A multi-energy optimization management and control method based on energy dynamic prediction

Abstract

The application belongs to the technical field of energy management and computing resource collaborative scheduling, and specifically discloses a multi-energy optimization management and control method based on energy dynamic prediction, which comprises the following steps: firstly, collecting renewable energy in the west and computing power demand data in the east and the west to perform space-time alignment; then, performing short-term power generation prediction and total computing power demand prediction in parallel; subsequently, based on the prediction results, performing collaborative optimization decision under the conditions of computing power task time delay constraint and west green power supply, generating a distribution scheme of computing power tasks among data centers in the east and the west, then executing the scheme and calculating the prediction error, and finally triggering and executing targeted regulation operation according to the deviation set; the application filters migratable tasks according to the time delay constraint, and optimally distributes the tasks under the constraint of green power, so that the migratable tasks are executed by using the west green power to the maximum extent under the premise of guaranteeing the time delay requirement of the computing power tasks, and the proportion of green energy consumption is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of energy management and computing resource collaborative scheduling technology, and relates to a multi-energy optimization and control method based on dynamic energy prediction. Background Technology

[0002] With the development of the digital economy, the surge in computing power demand in eastern my country has led to a sharp increase in energy consumption and carbon emissions, while the western region, rich in renewable energy, is facing insufficient absorption. The "East-to-West Computing" project aims to divert computing power demand westward to utilize green electricity. However, the output of renewable energy in the west is highly volatile and intermittent, while the computing power demand in the east also changes dynamically, resulting in a significant spatiotemporal mismatch. Therefore, it is necessary to construct a collaborative scheduling system that can accurately predict the dynamics of both supply and demand and achieve low-carbon optimization under multiple constraints.

[0003] For example, Chinese invention patent CN118798526A discloses an energy dynamic prediction and multi-energy flow coupling optimization method and energy management system. This technical solution first acquires historical energy production and consumption data, operating condition lists, and production plans within an industrial plant. Then, an improved neural network model is used for prediction, and a Gantt chart is used to manage operating condition curves, with adjustments made based on real-time and historical data. Finally, a multi-energy coupling optimization scheduling model is established for decision-making, aiming to reduce energy dissipation and operating costs for steel enterprises.

[0004] The existing technologies mentioned above have the following shortcomings: 1. Existing technologies mainly focus on collecting and optimizing energy data in local areas, and lack the ability to unify and align the energy supply side and computing power demand side that are separated in a wide geographical space. This leads to the problem of mismatch between renewable energy and computing power demand in both time and space in the East Data West Computing project, making it difficult to build a data foundation and decision-making foundation for cross-regional collaborative optimization.

[0005] 2. Current optimization scheduling models are mainly geared towards allocating continuous energy media, without considering the latency requirements of computing tasks. Furthermore, they do not incorporate latency sensitivity classification of tasks, network bandwidth, transmission latency constraints, and energy efficiency differences between different data centers as optimization considerations. Consequently, they cannot achieve fine-grained scheduling and allocation of computing tasks between eastern and western data centers while ensuring the quality of computing services and network performance. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a multi-energy optimization management method based on dynamic energy prediction is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: This invention provides a multi-energy optimization and management method based on dynamic energy prediction, including: S1, synchronously collecting western renewable energy power generation data and eastern and western computing power demand data of the target area, and combining the power grid structure of energy transmission and the network architecture of task migration to perform spatiotemporal correlation alignment processing on the data to generate spatiotemporal aligned data.

[0008] S2. Based on the spatiotemporal aligned data, perform parallel short-term power generation forecasts for each new energy power station and total computing power demand forecasts for each eastern data center.

[0009] S3. Based on the prediction results, taking into account the latency constraints of computing tasks and the supply of green electricity in the west, a collaborative optimization decision is made to generate a dynamic allocation scheme for computing tasks between the eastern and western data centers.

[0010] S4. Execute the dynamic allocation scheme and collect actual operating data simultaneously to calculate the prediction error between the western power generation and the eastern computing power demand, and obtain the prediction deviation set.

[0011] S5. Determine whether to trigger regulation based on the predicted deviation set. If triggered, perform corresponding regulation operations according to the source and severity of the deviation.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention collects data from the east and west at the same time and adds timestamps and spatial location tags to all data in a unified manner. Then, based on timestamps and spatial locations, it performs time-series alignment and spatial association mapping to construct spatiotemporal aligned data, establishes a unified data foundation for cross-regional energy and computing power scheduling, and solves the problem of spatiotemporal mismatch.

[0013] (2) This invention identifies key meteorological factors of each new energy power station and selects meteorologically similar periods from historical data. Then, it calculates the similarity of power generation curves through dynamic time warping algorithm and uses the actual power generation data of the historical period with the highest similarity as the prediction result, which improves the prediction accuracy of intermittent energy output fluctuations and provides reliable input for optimized scheduling.

[0014] (3) This invention obtains the computing power requirements and task types of each computing power task in each eastern data center, summarizes the current requirements and superimposes the increase in computing power requirements based on historical similar scenarios, and generates an accurate prediction of the total computing power requirements of each eastern data center. This provides key and accurate computing power load input for subsequent collaborative optimization decisions under the dual constraints of latency and green electricity, and supports the realization of fine-grained scheduling.

[0015] (4) By obtaining the set deadline of the computing task and combining it with the estimated execution time, the present invention calculates the total processing delay and comprehensively considers the delay constraints of the computing task, thereby realizing a precise and objective assessment of the urgency and migration potential of the task, and providing a direct basis for accurately distinguishing between locally locked tasks and westward-migrating tasks in subsequent scheduling.

[0016] (5) This invention selects westward-migrating tasks based on latency constraints and then optimizes their allocation under green electricity constraints. This allows for the maximum utilization of western green electricity to execute migratable tasks while strictly ensuring the latency requirements of computing power tasks. This effectively increases the proportion of green energy consumption and reduces the overall carbon footprint. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0019] Figure 2 This is a schematic diagram showing the connection steps of the short-term power generation prediction execution process of the present invention.

[0020] Figure 3 This is a schematic diagram showing the connection steps of the dynamic allocation scheme generation method of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, the present invention provides a multi-energy optimization and management method based on dynamic energy prediction. The method includes: S1, synchronously collecting western renewable energy power generation data and eastern and western computing power demand data of the target area, and combining the power grid structure of energy transmission and the network architecture of task migration to perform spatiotemporal correlation alignment processing on the data to generate spatiotemporal aligned data.

[0023] For example, the generation of spatiotemporally aligned data includes: obtaining real-time power generation and meteorological data of each new energy power plant from western renewable energy power generation data, obtaining real-time computing power demand and associated task characteristics of each western data center from western computing power demand data, obtaining real-time computing power demand and associated task characteristics of each eastern data center from eastern computing power demand data, and adding timestamps and spatial location tags to all data.

[0024] Time-series alignment is performed based on timestamps, and spatial association mapping between each new energy power station and its associated data center is established based on the spatial location tags, according to the power grid structure of energy transmission and the network architecture of task migration. The data after time-series alignment and spatial association mapping is used as spatiotemporal aligned data.

[0025] It should be added that the process of establishing the spatial association mapping between each new energy power station and its associated data center includes the following steps: First, based on the power grid structure of energy transmission, combined with the spatial location labels of each new energy power station and all data centers, the data centers that each new energy power station can provide power to in the power grid are determined; through electrical connection path analysis, the feasible transmission path, the corresponding transmission loss, and the capacity constraints of key transmission lines between each new energy power station and each data center are calculated and recorded to form an energy transmission accessibility mapping table that reflects the energy transmission accessibility and constraints.

[0026] Secondly, based on the network architecture of task migration, and combined with the spatial location tags of each computing task and all data centers, network communication path analysis is performed for each computing task to each data center, and its basic transmission latency and available bandwidth are calculated to form a task migration connectivity mapping table that reflects the network connectivity and performance constraints of task migration.

[0027] Finally, the energy transmission reachability mapping table and the task migration connectivity mapping table are associated and integrated. For each pair of computing power tasks and data centers that have both energy transmission reachability and task migration connectivity, their corresponding energy transmission constraint parameters and network transmission constraint parameters are merged to generate a unified spatial association mapping record. All such records together constitute the spatial association mapping, which is used to characterize the physical basis and constraints of the collaborative scheduling of energy flow and computing power task flow.

[0028] S2. Based on the spatiotemporal aligned data, perform parallel short-term power generation forecasts for each new energy power station and total computing power demand forecasts for each eastern data center.

[0029] Please see Figure 2As shown, for example, the short-term power generation forecast of each new energy power station includes: S2-1, identifying key meteorological factors that affect the power generation of each new energy power station.

[0030] In one specific embodiment, the key meteorological factors are determined according to the type of new energy power plant, including but not limited to: for photovoltaic power plants, the key meteorological factors may include irradiance, ambient temperature, cloud cover, etc. Among these, irradiance directly determines the energy input for photoelectric conversion; ambient temperature affects the conversion efficiency of photovoltaic modules; and cloud cover causes dynamic fluctuations in solar irradiance.

[0031] For wind power plants, the key meteorological factors may include wind speed, wind direction, and air density. Among these, wind speed is the core variable that determines wind power output; wind direction affects the wind turbine's angle to the wind and its output; and air density is related to altitude, temperature, and humidity, thus affecting wind energy density.

[0032] S2-2. Obtain real-time data of each key meteorological factor from the spatiotemporal alignment data, and select historical periods with similar meteorological factors to the current values ​​from the historical data to form a meteorological similarity historical dataset for each new energy power station.

[0033] It should be added that the meteorological similarity historical datasets of each new energy power station are formed through the following steps: for each new energy power station in the target area, high temporal resolution historical power generation data and synchronously recorded meteorological observation data are obtained, and an independent historical database for each new energy power station is constructed.

[0034] The key meteorological factors that significantly affect power generation are identified and extracted from historical meteorological data of various new energy power plants. For photovoltaic power plants, this set includes at least: total irradiance, direct irradiance, ambient temperature, and module backsheet temperature; for wind farms, this set includes at least: wind speed, wind direction, and air density.

[0035] From the spatiotemporal aligned data, the real-time observation values ​​of key meteorological factors corresponding to each new energy power station at the current moment are extracted, and real-time meteorological feature vectors are constructed for each new energy power station based on the observation values.

[0036] From the historical database of each new energy power station, the historical time period of the same period is extracted according to the preset time window, and the observation values ​​of key meteorological factors corresponding to each historical time period are extracted. Based on the observation values, historical meteorological feature vectors are constructed for each historical time period.

[0037] The cosine similarity algorithm is used to calculate the similarity value between the real-time meteorological feature vector of each new energy power station and its historical meteorological feature vector. For each new energy power station, the top K historical periods with the highest similarity values ​​are selected as similar historical periods, where K is a positive integer that is pre-set based on historical data verification. At the same time, the complete meteorological data sequence and the actual power generation data sequence corresponding to each similar historical period are obtained.

[0038] The complete time-series data of all key meteorological factors and the corresponding actual power generation time-series data for each similar historical period selected from each new energy power plant are structured and stored to ultimately generate a meteorological similarity historical dataset for each new energy power plant.

[0039] S2-3. For each historical period in the meteorological similarity historical dataset, calculate the similarity between each historical period and the power generation curve of the corresponding new energy power station.

[0040] Furthermore, the calculation of the similarity between the power generation curves of each historical period and the corresponding new energy power station includes: S2-3-1, constructing the real-time power generation curves of each new energy power station based on the real-time power generation of each new energy power station.

[0041] S2-3-2. Based on the power generation data of each historical period in the meteorological similarity historical dataset, construct the historical power generation curve for each historical period.

[0042] S2-3-3. The dynamic time warping algorithm is used to calculate the similarity between the power generation curves of each historical period and the corresponding new energy power station.

[0043] In one specific embodiment, calculating the similarity between the power generation curves of each historical time period and the corresponding new energy power plant includes: obtaining a real-time power generation curve sequence. Compared with historical power generation curve series ,in and , respectively, are the lengths of the two sequences.

[0044] calculate and The Euclidean distance between each point forms Local distance matrix , where matrix elements ,in, For the first in the real-time sequence Data points, , The first in the historical sequence Data points, .

[0045] Initialize a Cumulative distance matrix And set boundary conditions: .

[0046] For the remaining positions Calculated using the following recursive formula: The process aims to find a way from arrive The optimal path is one that minimizes the sum of the local distances between all points on the path.

[0047] from Begin backtracking to determine the optimal path. This path represents the optimal nonlinear correspondence between the two curves on the time axis. The value of is the minimum cumulative distance between the two curves.

[0048] To facilitate comparison, the minimum cumulative distance is converted into a similarity value. For example, it can be done using... In the formula The similarity value indicates that the two curves are more similar in shape and fluctuation pattern.

[0049] For each historical power generation curve in the meteorological similarity historical dataset, repeat the above steps to calculate the similarity between each historical power generation curve and the real-time curve. Finally, select the historical period with the highest similarity as the optimal similar historical period.

[0050] S2-4. Select the historical period with the highest power curve similarity as the optimal similar historical period, and take the actual power generation curve of the historical period that follows the optimal similar historical period and is of the same length as the prediction period as the predicted power generation of the new energy power station in the corresponding prediction period in the future.

[0051] For example, the prediction of total computing power demand for each eastern data center includes: obtaining the computing power demand and task type of each computing power task in each eastern data center from the spatiotemporal alignment data.

[0052] The computing power requirements of each computing task in each eastern data center are aggregated to obtain the total computing power requirements of each eastern data center.

[0053] Based on historical data, select one or more historical dates that are in the same time period and have the same task type as the current date, and calculate the average increase in computing power demand for the historical dates in a preset time period after the current time period, which will be used as the increase in computing power demand for each eastern data center.

[0054] The total computing power requirement is superimposed with the corresponding increase in computing power requirement to generate a prediction of the total computing power requirement for each eastern data center.

[0055] S3. Based on the prediction results, taking into account the latency constraints of computing tasks and the supply of green electricity in the west, a collaborative optimization decision is made to generate a dynamic allocation scheme for computing tasks between the eastern and western data centers.

[0056] For example, the comprehensive consideration of latency constraints for computing power tasks includes calculating the total processing latency by obtaining the set deadline and estimated execution time reference values ​​for each computing power task in each eastern data center.

[0057] Query the historical task database to obtain the actual execution time records of historical tasks with similar computing power requirements and type identifiers to the computing power task, and calculate the historical average task execution time.

[0058] The estimated execution time of a computing task is determined by taking the median value between the estimated execution time reference value and the historical average task execution time.

[0059] Based on the set deadline of the computing power task and the estimated task execution time, the total processing latency is calculated, wherein the total processing latency is: the time from the current time point to the task deadline point minus the estimated task execution time.

[0060] Please see Figure 3 As shown, exemplarily, the dynamic allocation scheme of the generated computing power tasks between the eastern and western data centers includes: for each computing power task in each eastern data center, if its total processing latency is less than or equal to the basic latency of the eastern and western network transmission, it is determined to be a locally locked task; otherwise, it is determined to be a task that can be migrated westward.

[0061] Specifically, if the total processing latency of a computing task is less than or equal to the baseline network transmission latency between the east and west regions, it indicates that the task's full execution time budget is extremely tight and cannot accommodate the additional network transmission time incurred due to migration to the western data center. Therefore, it must be executed in the local eastern data center to avoid timeout failure due to transmission delays. Conversely, if the total processing latency is greater than the baseline network transmission latency between the east and west regions, it indicates that the task has a certain latency margin, and its time budget is sufficient to cover the network transmission overhead between the east and west regions. Thus, it is technically possible to migrate it to the western data center for execution and is determined to be a westward migration task.

[0062] Based on the real-time computing power requirements of each western data center and their associated task characteristics, the predicted computing power requirements of each western data center are calculated, and combined with the corresponding energy efficiency coefficient, the total baseline energy consumption of the western data centers is calculated.

[0063] It should be noted that the method used for forecasting computing power demand in western data centers is consistent with the method used for forecasting computing power demand in eastern data centers, and the specific steps of the forecasting method will not be repeated here.

[0064] Furthermore, the energy efficiency coefficient is used to characterize the overall energy efficiency of a data center in converting computing resource consumption into electrical energy consumption. Its determination method includes: first, obtaining the total power consumption of the data center over the past complete statistical period from the data center infrastructure management system, denoted as... .

[0065] Simultaneously, the total computing resource consumption provided by all servers within the same statistical period is obtained from the computing resource scheduling system and recorded as follows: The consumption of computing resources can be uniformly quantified according to the resource measurement standards adopted by the data center. For example, standard core hours can be used for general CPU computing, while GPU-accelerated computing can be converted into equivalent CPU core hours or independent GPU hours for measurement.

[0066] Subsequently, the energy efficiency coefficient of the data center was calculated. , .

[0067] The total baseline energy consumption refers to the electrical energy consumed by each data center in the western region to process its local predicted computing power demand. The calculation steps are as follows: For each western data center, obtain its predicted computing power demand. This predicted computing power demand needs to be uniformly quantified as the total amount of computing resources required within a future scheduling cycle, such as core hours.

[0068] The predicted computing power demand of each western data center is multiplied by its corresponding energy efficiency coefficient to obtain the baseline energy consumption of that data center. Then, the baseline energy consumption of all western data centers is summed to obtain the total baseline energy consumption of the western data centers.

[0069] Based on the principle of prioritizing local consumption of green electricity in the west, the baseline energy consumption of data centers in the west is fully supplied by green electricity from the west. Under this premise, the total amount of available green electricity in the west that can be used to handle the relocation of power from the east to the west is obtained by superimposing the predicted power generation of each new energy power station during the scheduling cycle to calculate the total predicted amount of green electricity in the west, and then calculating the difference between this and the total baseline energy consumption to obtain the total amount of available green electricity in the west.

[0070] In one specific embodiment, the calculation process for the total predicted green electricity in the west is as follows: Assume that the west has a total of The first new energy power station, the first The projected power generation of each power station during the future dispatch cycle is: ,in, The total predicted amount of green electricity in the western region The calculation formula is: In the formula The duration of the scheduling cycle. This represents the maximum amount of green electricity that a western renewable energy system can provide to data center loads under ideal conditions.

[0071] Based on the computing power requirements of each westward migration task, the total amount of available green electricity in the west, and the real-time operating status of each western data center, under the dual constraints of green electricity supply and computing power resource capacity, the specific allocation results of westward migration tasks between eastern and western data centers are generated by executing task allocation rules.

[0072] Furthermore, the task allocation rules include: obtaining the real-time remaining computing power capacity of each western data center.

[0073] Based on the total processing latency of each westward migration task, they are sorted from low to high to form a task priority sequence. At this time, the shorter the total processing latency, the higher the ranking and the higher the priority.

[0074] According to the task priority sequence, the currently assigned westward relocation tasks are selected in order. Target data centers are selected from the western data centers that simultaneously meet the following conditions: the real-time remaining computing power capacity of the data center is not less than the computing power requirement of the task, and the remaining available green electricity in the power grid area where the data center is located is not less than the migration energy consumption of the task.

[0075] The method of sequentially filtering and allocating tasks according to their priority sequence is an efficient greedy allocation strategy that prioritizes the westward relocation of tasks with more pressing time delays while satisfying the dual constraints of real-time operation. In other embodiments of the present invention, other optimization algorithms may also be employed, such as constructing an integer programming model under the aforementioned dual constraints to find the globally optimal green power consumption scheme.

[0076] The migration energy consumption of a single westward-migrating task is calculated by multiplying the task's computing power requirement by the energy efficiency coefficient of the target western data center to which the task is to be migrated.

[0077] If the screening is successful, the computing power task will be assigned to the target data center for execution, and the remaining computing power capacity of the data center and the remaining available green electricity in its associated power grid area will be reduced simultaneously.

[0078] If no target meeting the above two conditions can be found among all western data centers, then all current and subsequent unassigned westward migration tasks will be assigned to eastern data centers for execution.

[0079] Combining the above allocation results of locally locked tasks and westward-migrating tasks, a dynamic allocation scheme for computing power tasks between the east and west is formed.

[0080] S4. Execute the dynamic allocation scheme and collect actual operating data simultaneously to calculate the prediction error between the western power generation and the eastern computing power demand, and obtain the prediction deviation set.

[0081] For example, the calculation of the prediction deviation set includes: calculating the error between the actual measured power generation of each new energy power station and the corresponding predicted power generation, and recording the new energy power station whose error exceeds a preset power generation error threshold as a deviation new energy power station.

[0082] It should be added that the preset power generation error threshold is set independently for each new energy power station. The method for obtaining it includes: statistically analyzing the absolute or relative error sequence between the short-term power generation prediction result and the actual value of the power station over a period of time; taking the 95th percentile of the error sequence as the preset error threshold to identify significant deviations that exceed the normal fluctuation range, thereby ensuring that the threshold can cover the vast majority of historical normal fluctuations and only alarms are triggered for significant deviations that exceed the normal range.

[0083] Calculate the error between the actual total computing power demand of each eastern data center and the predicted computing power demand, and record the eastern data centers whose error exceeds the preset computing power demand energy consumption threshold as deviation data centers.

[0084] It should be added that the preset computing power demand energy consumption threshold is set independently for each Eastern data center. The method for obtaining this threshold includes: statistically analyzing the absolute error sequence between the predicted computing power demand and the actual total energy consumption of the data center over a past period; and taking the 95th percentile of this error sequence as the energy consumption error threshold for that data center. This method allows the threshold to adapt to the normal range of data center business fluctuations.

[0085] A prediction deviation set is constructed based on the aforementioned deviation new energy power plants and deviation data centers.

[0086] S5. Determine whether to trigger regulation based on the predicted deviation set. If triggered, perform corresponding regulation operations according to the source and severity of the deviation.

[0087] It should be added that, based on the source and severity of the deviation, the corresponding control operation is performed, which specifically includes the following steps: According to the composition of the predicted deviation set, it is classified into one of the following three types: energy-side prediction deviation, computing power demand-side prediction deviation, or comprehensive prediction deviation; wherein, if the predicted deviation set only contains deviation new energy power plants, it is determined to be an energy-side prediction deviation; if it only contains deviation data centers, it is determined to be a computing power demand-side prediction deviation; if it contains both deviation new energy power plants and deviation data centers, it is determined to be a comprehensive prediction deviation.

[0088] Regarding energy-side prediction bias: For each power station marked as a biased renewable energy power station, its predicted power generation for the next scheduling cycle is corrected to the arithmetic mean of the actual power generation for the current scheduling cycle and the historical average power generation for the same period of the power station. If a renewable energy power station is identified as a biased renewable energy power station for N consecutive scheduling cycles, an equipment status verification warning is generated and sent to the operation and maintenance system. In subsequent scheduling cycles, the predicted power generation of the power station is suspended from participating in the generation of the task allocation scheme. Here, N is a positive integer preset according to system stability requirements, such as 3.

[0089] To address the bias in computing demand prediction: For westward-migrating tasks originating from various biased data centers, when generating the task priority sequence for the next scheduling cycle, the total processing latency of the tasks is uniformly increased by a preset penalty latency value to reduce their priority order for westward allocation; at the same time, the actual computing demand data of the biased data center in the current scheduling cycle is used to replace the oldest data point in the historical demand sequence used for historical trend analysis, so as to update the benchmark for computing demand prediction.

[0090] In response to comprehensive prediction deviations: immediately suspend the current prediction-based optimization allocation process and switch to a preset backup allocation strategy; the backup allocation strategy allocates westward relocation tasks to the western data center according to a preset fixed ratio, such as 50%; simultaneously, clear the historical meteorological similarity dataset and power curve similarity calculation records on which short-term power generation prediction depends, and reset the historical average task execution time and computing power demand increase calculation benchmark on which total computing power demand prediction depends; the cleared and reset data and parameters must be verified in at least one complete offline scheduling simulation cycle before they can be used for subsequent online prediction and scheduling.

[0091] After any of the above-mentioned control operations are completed, the system enters a control cooling period lasting M scheduling cycles. During the control cooling period, the system suspends the triggering judgment of deviation events from the same source, maintains only the monitoring state, and continues to execute the scheduling strategies that took effect before the cooling period to prevent frequent control oscillations caused by instantaneous disturbances. At the same time, the system monitors the actual processing progress of each computing task and the real-time power balance of each power grid area in real time. If the actual completion delay of any task exceeds its set deadline, or if any power grid area experiences an actual power deficit, the control cooling period is immediately forcibly interrupted, and the system switches to a conservative scheduling mode based on the real-time operating status. The current state is then remarked as a state requiring control evaluation. Here, M is a preset positive integer, such as 2.

[0092] For example, determining whether to trigger regulation includes: obtaining the number of deviation new energy power plants and the number of deviation data centers from the prediction deviation set.

[0093] If the number of deviation new energy power stations is greater than a preset threshold for the number of power stations or the number of deviation data centers is greater than a preset threshold for the number of data centers, then regulation is triggered; otherwise, regulation is not triggered.

[0094] It should be added that the preset threshold for the number of power plants and the threshold for the number of data centers are obtained by retrieving the system operation records of the past year and filtering out all abnormal events that triggered manual control or caused key performance indicators.

[0095] For each abnormal event, the minimum value of the number of new energy power plants and the number of data centers in a deviated state is selected from all abnormal events as the preset threshold for the number of power plants and the threshold for the number of data centers, respectively.

[0096] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0097] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0098] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0099] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0100] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-energy optimization management method based on dynamic energy prediction, characterized in that: The method includes: S1. Synchronously collect renewable energy power generation data in the western part of the target area and computing power demand data in the eastern and western parts, and combine the power grid structure of energy transmission and the network architecture of task migration to perform spatiotemporal correlation and alignment processing on the data to generate spatiotemporal aligned data. S2. Based on the spatiotemporal aligned data, perform parallel short-term power generation prediction for each new energy power station and total computing power demand prediction for each eastern data center. S3. Based on the prediction results, taking into account the latency constraints of computing tasks and the supply of green electricity in the west, a collaborative optimization decision is made to generate a dynamic allocation scheme for computing tasks between the eastern and western data centers. S4. Execute the dynamic allocation scheme and collect actual operating data simultaneously to calculate the prediction error between the western power generation and the eastern computing power demand, and obtain the prediction deviation set. S5. Determine whether to trigger regulation based on the predicted deviation set. If triggered, perform corresponding regulation operations according to the source and severity of the deviation.

2. The multi-energy optimization and management method based on dynamic energy prediction according to claim 1, characterized in that: The generated spatiotemporally aligned data includes: The system obtains real-time power generation and meteorological data of each new energy power station from the western renewable energy power generation data, real-time computing power demand and related task characteristics of each western data center from the western computing power demand data, and real-time computing power demand and related task characteristics of each eastern data center from the eastern computing power demand data, and adds timestamps and spatial location tags to all data. Time-series alignment is performed based on timestamps, and spatial association mapping between each new energy power station and its associated data center is established based on the spatial location tags, according to the power grid structure of energy transmission and the network architecture of task migration. The data after time-series alignment and spatial association mapping is used as spatiotemporal aligned data.

3. The multi-energy optimization management method based on dynamic energy prediction according to claim 1, characterized in that: The execution of short-term power generation forecasts for each new energy power plant includes: For each new energy power plant, identify the key meteorological factors that affect its power generation capacity; Real-time data of each key meteorological factor is obtained from the spatiotemporal aligned data, and historical periods with similar meteorological factors to the current values ​​are selected from the historical data to form a meteorological similarity historical dataset for each new energy power station. For each historical period in the meteorological similarity historical dataset, calculate the similarity between each historical period and the power generation curve of the corresponding new energy power station; The historical period with the highest power curve similarity is selected as the optimal similar historical period, and the actual power generation curve of the historical period immediately following the optimal similar historical period, which is of the same length as the prediction period, is taken as the predicted power generation of the new energy power station in the corresponding prediction period in the future.

4. The multi-energy optimization and management method based on dynamic energy prediction according to claim 3, characterized in that: The calculation of the similarity between the power generation curves of each historical period and the corresponding new energy power station includes: Based on the real-time power generation of each new energy power station, construct the real-time power generation curve of each new energy power station. Based on the power generation data of each historical period in the meteorological similarity historical dataset, historical power generation curves for each historical period are constructed respectively; The similarity between the power generation curves of each historical period and the corresponding new energy power station is calculated using a dynamic time warping algorithm.

5. The multi-energy optimization and management method based on dynamic energy prediction according to claim 1, characterized in that: The prediction of total computing power demand for each eastern data center includes: The computing power requirements and task types of each computing power task in each eastern data center are obtained from the spatiotemporal alignment data. The computing power requirements of each computing task in each eastern data center are summarized to obtain the total computing power requirements of each eastern data center. Based on historical data, select one or more historical dates that are in the same time period and have the same task type as the current time, and calculate the average increase in computing power demand for the historical dates in a preset time period after the current time period, which will be used as the increase in computing power demand for each eastern data center. The total computing power requirement is superimposed with the corresponding increase in computing power requirement to generate a prediction of the total computing power requirement for each eastern data center.

6. The multi-energy optimization and management method based on dynamic energy prediction according to claim 1, characterized in that: The latency constraints for comprehensively considering computing power tasks include calculating the total processing latency in the following ways: For each computing task in each eastern data center, obtain the set deadline and estimated execution time reference values; Query the historical task database to obtain the actual execution time records of historical tasks with similar computing power requirements and type identifiers to the computing power task, and calculate the historical average task execution time. The median value between the estimated execution time reference value and the historical average task execution time is taken as the estimated task execution time of the computing power task; The total processing latency is calculated based on the set deadline of the computing task and the estimated task execution time.

7. The multi-energy optimization management method based on dynamic energy prediction according to claim 1, characterized in that: The dynamic allocation scheme for generating computing power tasks between the eastern and western data centers includes: For each computing task in the eastern data center, if the total processing latency is less than or equal to the basic transmission latency of the east-west network, it is determined to be a locally locked task; otherwise, it is determined to be a task that can be migrated westward. Based on the real-time computing power requirements of each western data center and their associated task characteristics, the predicted computing power requirements of each western data center are calculated, and combined with the corresponding energy efficiency coefficient, the total baseline energy consumption of the western data center is calculated. The predicted power generation of each new energy power station during the dispatch cycle is superimposed to calculate the total predicted green power in the west. The difference between this and the total benchmark energy consumption is calculated to obtain the total available green power in the west. Based on the computing power requirements of each westward migration task, the total amount of available green electricity in the west, and the real-time operating status of each western data center, under the dual constraints of green electricity supply and computing power resource carrying capacity, the specific allocation results of westward migration tasks between the eastern and western data centers are generated by executing task allocation rules. Combining the above allocation results of locally locked tasks and westward-migrating tasks, a dynamic allocation scheme for computing power tasks between the east and west is formed.

8. The multi-energy optimization management method based on dynamic energy prediction according to claim 7, characterized in that: The task allocation rules include: Obtain the real-time remaining computing power capacity of each western data center; The tasks are sorted from low to high based on their total processing latency to form a task priority sequence. According to the task priority sequence, the currently assigned westward relocation tasks are selected in order. Target data centers are selected from the western data centers that simultaneously meet the following conditions: the real-time remaining computing power capacity of the data center is not less than the computing power requirement of the task, and the remaining available green electricity in the power grid area where the data center is located is not less than the migration energy consumption of the task. If the screening is successful, the computing power task will be assigned to the target data center for execution, and the remaining computing power capacity of the data center and the remaining available green electricity in its associated power grid area will be reduced simultaneously. If no target meeting the above two conditions can be found among all western data centers, then all current and subsequent unassigned westward migration tasks will be assigned to eastern data centers for execution.

9. The multi-energy optimization management method based on dynamic energy prediction according to claim 1, characterized in that: The calculation of the prediction deviation set includes: Calculate the error between the actual measured power generation of each new energy power station and the corresponding predicted power generation, and record the new energy power station whose error exceeds the preset power generation error threshold as a deviation new energy power station; Calculate the error between the actual total computing power demand of each eastern data center and the predicted computing power demand, and record the eastern data centers whose error exceeds the preset computing power demand energy consumption threshold as deviation data centers; A prediction deviation set is constructed based on the aforementioned deviation new energy power plants and deviation data centers.

10. The multi-energy optimization management method based on dynamic energy prediction according to claim 1, characterized in that: The determination of whether regulation is triggered includes: The number of deviation new energy power plants and the number of deviation data centers are obtained from the predicted deviation set; If the number of deviation new energy power stations is greater than a preset threshold for the number of power stations or the number of deviation data centers is greater than a preset threshold for the number of data centers, then regulation is triggered; otherwise, regulation is not triggered.

Images