An algorithmic collaboration system and method

Abstract

The application provides an algorithm and electricity collaborative system and method, belonging to the field of algorithm and electricity collaboration, comprising: a data acquisition module that acquires power data, algorithm data and price data, and constructs a power supply time sequence diagram and an algorithm demand time sequence diagram; a prediction module that generates power supply prediction values and algorithm demand prediction values based on the power supply time sequence diagram and the algorithm demand time sequence diagram; an intelligent matching module that generates a matching decision scheme based on the power supply prediction values and the algorithm demand prediction values; a task scheduling module that performs differentiated scheduling based on the matching decision scheme; and a feedback optimization module that is used for evaluating the task scheduling result and feeding back the evaluation result to the prediction module and the intelligent matching module. The application predicts power supply and algorithm demand, generates active scheduling decisions based on the prediction results, and the task scheduling module performs spatiotemporal flexible scheduling of tasks, thereby significantly improving the consumption utilization rate of new energy power and the supply stability of algorithm services.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of algorithm and electricity collaboration, in particular to an algorithm and electricity collaboration system and method. BACKGROUND

[0002] With the rapid development of digital economy and the deepening of the "double carbon" target, the energy consumption of large-scale computing power centers continues to rise in the proportion of total social electricity consumption. The diversification of new energy power generation, energy storage systems, and computing power infrastructure into the power system has brought new challenges to power supply and demand balance. The collaborative optimization of computing power infrastructure and the power system has become an important way to reduce carbon emissions and improve energy efficiency.

[0003] Existing algorithm and electricity collaboration technologies mainly make scheduling decisions based on real-time power supply and demand states, lacking the ability to predict future changes in power supply and computing power demand trends. This static matching mode is prone to frequent supply and demand adjustments when facing the volatility of new energy generation and the spatiotemporal changes in computing power demand, leading to system instability. At the same time, existing computing power task classification and scheduling strategies are relatively extensive, and the potential for flexible adjustment of different types of tasks is not fully exploited, failing to fully utilize the adjustable load role of the computing power system in the power system.

[0004] Chinese Patent Publication No. CN119518737A discloses a power grid and energy storage collaborative power distribution network planning method. This method forms a net load curve through source and load prediction, determines the energy storage configuration capacity based on power capacity deficiency calculation, and establishes a cost optimization model for grid and storage collaboration to achieve comprehensive optimization of power grid construction cost and energy storage configuration cost. This method mainly addresses the problem of insufficient power supply capacity or insufficient new energy carrying capacity, and uses the peak shaving function of the energy storage system to alleviate the pressure on the power grid, and considers the cost comparison of various energy storage configuration methods such as leasing energy storage, responding energy storage, and newly built energy storage. However, this method mainly focuses on energy storage configuration planning on the grid side and does not involve active scheduling of computing power loads or algorithm and electricity collaboration optimization, lacking consideration and utilization of the spatiotemporal migration characteristics of computing power tasks. SUMMARY

[0005] Therefore, the present application proposes an algorithm and electricity collaboration system and method, which solves the defect that the computing power system and the power system run independently in the prior art, realizes algorithm and electricity collaboration, and significantly improves the utilization rate of new energy power and the stability of computing power service supply.

[0006] The technical solution of the present application is as follows: The present application provides an algorithm and electricity collaboration system, comprising:

[0007] a data acquisition module for acquiring power data, computing power data, and price data, and constructing power supply time series graphs and computing power demand time series graphs based on the power data and computing power data, respectively;

[0008] a prediction module configured to generate power supply prediction values and computing power demand prediction values based on the power supply time series graph and the computing power demand time series graph, in combination with historical data statistical analysis and real-time weather data correction;

[0009] an intelligent matching module configured to generate a matching decision scheme of computing power tasks and power resources based on the power supply prediction values and the computing power demand prediction values, in combination with time-of-use and regional power price data, through supply and demand balance analysis, cost-benefit analysis and matching threshold judgment;

[0010] a task scheduling module configured to obtain computing power tasks and perform task classification, and to perform differential scheduling of different types of computing power tasks based on the matching decision scheme, computing power data and price data, to generate a task scheduling result;

[0011] a feedback optimization module configured to evaluate the task scheduling result to generate an evaluation result, and to feed back the evaluation result to the prediction module and the intelligent matching module to dynamically adjust prediction parameters of the prediction module and matching thresholds of the intelligent matching module.

[0012] In the above technical solution, preferably, the power data includes adjustable power generation capacity data and non-adjustable power generation capacity data, wherein the adjustable power generation capacity data includes coal power data, gas power data, hydropower data and nuclear power data, and the non-adjustable power generation capacity data includes wind power data and photovoltaic data.

[0013] The computing power data includes computing power node power consumption data, task execution duration data, network transmission delay data and task resource demand data.

[0014] The price data includes time-of-use and regional power price data and computing power service price data.

[0015] In the above technical solution, preferably, the prediction module includes a power supply prediction unit and a computing power demand prediction unit, wherein,

[0016] The power supply prediction unit is configured to calculate an adjustment capacity correction coefficient based on the adjustable power generation capacity data, calculate a weather correction coefficient in combination with weather data, calculate a trend correction coefficient based on historical power generation deviation, and generate power supply prediction values according to the power supply time series graph and current power supply data.

[0017] The computing power demand prediction unit is configured to calculate a mode correction coefficient through load pattern recognition based on the computing power demand time series graph and current computing power demand data, in combination with computing power node power consumption data and task execution duration data, calculate a resource correction coefficient based on task resource demand data and network transmission delay data, calculate a business correction coefficient based on business volume changes, and generate computing power demand prediction values.

[0018] On the basis of the above technical solutions, preferably, the specific steps of the power supply prediction unit include:

[0019] Based on the power supply timing diagram, historical power supply data is obtained, and current meteorological data including wind speed data, wind direction data, solar radiation data and temperature data is obtained, and adjustable power generation capacity data including the current operating state, available capacity and rated capacity information of various types of adjustable generator sets is obtained;

[0020] Based on the wind speed data and the wind direction data, the wind power theoretical value is calculated through the wind turbine power curve, and based on the solar radiation data and the temperature data, the photovoltaic power theoretical value is calculated through the photovoltaic power generation physical model, and the adjustable capacity correction coefficient is calculated by combining the available capacity and the rated capacity ratio of the adjustable generator set;

[0021] Based on the historical power supply data, the historical average power generation is calculated, and based on the historical average power generation and the theoretical power generation, the meteorological correction coefficient is calculated;

[0022] Actual power generation data is obtained, and the trend correction coefficient is calculated based on the deviation of the actual power generation data and the historical power generation;

[0023] Based on the historical power supply data, the adjustable capacity correction coefficient, the meteorological correction coefficient and the trend correction coefficient, the power supply prediction value is calculated.

[0024] On the basis of the above technical solutions, preferably, the specific steps of the computing power demand prediction unit include:

[0025] According to the computing power demand timing diagram, historical computing power demand data is obtained, and a comprehensive load mode sample library is constructed by combining computing power node power consumption data and task execution duration data, the load mode sample library including daily load change mode based on power consumption characteristics classification, weekly load change mode based on execution duration characteristics classification and monthly load change mode based on comprehensive resource demand characteristics classification;

[0026] The computing power demand data within a preset time window at the current time is obtained, and the computing power demand change curve is calculated by combining the current computing power node power consumption data and the task execution duration data, and the current computing power demand change mode is generated;

[0027] The current computing power demand change mode is calculated for similarity with each load change mode in the comprehensive load mode sample library and sorted, the historical computing power demand value corresponding to the highest similarity is determined, and the mode correction coefficient is calculated based on the historical computing power demand value and the computing power node power consumption data;

[0028] Based on the task resource demand data, the resource demand distribution characteristics of the current task are analyzed, the resource consumption influence of cross-node task execution is evaluated in combination with network transmission delay data, and the resource correction coefficient is calculated.

[0029] Obtain the number of computing power tasks submitted at the current moment, and calculate the business correction coefficient based on the deviation between the current number of tasks and the historical average number of tasks;

[0030] The predicted computing power demand is generated based on historical computing power demand data, pattern correction coefficient, resource correction coefficient, and business correction coefficient.

[0031] Based on the above technical solutions, the preferred formula for calculating the predicted power supply value is: ; ; ; in, express Forecasted power supply at any given time. Indicates the current moment. Indicates the prediction time step. express Historical power supply values ​​at any given time This represents the meteorological correction factor. This represents the trend correction coefficient. express Theoretical power generation at any given time express Historical average power generation at any given time Indicates the trend sensitivity coefficient. This indicates the i-th day before the current time. express Actual power generation data at any given time express Historical power generation at any given time This represents the long-term average of historical power generation. express The adjustment capability correction coefficient at any time. express Available capacity of the adjustable generator set at any time. express The rated capacity of the adjustable generator set at any time. This represents the adjustment capacity weighting coefficient. Indicates the length of the historical time window for trend analysis.

[0032] Based on the above technical solutions, preferably, the formula for calculating the predicted computing power demand is as follows: ; ; ;

[0033] in, express Predicted computing power demand at any given time. represents a historical computing power demand value at a time point, represents a mode correction coefficient, represents a service correction coefficient, represents a historical computing power demand value corresponding to the most similar mode at a time point, represents a historical average computing power demand value at a time point, represents a service sensitivity coefficient, represents an actual task submission quantity at a time point, a historical average task submission quantity, represents a resource correction coefficient.

[0034] On the basis of the above technical solutions, preferably, the intelligent matching module includes three matching rules, namely a first matching rule, a second matching rule, and a third matching rule, and the specific steps of the intelligent matching module include:

[0035] A supply-demand safety threshold and a supply-demand gap threshold are set, a supply-demand difference value at a current time point is obtained based on the power supply prediction value and the computing power demand prediction value, and a power cost coefficient at the current time point is calculated based on the time-of-use and regional power price data; wherein the supply-demand safety threshold is less than the supply-demand gap threshold;

[0036] When the power supply prediction value is greater than the computing power demand prediction value, and the supply-demand difference value at the current time point is greater than the supply-demand safety threshold, the adjustable capacity range of the generator set at the current time point is evaluated in combination with the adjustable power generation capacity data, the first matching rule is used, the supply surplus is calculated by accumulating the supply-demand difference value and the power generation adjustment capacity within the prediction time step, the cost-benefit analysis is performed in combination with the computing power node power consumption data and the power cost coefficient at the current time point, and the positive dispatching decision scheme is generated based on the supply surplus and the cost-benefit analysis result; the positive dispatching decision scheme is to instruct the task scheduling module to preferentially schedule high-power-consumption tasks for execution;

[0037] When the power supply prediction value is less than the computing power demand prediction value, and the absolute value of the supply-demand difference value at the current time point is greater than the supply-demand gap threshold, the second matching rule is used, the time point at which the power supply meets the computing power demand is determined by calculating forward by time point, the duration of the power supply gap is determined, the time cost of task migration is calculated in combination with the task execution duration data and the network transmission delay data, and the conservative dispatching decision scheme is generated based on the duration of the power supply gap and the task migration cost; the conservative dispatching decision scheme is to instruct the task scheduling module to perform task delay scheduling or spatial migration scheduling;

[0038] When the absolute value of the supply-demand difference at the current time is between the supply-demand safety threshold and the supply-demand gap threshold, a third matching rule is used to statistically analyze the historical supply-demand prediction accuracy, combine the task resource demand data and the computing power service price data for comprehensive cost analysis, and generate a balanced scheduling decision scheme based on the historical supply-demand prediction accuracy and the comprehensive cost analysis result; the balanced scheduling decision scheme is a scheduling decision scheme that indicates that the task scheduling module adjusts the scheduling decision scheme according to the historical supply-demand prediction accuracy.

[0039] Further preferably, the specific steps of the task scheduling module include:

[0040] The computing power tasks are obtained and divided into non-migratable tasks, time-migratable tasks and space-migratable tasks.

[0041] When the matching decision scheme is an aggressive scheduling decision scheme, the power consumption information of each task in the current period is obtained, the non-migratable tasks are sorted in descending order of power consumption and high-power-consumption tasks are preferentially scheduled, and the execution time of the time-migratable tasks is advanced to the current period of sufficient power supply;

[0042] When the matching decision scheme is a conservative scheduling decision scheme, the non-migratable tasks are kept normally executed, the time-migratable tasks are delayed to be executed after the power supply gap ends, and the space-migratable tasks are transferred to other computing power nodes with sufficient power supply for execution;

[0043] When the matching decision scheme is a balanced scheduling decision scheme, the historical supply-demand prediction accuracy is obtained, when the prediction accuracy is higher than a set threshold, the aggressive scheduling decision scheme is executed, and when the prediction accuracy is lower than the set threshold, the scheduling proportion of high-power-consumption tasks is reduced and the time interval of task execution is increased.

[0044] The present application provides a kind of algorithm electricity collaborative algorithm, applied to the collaborative model described above, including the following steps:

[0045] S1, power data, computing power data and price data are collected, and power supply time series diagram and computing power demand time series diagram are respectively constructed based on power data and computing power data;

[0046] S2, based on the power supply time series diagram and the computing power demand time series diagram, combined with historical data statistical analysis and real-time weather data correction, power supply prediction value and computing power demand prediction value are generated;

[0047] S3, based on the power supply prediction value and the computing power demand prediction value, combined with time-sharing and regional power price data, through supply-demand balance analysis, cost-benefit analysis and matching threshold judgment, a matching decision scheme of computing power tasks and power resources is generated;

[0048] S4, acquire computing power tasks and perform task classification, different types of computing power tasks are differentially scheduled based on the matching decision scheme, computing power data and price data, and a task scheduling result is generated;

[0049] S5, evaluate the task scheduling result and generate an evaluation result, and dynamically adjust the power supply prediction value and the computing power demand prediction value and the matching threshold of the intelligent matching module according to the evaluation result.

[0050] The algorithm and electricity collaborative system and method of the present application have the following beneficial effects compared with the prior art:

[0051] (1) The power supply and computing power demand are predicted by the prediction module, the intelligent matching module generates an active scheduling decision based on the prediction result, and the task scheduling module performs time and space flexible scheduling of the task. Compared with the traditional passive response scheduling method, the utilization rate of new energy power and the stability of computing power service supply are significantly improved;

[0052] (2) The double correction mechanism of the meteorological correction coefficient and the trend correction coefficient significantly improves the accuracy and practicality of new energy power generation power prediction, provides a more reliable data basis for algorithm and electricity collaborative decision-making, and effectively reduces the adverse effects of prediction errors on scheduling decisions;

[0053] (3) Multi-level risk identification is realized by setting supply and demand safety thresholds and supply and demand gap thresholds, precise identification and differential treatment of different degrees of supply and demand imbalance are realized, and when the supply and demand difference is in different threshold intervals, the corresponding matching strategy is automatically selected, so that the computing power task scheduling can actively adapt to the changes in the operation state of the power system, and the algorithm and electricity collaborative ability is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0055] Figure 1 The system block diagram of the algorithm and electricity collaborative system of the present application;

[0056] Figure 2 The flowchart of the algorithm and electricity collaborative method of the present application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0058] As shown in Figure 1 The present application provides an algorithm power collaborative system, comprising:

[0059] A data acquisition module is configured to acquire power data, algorithm power data and price data, and construct power supply time series graphs and algorithm power demand time series graphs based on the power data and the algorithm power data, respectively;

[0060] A prediction module is configured to generate power supply prediction values and algorithm power demand prediction values based on the power supply time series graphs and the algorithm power demand time series graphs, in combination with historical data statistical analysis and real-time weather data correction;

[0061] An intelligent matching module is configured to generate a matching decision scheme of algorithm power tasks and power resources based on the power supply prediction values and the algorithm power demand prediction values, in combination with time-of-use and regional power price data, through supply and demand balance analysis, cost-benefit analysis and matching threshold judgment;

[0062] A task scheduling module is configured to acquire algorithm power tasks and perform task classification, and perform differential scheduling of different types of algorithm power tasks based on the matching decision scheme, the algorithm power data and the price data, to generate a task scheduling result;

[0063] A feedback optimization module is configured to evaluate the task scheduling result to generate an evaluation result, and feed back the evaluation result to the prediction module and the intelligent matching module to dynamically adjust prediction parameters of the prediction module and matching thresholds of the intelligent matching module.

[0064] The present application predicts power supply and algorithm power demand through the prediction module, generates an active scheduling decision based on the prediction result through the intelligent matching module, and performs spatiotemporal flexible scheduling of tasks through the task scheduling module. Compared with the traditional passive response scheduling method, the present application significantly improves the consumption and utilization rate of new energy power and the supply stability of algorithm power service.

[0065] In an embodiment of the present application, the power data includes adjustable power generation capacity data and non-adjustable power generation capacity data, wherein the adjustable power generation capacity data includes coal power data, gas power data, hydro power data and nuclear power data, and the non-adjustable power generation capacity data includes wind power data and photovoltaic data;

[0066] The algorithm power data includes algorithm power node power consumption data, task execution duration data, network transmission delay data and task resource demand data.

[0067] The price data includes time-sharing and regional power price data and computing power service price data.

[0068] It can be understood that the adjustable power generation capacity data refers to a power generation resource in a power system that can accept dispatching instructions and change the output level within a certain time to maintain the balance between supply and demand of the power system, and the characteristics of such power source are controllable output, timely response and wide adjustment range, and the power generation power can be actively adjusted according to the dispatching demand of the power grid; the non-adjustable power generation capacity data refers to a power generation resource whose output is mainly restricted by external natural conditions and is difficult to actively adjust according to the dispatching demand, and the characteristics of such power source are large output fluctuation, strong intermittence and relatively poor predictability, and the power generation power mainly depends on natural factors such as weather conditions.

[0069] Among them, the coal power data is obtained by the power plant monitoring system to obtain the rated capacity, current load rate, climbing rate and other key parameters of the power plant, and is quantified as the power range that can be adjusted within every 15 minutes; the gas power data is obtained by the operation monitoring system of the gas turbine unit to obtain the start-stop state, fuel supply condition, adjustment rate and other parameters, and is quantified as the rapid adjustment capacity; the hydropower data is obtained by the automation system of the hydropower plant to obtain the reservoir water level, discharge flow, unit operation state and other parameters, and is quantified as the adjustable output range; the nuclear power data is obtained by the control system of the nuclear power plant to obtain the reactor power level, adjustment mode and other parameters, and is quantified as the limited adjustment capacity.

[0070] The wind power data is obtained by the wind farm monitoring system and weather observation equipment to obtain the wind speed, wind direction, unit availability and other parameters, and is quantified by combining the wind turbine power curve to obtain the real-time power generation power; the photovoltaic data is obtained by the photovoltaic power station monitoring system and weather station to obtain the solar radiation intensity, component temperature, inverter efficiency and other parameters, and is quantified by combining the photovoltaic power generation physical model to obtain the real-time power generation power.

[0071] The power price data is obtained by the power trading platform to obtain the price information of different time periods and different regions, and is quantified as a cost parameter; the computing power service price data is obtained by the cloud service platform to obtain the price system of different computing power services, and provides a reference for cost optimization of task scheduling.

[0072] In an embodiment of the present application, the prediction module includes a power supply prediction unit and a computing power demand prediction unit, wherein,

[0073] The power supply prediction unit is configured to calculate an adjustment capacity correction coefficient based on the adjustable power generation capacity data, calculate a weather correction coefficient in combination with weather data, calculate a trend correction coefficient based on historical power generation deviation, and generate a power supply prediction value according to the power supply time sequence diagram and the current time power supply data;

[0074] The computing power demand prediction unit is used for calculating a mode correction coefficient through load mode recognition according to the computing power demand time sequence diagram, the current time computing power demand data, the computing power node power consumption data and the task execution time length data, calculating a resource correction coefficient based on the task resource demand data and the network transmission delay data, calculating a service correction coefficient based on the service volume change, and generating a computing power demand prediction value.

[0075] It can be understood that the meteorological data is collected by using a multi-source fusion technology, and high temporal and spatial resolution meteorological element fields are formed by comprehensively utilizing ground meteorological station observation data, upper air sounding data, satellite remote sensing data and numerical weather prediction model output.

[0076] The prediction module of the present application significantly improves the accuracy and practicability of new energy power prediction through the double correction mechanism of meteorological correction coefficient and trend correction coefficient, provides more reliable data basis for coordinated decision-making of computing and electricity, and effectively reduces the adverse effects of prediction error on dispatching decision.

[0077] In an embodiment of the present application, the specific steps of the power supply prediction unit include:

[0078] Based on the power supply time sequence diagram, historical power supply data is obtained, and current meteorological data including wind speed data, wind direction data, solar radiation data and temperature data is obtained, and adjustable generating capacity data including the current operating state, available capacity and rated capacity information of various types of adjustable generating units is obtained;

[0079] Based on the wind speed data and the wind direction data, the wind power theoretical value is calculated through the wind turbine power curve, and based on the solar radiation data and the temperature data, the photovoltaic power theoretical value is calculated through the photovoltaic power generation physical model, and the adjustment capacity correction coefficient is calculated by combining the available capacity and the rated capacity ratio of the adjustable generating unit;

[0080] Based on the historical power supply data, the historical average generating power is calculated, and based on the historical average generating power and the theoretical generating power, the meteorological correction coefficient is calculated;

[0081] Actual generating data is obtained, and the trend correction coefficient is calculated based on the deviation of the actual generating data and the historical generating power;

[0082] Based on the historical power supply data, the adjustment capacity correction coefficient, the meteorological correction coefficient and the trend correction coefficient, the power supply prediction value is calculated, and the calculation formula is: ; ; ; Wherein, represents the power supply prediction value at the time t, represents the current time, denotes a predicted time step, denotes a historical power supply value at a time point, denotes a meteorological correction coefficient, denotes a trend correction coefficient, denotes a theoretical power generation at a time point, denotes a historical average power generation at a time point, denotes a trend sensitivity coefficient, denotes the i-th day before the current time point, denotes actual power generation data at a time point, denotes a historical power generation at a time point, denotes a long-term average of the historical power generation, denotes a regulation capacity correction coefficient at a time point, denotes an available capacity of an adjustable generator set at a time point, denotes a rated capacity of an adjustable generator set at a time point, denotes a regulation capacity weight coefficient, denotes a historical time window length for trend analysis.

[0083] It can be understood that, is the historical power generation average value up to the time point t. The wind power prediction adopts the wind turbine power curve method to calculate the theoretical value of wind power based on wind speed data and wind direction data. The power generation of the wind turbine is closely related to the wind speed. When the wind speed is lower than the cut-in wind speed (i.e. the minimum wind speed at which the wind turbine can start and begin to generate power), the wind turbine cannot start and the power generation is zero; when the wind speed exceeds the cut-out wind speed (the maximum wind speed at which the wind turbine must be stopped for safety), the wind turbine stops to protect, and the power generation decreases to zero; only between the cut-in wind speed and the cut-out wind speed can the wind turbine generate power normally, and the power generation remains at the rated power level; the photovoltaic power generation is mainly affected by the solar radiation intensity, showing obvious daily variation law, and is also affected by factors such as cloud cover and temperature change, and has strong volatility.

[0084] The meteorological correction coefficient reflects the deviation of the current meteorological condition from the historical average meteorological condition. When the meteorological condition is better than the historical average level, the correction coefficient is greater than 1, indicating that the power generation will be higher than the historical average; when the meteorological condition is worse than the historical average level, the correction coefficient is less than 1, indicating that the power generation will be lower than the historical average.

[0085] The trend correction coefficient is used to capture systematic bias and short-term trend changes, that is, by analyzing the deviation of recent actual power generation from the historical average value, the changing trend of the system operation state is identified. When the recent actual power generation is continuously higher than the historical average value, the trend correction coefficient is greater than 1, indicating that the predicted value should be appropriately adjusted upward; when the recent actual power generation is continuously lower than the historical average value, the trend correction coefficient is less than 1, indicating that the predicted value should be appropriately adjusted downward. The trend sensitive coefficient is determined by using the least square regression method, and the parameter value is optimized by minimizing the prediction error. Through regression analysis of historical data, the optimal can be obtained, and is updated regularly according to the operation data.

[0086] In an embodiment of the present application, the historical average power generation is calculated by using the sliding window statistical method to sum and average the power generation, so that the seasonal variation of the power generation can be captured, and the sensitivity to the recent trend change is maintained through the sliding window.

[0087] In an embodiment of the present application, the adjustable power generation data is evaluated by using a hierarchical evaluation method in power supply prediction, and the adjustment capacity correction coefficient is dynamically calculated by monitoring the operation state of each type of adjustable generator in real time. For coal-fired units, the current load rate, boiler temperature, turbine state and other operation parameters are obtained, and the adjustable capacity in the prediction period is calculated in combination with the climbing rate and adjustment range of the unit; for gas-fired units, the gas supply condition, combustion chamber temperature and turbine blade state are focused on; for hydroelectric units, the current water level, inflow, outflow and key parameters such as the operation state of the unit turbine are obtained; for nuclear power units, the core operation parameters such as reactor power level, steam generator state and control rod position are monitored.

[0088] The calculation of the adjustment capacity correction coefficient adopts a weighted comprehensive evaluation method, and the system calculates the coal-fired adjustment coefficient, the gas-fired adjustment coefficient, the hydroelectric adjustment coefficient and the nuclear power adjustment coefficient according to the real-time available capacity, the adjustment rate characteristics and the operation constraints of each type of adjustable unit, and then performs weighted synthesis according to the installed capacity proportion and the current operation state of each type of unit in the system.

[0089] In an embodiment of the present application, the specific steps of the computing power demand prediction unit include:

[0090] The historical computing power demand data is obtained according to the computing power demand time sequence diagram, and a comprehensive load mode sample library is constructed in combination with the computing power node power consumption data and the task execution duration data, the load mode sample library includes daily load change modes classified based on power consumption characteristics, weekly load change modes classified based on execution duration characteristics, and monthly load change modes classified based on comprehensive resource demand characteristics;

[0091] Obtain computing power demand data within a set time window before the current moment, combine it with the current computing power node power consumption data and task execution duration data to calculate the computing power demand change curve, and generate the current computing power demand change pattern;

[0092] The similarity between the current computing power demand change pattern and each load change pattern in the comprehensive load pattern sample library is calculated and sorted. The historical computing power demand value corresponding to the highest similarity is determined. The pattern correction coefficient is calculated based on the historical computing power demand value and the power consumption data of the computing power node.

[0093] Based on the analysis of task resource demand data, the resource demand distribution characteristics of the current task are analyzed, and the impact of cross-node task execution on resource consumption is evaluated in combination with network transmission delay data, and the resource correction coefficient is calculated.

[0094] Obtain the number of computing power tasks submitted at the current moment, and calculate the business correction coefficient based on the deviation between the current number of tasks and the historical average number of tasks;

[0095] The predicted computing power demand is generated based on historical computing power demand data, pattern correction coefficient, resource correction coefficient, and business correction coefficient. The calculation formula is as follows: ; ; ; ; ;in, express Predicted computing power demand at any given time. express Historical computing power requirements at any given time Indicates the mode correction coefficient, This represents the business adjustment factor. express The historical computing power requirement corresponding to the most similar pattern at any given time. express Historical average computing power requirement at any given time. This represents the business sensitivity coefficient, with a value range of [0, 1]. express The actual number of tasks submitted at any given time. This represents the historical average number of task submissions. Represents the resource adjustment factor. express The average resource demand intensity of the current task at any given time. This indicates the historical average resource demand intensity for the task. The standard deviation of the number of historical task submissions.

[0096] Understandably, the average resource demand intensity is obtained by weighting the CPU cores, memory, and storage requirements of a computing task.

[0097] The mode correction coefficient reflects the deviation of the current demand mode from the historical average mode, and when the current mode is similar to the historical high demand mode, the correction coefficient is greater than 1; and when the current mode is similar to the historical low demand mode, the correction coefficient is less than 1.

[0098] In an embodiment of the present application, the computing power node power consumption data and the task execution time data are fused in multiple dimensions in the computing power demand prediction, and more accurate load mode sample libraries are constructed by analyzing the power consumption characteristics and time characteristics of different types of tasks. The computing power tasks are divided into high-power tasks (power consumption exceeding 1000W), medium-power tasks (power consumption between 500-1000W) and low-power tasks (power consumption less than 500W) according to the power consumption level, and at the same time, short-time tasks (execution time less than 1 hour), medium-time tasks (execution time 1-8 hours) and long-time tasks (execution time more than 8 hours) are divided according to the execution time, and nine basic task type modes are established by this two-dimensional classification method.

[0099] In an embodiment of the present application, the generation of the current computing power demand change mode adopts a sliding window technology, and the computing power demand data within a set time window before the current time is obtained to calculate the computing power demand change curve. The selection of the time window is determined according to the prediction time scale. For hour-level prediction, the data of the past 24 hours is used; for day-level prediction, the data of the past 7 days is used; and for week-level prediction, the data of the past 4 weeks is used.

[0100] In an embodiment of the present application, the similarity calculation adopts a dynamic time warping algorithm, and the calculation formula is: Wherein, represents the similarity between the current computing power demand change mode and the historical load change mode , represents the current computing power demand change mode, represents the historical load change mode, represents the dynamic time warping distance between the current computing power demand change mode and the historical load change mode , represents the maximum value of the absolute value of all elements in the sequence between the current computing power demand change mode and the historical load change mode .

[0101] In an embodiment of the present application, the intelligent matching module includes three matching rules, namely a first matching rule, a second matching rule and a third matching rule, and the specific steps of the intelligent matching module include:

[0102] The supply-demand safety threshold and the supply-demand gap threshold are set, and a supply-demand difference value at the current time is obtained based on the power supply prediction value and the computing power demand prediction value, and a power cost coefficient at the current time is calculated based on the time-of-use and regional power price data; wherein the supply-demand safety threshold is less than the supply-demand gap threshold;

[0103] When the power supply prediction value is greater than the computing power demand prediction value, and the supply-demand difference value at the current time is greater than the supply-demand safety threshold, the adjustable capacity range of the generator set at the current time is evaluated in combination with the adjustable power generation capacity data, the supply surplus amount is calculated by accumulating the supply-demand difference value and the power generation adjustment capacity within the prediction time step using the first matching rule, the cost-benefit analysis is performed in combination with the computing power node power consumption data and the power cost coefficient at the current time, and the positive scheduling decision scheme is generated based on the supply surplus amount and the cost-benefit analysis result; the positive scheduling decision scheme is to instruct the task scheduling module to preferentially schedule high-power-consumption tasks.

[0104] When the power supply prediction value is less than the computing power demand prediction value, and the absolute value of the supply-demand difference value at the current time is greater than the supply-demand gap threshold, the time at which the power supply meets the computing power demand is determined by calculating forward by time step using the second matching rule, the duration of the power supply gap is determined, the time cost of task migration is calculated in combination with the task execution time length data and the network transmission delay data, and the conservative scheduling decision scheme is generated based on the duration of the power supply gap and the task migration cost; the conservative scheduling decision scheme is to instruct the task scheduling module to perform task delay scheduling or spatial migration scheduling.

[0105] When the absolute value of the supply-demand difference value at the current time is between the supply-demand safety threshold and the supply-demand gap threshold, the third matching rule is used to statistically analyze the historical supply-demand prediction accuracy, the comprehensive cost analysis is performed in combination with the task resource demand data and the computing power service price data, and the balanced scheduling decision scheme is generated based on the historical supply-demand prediction accuracy and the comprehensive cost analysis result; the balanced scheduling decision scheme is to instruct the task scheduling module to adjust the scheduling decision scheme according to the historical supply-demand prediction accuracy.

[0106] It can be understood that the preferential scheduling of high-power-consumption tasks is realized based on the power consumption sorting mechanism, the task scheduling module obtains the power consumption information of each task in the current period, sorts the tasks according to the power consumption from high to low, and preferentially executes the high-power-consumption computing power tasks, so as to maximize the utilization efficiency of power resources.

[0107] The supply-demand safety threshold is determined based on statistical analysis of historical power supply and demand data, and the supply-demand gap threshold is determined based on the maximum bearing capacity of the power system and the risk control requirement, and the threshold value thereof can timely trigger the conservative scheduling strategy.

[0108] In an embodiment of the present application, the specific steps of the feedback optimization module include evaluation index calculation, deviation analysis and parameter adjustment, and specifically include:

[0109] By regularly counting the task scheduling success rate, average task completion time, power resource utilization rate and system load balancing degree in a certain time window, and comparing and analyzing with the preset target value, the achievement degree and deviation degree of each index are calculated;

[0110] By analyzing the deviation direction and amplitude of each evaluation index, the main problems and improvement space in the current scheduling strategy are identified, for example, when the task scheduling success rate is lower than the target value, it needs to be further analyzed whether it is caused by insufficient prediction accuracy, improper matching threshold setting or too strict task classification standard;

[0111] By analyzing the distribution characteristics and variation trend of the prediction error, an adaptive adjustment strategy is adopted to gradually optimize the parameter value, so that the prediction result is closer to the actual situation.

[0112] The present application realizes accurate identification and differential treatment of different degrees of supply and demand imbalance state by setting supply and demand safety threshold and supply and demand gap threshold, and automatically selects the corresponding matching strategy when the supply and demand difference is in different threshold intervals, so that the computing power task scheduling can actively adapt to the change of the operation state of the power system, and effectively improve the computing power and power coordination ability.

[0113] In an embodiment of the present application, when the meteorological data shows that the wind speed or solar radiation intensity in the future period is high, the wind power and photovoltaic power generation power is expected to increase, and the supply surplus increases accordingly, so the intensity of active scheduling can be appropriately strengthened.

[0114] In an embodiment of the present application, the conservative scheduling decision scheme includes task delay scheduling and space migration scheduling, the task delay scheduling is for time-migratable tasks, that is, the execution time of the time-migratable tasks is postponed after the power supply gap ends; the space migration scheduling is for space-migratable tasks, that is, the space-migratable tasks are transferred to other computing power nodes with sufficient power supply for execution.

[0115] In an embodiment of the present application, the statistical method of historical supply and demand prediction accuracy rate adopts a sliding window method, and the calculation formula is: Wherein, represents the historical supply and demand prediction accuracy rate, represents the statistical window length, usually takes 168, which is the number of hours in a week, represents the time serial number, represents the power supply prediction value at the jth moment, represents the actual power supply value at the jth moment.

[0116] In an embodiment of the present application, the specific steps of the task scheduling module include:

[0117] The computing power task is acquired and divided into non-migratable tasks, time-migratable tasks and space-migratable tasks.

[0118] When the matching decision scheme is the positive scheduling decision scheme, the power consumption information of each task in the current period is acquired, the non-migratable tasks are sorted in descending order of power consumption and the high-power consumption tasks are preferentially scheduled, and the execution time of the time-migratable tasks is advanced to the current power sufficient period.

[0119] When the matching decision scheme is the conservative scheduling decision scheme, the normal execution of the non-migratable tasks is maintained, the time-migratable tasks are delayed to be executed after the power supply gap ends, and the space-migratable tasks are transferred to other computing power nodes with sufficient power supply for execution.

[0120] When the matching decision scheme is the balanced scheduling decision scheme, the historical supply-demand prediction accuracy is acquired, when the prediction accuracy is higher than the set threshold, the positive scheduling decision scheme is executed, and when the prediction accuracy is lower than the set threshold, the scheduling proportion of the high-power consumption tasks is reduced and the time interval of task execution is increased.

[0121] It can be understood that the non-migratable tasks refer to tasks with strict requirements for execution time and execution location, which must be executed at a specified time and at a specified node; the time-migratable tasks refer to tasks with relatively fixed execution location but certain flexibility in execution time; and the space-migratable tasks refer to tasks with relatively determined execution time but flexible execution location.

[0122] The generation of the balanced scheduling decision scheme is realized based on a prediction accuracy threshold judgment mechanism, when the historical supply-demand prediction accuracy is greater than the set threshold, it indicates that the prediction module has high reliability, the balanced scheduling decision scheme tends to positive scheduling, and the first matching rule is executed according to the positive scheduling decision scheme, that is, the high-power consumption tasks are preferentially scheduled and the time-migratable tasks are executed in advance; when the historical supply-demand prediction accuracy is not greater than the set threshold, it indicates that the reliability of the prediction module is limited, the balanced scheduling decision scheme tends to conservative scheduling, which is specifically manifested as reducing the scheduling proportion of the high-power consumption tasks and increasing the time interval of task execution.

[0123] The present application overcomes the technical limitation that the traditional scheduling method is not accurate enough in distinguishing task types by dividing the computing power tasks and adopting differentiated scheduling decision schemes for different types of tasks, and improves the power resource utilization rate and reduces the computing power task execution cost through the collaborative optimization of task execution time and space.

[0124] In an embodiment of the present application, the time window optimization algorithm is adopted to realize the advance execution strategy of the time-migratable tasks, and the optimal task execution time is determined by analyzing the power sufficient period and the task time constraint.

[0125] In an embodiment of the present application, the evaluation of the adjustable power generation capacity data adopts a unit regulation capacity analysis method to determine the power generation regulation space of the system by analyzing the operating state of each type of adjustable power generation unit at the current time. For coal-fired power units, the current load rate, maximum output and minimum output data of the unit are obtained, and the range of power generation that can be increased or decreased within the predicted time period is calculated; for gas-fired power units, the start-stop state and fuel supply of the gas turbine are obtained; for hydroelectric power units, the regulation capacity of the hydroelectric power is evaluated in combination with the reservoir water level, discharge flow constraint and unit operating state; for nuclear power units, the reactor power level, control rod position and steam generator operating parameters are monitored to evaluate the load tracking capacity of the nuclear power unit.

[0126] When the intelligent matching module detects that the power supply prediction value is greater than the computing power demand prediction value and the difference exceeds the safety threshold, it is analyzed whether the adjustable power generation unit has the space to reduce the output. If the adjustable power generation unit is currently operating in a high load state and has sufficient regulation margin, the aggressive scheduling decision scheme will instruct to appropriately reduce the output of part of the adjustable unit to reserve power space for the increase of computing power load, realizing the dynamic coordination of power supply and computing power demand.

[0127] In an embodiment of the present application, the time and regional power price data adopts a dynamic cost optimization strategy to guide the scheduling decision of the computing power task according to the periodical change and regional difference of the power price. The calculation of the power cost coefficient is based on the ratio of the current power price to the historical average power price. When the power cost coefficient is greater than 1, it indicates that the current electricity price is higher than the average level, and when the power cost coefficient is less than 1, it indicates that the current electricity price is lower than the average level. When generating the aggressive scheduling decision scheme, the intelligent matching module gives priority to increasing the execution proportion of high-power consumption tasks in the period when the power price is relatively low, and executes low-power consumption tasks or delays part of the time migration tasks in the period when the power price is relatively high.

[0128] The cost-benefit analysis includes task execution cost calculation and cross-regional cost comparison. The task execution cost calculation is based on the computing power node power consumption data and the current time power price data. The power cost of the task is obtained by multiplying the expected power consumption of the task by the corresponding power price, and then the total execution cost of the task is calculated by combining the computing power service price data. The cross-regional cost includes the power cost of the target node, the computing power service cost and the delay cost generated by network transmission.

[0129] In an embodiment of the present application, the time cost calculation of task migration adopts a comprehensive evaluation method, based on the execution time data analysis of different types of tasks, the execution time of tasks is divided into short-time tasks (execution time less than 1 hour), medium-time tasks (execution time 1-8 hours) and long-time tasks (execution time more than 8 hours), different types of tasks have different sensitivity to delay, short-time tasks are most sensitive to delay, and long-time tasks have relatively high delay tolerance.

[0130] By monitoring the network delay between different computing power nodes in real time, a dynamic network delay matrix is established, when the intelligent matching module needs to generate a space migration scheduling scheme, the power supply condition of the target node, the network transmission delay and the sensitivity of the task to the delay are comprehensively considered, for delay-sensitive tasks, the adjacent nodes with lower network delay are preferentially selected for migration, and for delay-tolerant tasks, the power cost advantage of the target node is considered.

[0131] In an embodiment of the present application, the comprehensive cost analysis adopts a multi-dimensional cost modeling method, and the power cost, computing power service cost, network transmission cost and time delay cost are included in a unified cost evaluation framework. When the intelligent matching module generates a balanced scheduling decision scheme, first, the weight coefficients of various costs are calculated, the weight coefficient of the power cost is determined based on the deviation degree of the current price from the historical price, the greater the deviation degree, the higher the weight coefficient; the weight coefficient of the computing power service cost is determined based on the resource supply-demand ratio, the more scarce the resource, the higher the weight coefficient; the weight coefficient of the network transmission cost is determined based on the delay sensitivity, the higher the delay sensitivity, the higher the weight coefficient; the weight coefficient of the time delay cost is determined based on the task urgency, the higher the urgency, the higher the weight coefficient.

[0132] The intelligent matching module divides the prediction time period into multiple time windows, analyzes the executable task combination and the corresponding comprehensive cost in each time window, and finds out the task scheduling sequence that minimizes the comprehensive cost in the whole prediction time period through recursive calculation. When the historical supply and demand prediction accuracy is high, task scheduling is performed based on the cost difference across time periods; when the historical supply and demand prediction accuracy is low, conservative decision is made based on the current cost situation.

[0133] As shown in Figure 2 The present application also provides an algorithm and power cooperation method, applied to the algorithm and power cooperation system as described above, comprising the following steps:

[0134] S1, collect power data, computing power data and price data, and construct power supply time series graph and computing power demand time series graph based on the power data and computing power data respectively;

[0135] S2, based on the power supply timing diagram and the computing power demand timing diagram, combined with historical data statistical analysis and real-time weather data correction, generate power supply prediction value and computing power demand prediction value;

[0136] S3, based on the power supply prediction value and the computing power demand prediction value, combined with the time-sharing and regional power price data, through supply and demand balance analysis, cost-benefit analysis and matching threshold judgment, generate the matching decision scheme of computing power task and power resource;

[0137] S4, obtain the computing power task and perform task classification, based on the matching decision scheme, computing power data and price data, different types of computing power tasks are differentially scheduled to generate task scheduling results;

[0138] S5, evaluate the task scheduling results and generate evaluation results, dynamically adjust the power supply prediction value and the computing power demand prediction value and the matching threshold of the intelligent matching module according to the evaluation results.

[0139] The above is only a preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An algorithmic co-simulation system, characterized by: The method comprises the following steps: a data collection module is used to collect power data, computing power data and price data, and to construct power supply time series and computing power demand time series based on the power data and computing power data respectively; a prediction module is used to generate power supply prediction values and computing power demand prediction values based on the power supply time series and computing power demand time series, combined with historical data statistical analysis and real-time weather data correction; an intelligent matching module is used to generate a matching decision scheme of computing power tasks and power resources based on the power supply prediction values and computing power demand prediction values, combined with time-of-use and regional power price data, through supply and demand balance analysis, cost-benefit analysis and matching threshold judgment; the specific steps of the intelligent matching module include: setting a supply and demand safety threshold and a supply and demand gap threshold, and obtaining a supply and demand difference value at the current time based on the power supply prediction values and computing power demand prediction values, and calculating a power cost coefficient at the current time based on the time-of-use and regional power price data; wherein the supply and demand safety threshold is less than the supply and demand gap threshold; comparing the supply and demand difference value at the current time with the supply and demand safety threshold and the supply and demand gap threshold to generate an aggressive scheduling decision scheme, a conservative scheduling decision scheme and a balanced scheduling decision scheme, wherein the aggressive scheduling decision scheme indicates that the task scheduling module prioritizes scheduling high-power-consumption tasks; the conservative scheduling decision scheme indicates that the task scheduling module performs task delay scheduling or spatial migration scheduling; and the balanced scheduling decision scheme indicates that the task scheduling module adjusts the scheduling decision scheme according to the historical supply and demand prediction accuracy; a task scheduling module is used to obtain computing power tasks and perform task classification, and to perform differentiated scheduling on different types of computing power tasks based on the matching decision scheme, computing power data and price data, and to generate a task scheduling result; a feedback optimization module is used to evaluate the task scheduling result to generate an evaluation result, and to feed back the evaluation result to the prediction module and the intelligent matching module to dynamically adjust the prediction parameters of the prediction module and the matching threshold of the intelligent matching module.

2. The system of claim 1, wherein: The power data includes adjustable power generation capacity data and non-adjustable power generation capacity data, wherein the adjustable power generation capacity data includes coal power data, gas power data, hydropower data and nuclear power data, and the non-adjustable power generation capacity data includes wind power data and photovoltaic data; The computing power data includes computing power node power consumption data, task execution duration data, network transmission delay data and task resource demand data; The price data includes time-of-use and regional power price data and computing power service price data.

3. The system of claim 1, wherein: The prediction module includes a power supply prediction unit and a computing power demand prediction unit, wherein the power supply prediction unit is used to calculate an adjustment capacity correction coefficient based on the adjustable power generation capacity data, calculate a weather correction coefficient combined with weather data, calculate a trend correction coefficient based on historical power generation deviation, and generate power supply prediction values based on the power supply time series and current power supply data; The computing power demand prediction unit is configured to calculate a mode correction coefficient by load mode recognition based on the computing power demand time sequence diagram, current time computing power demand data, computing power node power consumption data and task execution duration data, calculate a resource correction coefficient based on task resource demand data and network transmission delay data, calculate a business correction coefficient based on traffic volume changes, and generate a computing power demand prediction value.

4. The system of claim 3, wherein the at least one processor is further configured to: The specific steps of the power supply prediction unit include: Based on the power supply time sequence diagram, historical power supply data is obtained, and current time meteorological data including wind speed data, wind direction data, solar radiation data and temperature data is obtained, and adjustable power generation capacity data including the current operating state, available capacity and rated capacity information of various types of adjustable generator sets is obtained; Based on the wind speed data and the wind direction data, the wind power theoretical value is calculated through the wind turbine power curve, and based on the solar radiation data and the temperature data, the photovoltaic power theoretical value is calculated through the photovoltaic power generation physical model, and the adjustment capacity correction coefficient is calculated by combining the available capacity and the rated capacity ratio of the adjustable generator set; Based on the historical power supply data, the historical average power generation is calculated, and based on the historical average power generation and the theoretical power generation, the meteorological correction coefficient is calculated; Actual power generation data is obtained, and the trend correction coefficient is calculated based on the deviation of the actual power generation data and the historical power generation; The power supply prediction value is calculated based on the historical power supply data, the adjustment capacity correction coefficient, the meteorological correction coefficient and the trend correction coefficient.

5. The system of claim 3, wherein the at least one processor is further configured to: The specific steps of the computing power demand prediction unit include: Based on the computing power demand time sequence diagram, historical computing power demand data is obtained, and a comprehensive load mode sample library is constructed by combining computing power node power consumption data and task execution duration data, the load mode sample library including daily load change modes classified based on power consumption characteristics, weekly load change modes classified based on execution duration characteristics, and monthly load change modes classified based on comprehensive resource demand characteristics; The computing power demand data within a preset time window before the current time is obtained, the computing power demand change curve is calculated by combining the current computing power node power consumption data and the task execution duration data, and the current computing power demand change mode is generated; The similarity of the current computing power demand change mode and each load change mode in the comprehensive load mode sample library is calculated and sorted, the historical computing power demand value corresponding to the highest similarity is determined, and the mode correction coefficient is calculated based on the historical computing power demand value and the computing power node power consumption data; Based on the task resource demand data, the resource demand distribution characteristics of the current task are analyzed, the influence of resource consumption of cross-node task execution is evaluated by combining the network transmission delay data, and the resource correction coefficient is calculated; The number of computing power tasks submitted at the current time is obtained, and the business correction coefficient is calculated based on the deviation of the current task quantity and the historical average task quantity; The computing power demand prediction value is generated based on the historical computing power demand data, the mode correction coefficient, the resource correction coefficient and the business correction coefficient.

6. An electronic co-processing system as in claim 5, wherein: The calculation formula of the power supply prediction value is: U u，p (t+Δt) = U u，h (t+Δt) · f weather (Δt) · f trend (t) · f ab (Δt); wherein, U u，p (t+Δt) represents the predicted value of power supply at time t+Δt, t represents the current time, Δt represents the prediction time step, U u，h (t+Δt) represents the historical power supply value at time t+Δt, f weather (Δt) represents the weather correction coefficient, f trend (t) represents the trend correction coefficient, P f (Δt) represents the theoretical power generation at time Δt, P h (Δt) represents the historical average power generation at time Δt, β represents the trend sensitivity coefficient, i represents the day i days before the current time, U u (t-i) represents the actual power generation data at time t-i, U u，h (t-i) represents the historical power generation at time t-i, represents the long-term average of the historical power generation, f ab (Δt) represents the adjustment capacity correction coefficient at time Δt, U a (Δt) represents the available capacity of the adjustable generator set at time Δt, U r (Δt) represents the rated capacity of the adjustable generator set at time Δt, α represents the adjustment capacity weight coefficient, n represents the length of the historical time window of trend analysis.

7. An electronic co-processing system as in claim 6, wherein: The calculation formula of the computing power demand prediction value is: P i，p (t+Δt) = P i，h (t+Δt) · g pattern (Δt) · g business (t) · g r (t) ; wherein P i,p (t+Δt) represents the predicted value of the computing power demand at time t+Δt, P i，h (t+Δt) represents the historical computing power demand value at time t+Δt, g pattern (Δt) represents the mode correction coefficient, g business (t) represents the service correction coefficient, P s (Δt) represents the historical computing power demand value corresponding to the most similar mode at time Δt, P a (Δt) represents the historical average computing power demand value at time Δt, γ represents the service sensitivity coefficient, N task (t) represents the actual number of task submissions at time t, represents the historical average number of task submissions, g r (t) represents the resource correction coefficient.

8. An electronic co-processing system as in claim 7, wherein: The intelligent matching module includes three matching rules, namely a first matching rule, a second matching rule and a third matching rule, When the power supply prediction value is greater than the computing power demand prediction value, and the supply-demand difference value at the current time is greater than the supply-demand safety threshold, the adjustable capacity range of the generator set at the current time is evaluated in combination with the adjustable power generation capacity data, the first matching rule is used, the supply surplus is calculated by accumulating the supply-demand difference value and the power generation adjustment capacity within the prediction time step, the cost-benefit analysis is performed in combination with the computing power node power consumption data and the power cost coefficient at the current time, and the positive scheduling decision scheme is generated based on the supply surplus and the cost-benefit analysis result; When the power supply prediction value is less than the computing power demand prediction value, and the absolute value of the supply-demand difference value at the current time is greater than the supply-demand gap threshold, the second matching rule is used, the time at which the power supply meets the computing power demand is determined by calculating forward by time, the duration of the power supply gap is determined, the time cost of task migration is calculated in combination with the task execution time length data and the network transmission delay data, and the conservative scheduling decision scheme is generated based on the duration of the power supply gap and the task migration cost. When the absolute value of the supply-demand difference value at the current time is between the supply-demand safety threshold and the supply-demand gap threshold, the third matching rule is used, the historical supply-demand prediction accuracy is counted, the comprehensive cost analysis is performed in combination with the task resource demand data and the computing power service price data, and the balanced scheduling decision scheme is generated based on the historical supply-demand prediction accuracy and the comprehensive cost analysis result.

9. An electronic co-processing system as in claim 8, wherein: The specific steps of the task scheduling module include: acquiring computing power tasks and dividing the computing power tasks into non-migratable tasks, time-migratable tasks, and space-migratable tasks; When the matching decision scheme is the positive scheduling decision scheme, the power consumption information of each task in the current period is acquired, the non-migratable tasks are sorted in descending order of power consumption and high-power-consumption tasks are preferentially scheduled, and the execution time of the time-migratable tasks is advanced to the current power sufficient period; When the matching decision scheme is the conservative scheduling decision scheme, the normal execution of the non-migratable tasks is maintained, the execution of the time-migratable tasks is delayed until after the end of the power supply gap, and the space-migratable tasks are transferred to other computing power nodes with sufficient power supply for execution; When the matching decision scheme is the balanced scheduling decision scheme, the historical supply-demand prediction accuracy is acquired, when the prediction accuracy is higher than a set threshold, the positive scheduling decision scheme is executed, and when the prediction accuracy is lower than the set threshold, the scheduling proportion of high-power-consumption tasks is reduced and the time interval of task execution is increased.

10. A method of computing synergies, characterized by, The application is applied to the computing power collaborative system of any one of claims 1-9, and includes the following steps: S1, collecting power data, computing power data, and price data, and constructing power supply time series graphs and computing power demand time series graphs based on the power data and the computing power data, respectively; S2, based on the power supply time series graph and the computing power demand time series graph, in combination with historical data statistical analysis and real-time weather data correction, generating a power supply prediction value and a computing power demand prediction value; S3, based on the power supply prediction value and the computing power demand prediction value, in combination with time-of-use and regional power price data, through supply-demand balance analysis, cost-benefit analysis, and matching threshold judgment, generating a matching decision scheme of computing power tasks and power resources; S4, acquire computing power tasks and perform task classification, different types of computing power tasks are differentially scheduled based on the matching decision scheme, computing power data and price data, and a task scheduling result is generated; S5, evaluate the task scheduling result and generate an evaluation result, and dynamically adjust the power supply prediction value and the computing power demand prediction value and the matching threshold of the intelligent matching module according to the evaluation result.

Images