratio of the predicted load.

[0111] and These are the current times. Maximum power demand limits and maximum computing task load limits are set to ensure the system does not overload. and It can be calculated based on historical data.

[0112] After obtaining the predicted results of future power demand and computing task load from the dynamic recursive state prediction models of power demand and computing task load constructed according to S100, respectively, resource scheduling optimization is performed. First, preliminary scheduling is conducted. The goal of preliminary scheduling is to allocate preliminary power demand and computing task load for each time period based on the prediction results, ensuring that the power demand and computing task load at each time period are met to a certain extent in the preliminary scheduling. To ensure the rationality of the preliminary scheduling, a real-time load adaptation factor is added to the preliminary scheduling formula.

[0113] The dynamic adjustment coefficient and real-time load adaptation factor can be flexibly adjusted according to the fluctuations in power load and computing task load, ensuring that the system can respond quickly to load changes and avoid over-scheduling or under-scheduling.

[0114] Furthermore, in S300, based on the initial scheduling, to further optimize the allocation of power demand and computing load at the current moment, it is necessary to consider minimizing scheduling error and the synergistic effect between resources. A Lagrange optimization algorithm is used to synergistically optimize the scheduling of power demand and computing load. The optimization objective is to minimize the deviation between power demand and computing load scheduling, and the stability of the system is guaranteed by constraints. By introducing the Lagrange multiplier method, the scheduling error of power demand and computing load in the optimization problem is quantified as a penalty term, and the optimal resource allocation strategy is obtained by minimizing the objective function.

[0115] Specifically, the objective function for the scheduling error between the power demand and the computational task load is:

[0116]

[0117] In the formula, It is the objective function for scheduling optimization, used to measure electricity demand. and computational task load The goal is to minimize scheduling errors and penalize them, thereby achieving more efficient resource scheduling. It represents the total length of the scheduling period, indicating the number of moments in the entire scheduling cycle; and It is the Lagrange multiplier, used to control the penalty for power demand scheduling errors and calculation task load scheduling errors.

[0118] The objective function calculates the scheduling error between power demand and computing task load. The scheduling deviation at each time step is squared and weighted with the corresponding Lagrange coefficient. The goal is to balance the penalty of scheduling error by adjusting the coefficients so that the resource allocation of power and computing tasks is as close as possible to the initial scheduling value, thereby achieving optimal resource scheduling.

[0119] The optimized power demand scheduling quantity is obtained by minimizing the scheduling optimization objective function. and computational task load scheduling .

[0120] Preferably, after obtaining the optimized scheduling result and Then, the scheduling results are adjusted using a stability correction factor to ensure stable system operation and adaptability to load fluctuations. The stability correction factor is dynamically adjusted using the following formula:

[0121]

[0122] In the formula, It is a stability correction factor, representing the value at time t. The scheduling stability adjustment factor; It is a stability correction coefficient, used to control the degree of influence of errors on the stability correction factor. It is a weighting coefficient used to adjust the strength of the impact of the difference between the current actual load and the future predicted load on stability correction. and The time is obtained based on real-time power system parameter information and real-time calculation task information. The power demand and computing workload.

[0123] The dynamic adjustment of scheduling results through stability correction factors includes determining the final power demand and computational task load scheduling results using the following formula:

[0124]

[0125]

[0126] To ensure system stability, constraints are imposed on the optimization process. In this embodiment, the constraints include power demand constraints, computational task timeliness constraints, and system stability constraints.

[0127] Specifically, the constraints are as follows:

[0128] Electricity demand constraints:

[0129]

[0130] Time constraints for computation tasks:

[0131]

[0132] System stability constraints:

[0133]

[0134]

[0135] In the formula, It is a moment The electricity demand; It is a moment Maximum electricity demand limit; It is a moment The execution time of the corresponding computational task; It is the maximum timeliness requirement for computing tasks, that is, the maximum tolerable execution time for each task; It is the tolerance for power load error, representing the time interval. electricity demand Compared with the predicted value The maximum error between; It calculates the tolerance for prediction errors in the task, representing the time interval. Computational workload Compared with the predicted value The maximum error between them. This formula reflects the current time. Scheduling and the next moment Stability constraints between predictions, at the current time The power demand and computing load of the scheduling should not deviate too far from the future forecast to ensure that the scheduling has forecast consistency and forward stability of the system.

[0136] The constraints ensure that power dispatch does not exceed the system's maximum supply capacity. Furthermore, the computational tasks can be completed on time, ensuring system stability. By combining constraints with the objective function, a holistic optimization problem is formed, aiming to minimize the error while satisfying the aforementioned constraints. Ultimately, the optimal scheduling result that meets all requirements is obtained. and .

[0137] By using the Lagrange optimization algorithm, which comprehensively considers the scheduling errors of power demand and computing task load, the allocation scheme of power resources and computing task load can be accurately optimized, reducing resource waste and making the scheduling result closer to the optimal state. It not only optimizes the scheduling scheme, but also ensures that constraints such as power demand and computing task timeliness are met, avoiding system instability caused by overload or task delay, and improving the reliability and sustainability of the system.

[0138] Furthermore, based on the final power demand at the current moment calculated by S300 and the results of the computational task load scheduling... and The optimized scheduling scheme is output, which ensures the optimal allocation of power demand and computing task load, and meets the system stability and load constraints.

[0139] In this embodiment, the final power demand and computing task load scheduling results are obtained. and It is the current moment. The scheduling output, although using the next time step The purpose of the prediction is to serve as an adjustment reference so that the current moment... The scheduling output is more robust and forward-looking.

[0140] The computing-power co-optimization method based on digital twins provided in this implementation, on the basis of preliminary scheduling, uses the Lagrange optimization algorithm to co-optimize the scheduling of power demand and computing task load, minimizes scheduling error and balances the synergistic effect between resources, and dynamically adjusts the scheduling results by introducing a stability correction factor, taking into account the constraints of power demand, computing task timeliness and system stability, and finally obtains the optimal scheduling scheme that satisfies all constraints.

[0141] The computing-electricity co-optimization method based on digital twins provided in this embodiment solves the problems of existing technologies that usually ignore the dynamic characteristics of load fluctuations, resulting in the system being unable to adjust the scheduling strategy in a timely manner when the power load fluctuates significantly, thus causing over-schedule or under-schedule; and the lack of flexibility of traditional scheduling methods, which cannot effectively integrate the fluctuation characteristics of renewable energy and the real-time scheduling needs of green energy, resulting in unbalanced resource utilization and affecting the stability and sustainability of the system.

[0142] Example 2

[0143] As another specific embodiment of the present invention, this embodiment provides a computing-power collaborative optimization system based on digital twins, referring to... Figure 2 It includes a data acquisition module 10, a prediction model construction module 20, a preliminary scheduling module 30, a resource scheduling optimization module 40, and a scheduling scheme output module 50.

[0144] The data acquisition module 10 is used to acquire historical power system parameter information and historical calculation task information in the first historical period, historical power system parameter information and historical calculation task information in the second historical period, and real-time power system parameter information and real-time calculation task information.

[0145] The prediction model construction module 20 is used to construct a dynamic recursive state prediction model based on digital twin by utilizing the historical power system parameter information and historical calculation task information within the first historical period.

[0146] The preliminary scheduling module 30 is used to use the constructed dynamic recursive state prediction model to predict the power demand and computing task load of the next moment based on the historical power system parameter information and historical computing task information in the second historical period, and to perform preliminary scheduling of the power demand and computing task load of the current moment according to the prediction results.

[0147] The resource scheduling optimization module 40 is used to perform coordinated optimization of the scheduling of power demand and computing task load at the current moment based on the preliminary scheduling and the Lagrange optimization algorithm, so as to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load, and dynamically adjust the scheduling result by means of a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0148] The scheduling scheme output module 50 is used to output the adjusted scheduling result.

[0149] Furthermore, the data acquisition module 10 includes a first historical data acquisition unit, a second historical data acquisition unit, and a real-time data acquisition unit, wherein,

[0150] The first historical data acquisition unit is used to acquire historical power system parameter information and historical calculation task information within the first historical time period;

[0151] The second historical data acquisition unit is used to acquire historical power system parameter information and historical calculation task information within the second historical period.

[0152] The real-time data acquisition unit is used to acquire real-time power system parameter information and real-time calculation task information.

[0153] Furthermore, the preliminary scheduling module 30 includes a prediction unit and a scheduling unit, wherein,

[0154] The prediction unit is used to use the constructed dynamic recursive state prediction model to predict the power demand and computing task load of the next moment based on the historical power system parameter information and historical computing task information in the second historical period, and to transmit the prediction results to the scheduling unit.

[0155] The scheduling unit is used to perform preliminary scheduling of the current power demand and computing task load based on the prediction results of the prediction unit.

[0156] The goal of preliminary scheduling is to allocate preliminary power demand and computing load for each time period based on the forecast results, and to perform preliminary resource allocation to ensure that the power demand and computing load for each time period are met to a certain extent in the preliminary scheduling.

[0157] Furthermore, the resource scheduling optimization module 40 includes an optimization unit and a constraint unit, wherein,

[0158] The optimization unit is used to perform coordinated optimization of the scheduling of power demand and computing task load at the current moment based on the initial scheduling and according to the Lagrange optimization algorithm, so as to generate a scheduling result that minimizes the scheduling deviation between power demand and computing task load, and dynamically adjust the scheduling result by means of a stability correction factor based on real-time power system parameter information and real-time computing task information.

[0159] The constraint unit is used to constrain the optimization process using constraint conditions.

[0160] The resource scheduling optimization module 40 in this embodiment uses the Lagrange optimization algorithm to optimize resource scheduling, minimize the scheduling error of power demand and computing task load, and consider the synergistic effect between loads; through optimization, the scheduling result of power demand and computing task load at the current moment is obtained, and a stability correction factor is introduced to ensure the stability of the scheduling result.

[0161] To ensure that the scheduling results are optimized while meeting power demand, computational timeliness, and system stability, constraints are used to verify the results during the optimization process, thus ensuring the feasibility and stability of the scheduling results.

[0162] Example 3

[0163] As another specific embodiment of the present invention, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the computational-electronic co-optimization method based on digital twins described in Embodiment 1.

[0164] Example 4

[0165] As another specific embodiment of the present invention, this embodiment provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the computer-computer collaborative optimization method based on digital twins described in Embodiment 1.

[0166] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A digital-twin-based algorithmic-electricity collaborative optimization method, characterized in that, The method comprises: acquiring historical power system parameter information and historical computing task information in a first historical period, historical power system parameter information and historical computing task information in a second historical period, and real-time power system parameter information and real-time computing task information; constructing a dynamic recursive state prediction model based on digital twinning using the historical power system parameter information and the historical computing task information in the first historical period; using the constructed dynamic recursive state prediction model based on digital twinning, predicting the power demand and the computing task load at the next moment based on the historical power system parameter information and the historical computing task information in the second historical period, and preliminarily scheduling the power demand and the computing task load at the current moment according to the prediction result; on the basis of the preliminary scheduling, cooperatively optimizing the scheduling of the power demand and the computing task load at the current moment according to a Lagrange optimization algorithm to generate a scheduling result that minimizes the scheduling deviation of the power demand and the computing task load, and dynamically adjusting the scheduling result through a stability correction factor based on the real-time power system parameter information and the real-time computing task information; wherein the constructing a dynamic recursive state prediction model based on digital twinning using the historical power system parameter information and the historical computing task information in the first historical period comprises constructing a dynamic recursive state prediction model of power demand and a dynamic recursive state prediction model of computing task load, and the constructed dynamic recursive state prediction model of power demand is as follows: In the formula, It is a forecast of electricity demand; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The electricity demand; It is a historic moment The weighting coefficients of the electricity demand data represent the degree of influence of past time points on electricity demand forecasts; It is a historic moment The dynamic adjustment coefficient of the power demand data is adjusted in real time as the load fluctuation rate changes; the constructed dynamic recursive state prediction model of computing task load is as follows: In the formula, is the predicted value of the computing task load; is the size of the historical data window, representing the number of historical time points considered; represents the current time point, represents the next time point; is the computing task load at the historical time point ; is the computing task load at the historical time point ; is the weighting coefficient of the computing task load data at the historical time point , representing the degree of influence of past time points on the prediction of the computing task load; is the dynamic adjustment coefficient of the computing task load data at the historical time point , used to adapt to the rate of fluctuations in the computing task load.

2. The digital-twin-based algorithm-electricity co-optimization method of claim 1, wherein, The power system parameter information includes power load, current, voltage, power, and temperature, and the computing task information includes execution progress, load distribution, response time, and energy consumption of the computing task.

3. The digital-twin-based algorithm-electronics co-optimization method of claim 1, wherein, The preliminary scheduling of the power demand and the computing task load at the current moment according to the prediction result comprises preliminarily scheduling the power demand and the computing task load at the current moment through the following formula: In the formula, and is the maximum power demand limit and the maximum computing task load limit at time The preliminary allocation of power demand and computing task load; and are the predicted values of power demand and computing task load; is the real-time load adaptation factor, which is used to dynamically adjust the power demand scheduling according to the changes of real-time load to cope with load fluctuations; and are the maximum power demand limit and the maximum computing task load limit at time , respectively, to ensure that the system is not overloaded.

4. The digital-twin-based algorithm-electronics co-optimization method of claim 3, wherein, The cooperatively optimizing the scheduling of the power demand and the computing task load at the current moment according to the Lagrange optimization algorithm on the basis of the preliminary scheduling to generate a scheduling result that minimizes the scheduling deviation of the power demand and the computing task load comprises determining a target function of the scheduling error of the power demand and the computing task load, and obtaining an allocation strategy by minimizing the target function, wherein the target function is: In the formula, It is the objective function for scheduling optimization, used to measure electricity demand. and computational task load The scheduling error is identified and penalized, with the goal of minimizing the scheduling error. It represents the total length of the scheduling period, indicating the number of moments in the entire scheduling cycle; and It is the Lagrange multiplier, used to control the penalty for power demand scheduling errors and calculation task load scheduling errors.

5. An algorithmic-electricity collaborative optimization system based on digital twinning, characterized in that, The system comprises a data acquisition module (10), a prediction model construction module (20), a preliminary scheduling module (30), a resource scheduling optimization module (40), and a scheduling scheme output module (50); The data acquisition module (10) is configured to acquire historical power system parameter information and historical computing task information in a first historical period, historical power system parameter information and historical computing task information in a second historical period, and real-time power system parameter information and real-time computing task information; The prediction model construction module (20) is configured to construct a dynamic recursive state prediction model based on digital twinning using the historical power system parameter information and the historical computing task information in the first historical period; The prediction model construction module (20) is configured to construct a dynamic recursive state prediction model based on digital twinning using the historical power system parameter information and the historical computing task information in the first historical period; The preliminary scheduling module (30) is configured to utilize the constructed digital-twin-based dynamic recursive state prediction model to predict the power demand and the computing task load at the next time based on the historical power system parameter information and the historical computing task information in the second historical period, and to preliminarily schedule the power demand and the computing task load at the current time according to the prediction result. The resource scheduling optimization module (40) is configured to, based on the preliminary scheduling, cooperatively optimize the scheduling of the power demand and the computing task load at the current time according to a Lagrange optimization algorithm, to generate a scheduling result that minimizes the scheduling deviation of the power demand and the computing task load, and to dynamically adjust the scheduling result by a stability correction factor based on real-time power system parameter information and real-time computing task information. The scheduling scheme output module (50) is configured to output the adjusted scheduling result. The constructing of the digital-twin-based dynamic recursive state prediction model based on the historical power system parameter information and the historical computing task information in the first historical period comprises constructing a dynamic recursive state prediction model of the power demand and a dynamic recursive state prediction model of the computing task load, and the constructed dynamic recursive state prediction model of the power demand is as follows: In the formula, It is a forecast of electricity demand; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The electricity demand; It is a historic moment The weighting coefficients of the electricity demand data represent the degree of influence of past time points on electricity demand forecasts; It is a historic moment The dynamic adjustment coefficient of the power demand data is adjusted in real time as the load fluctuation rate changes; The constructed dynamic recursive state prediction model of the computing task load is as follows: In the formula, It is a predicted value for calculating task load; It represents the size of the historical data window, indicating the number of historical moments considered. Indicates the current moment. Indicates the next moment; It is a historic moment The computational workload; It is a historic moment The weighting coefficients of the computational task load data represent the degree of influence of past time points on the prediction of computational task load. It is a historic moment The dynamic adjustment coefficient of the computational task load data is used to adapt to the rate of fluctuation of the computational task load.

6. The digital-twin-based cyber-physical co-optimization system of claim 5, wherein, The data acquisition module (10) comprises a first historical data acquisition unit, a second historical data acquisition unit, and a real-time data acquisition unit, wherein The first historical data acquisition unit is configured to acquire the historical power system parameter information and the historical computing task information in the first historical period. The second historical data acquisition unit is configured to acquire the historical power system parameter information and the historical computing task information in the second historical period. The real-time data acquisition unit is configured to acquire real-time power system parameter information and real-time computing task information.

7. The digital-twin-based cyber-physical co-optimization system of claim 5, wherein, The preliminary scheduling module (30) comprises a prediction unit and a scheduling unit, wherein The prediction unit is configured to utilize the constructed digital-twin-based dynamic recursive state prediction model to predict the power demand and the computing task load at the next time based on the historical power system parameter information and the historical computing task information in the second historical period, and to pass the prediction result to the scheduling unit. The scheduling unit is configured to preliminarily schedule the power demand and the computing task load at the current time according to the prediction result of the prediction unit.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein, The processor executes the computer program to implement the steps of the method of any one of claims 1-4.

9. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1-4.

Images