A synergistically optimized data center energy scheduling method and system

Abstract

This invention discloses a collaborative optimization method and system for data center energy scheduling, applied in the field of energy scheduling and intelligent optimization technology. The method includes: extracting features from historical renewable energy power generation data to obtain renewable energy power generation feature vectors for the data center; performing cluster analysis on these feature vectors to obtain typical renewable energy power generation scenarios for the data center; determining the probability distribution of these scenarios based on their probability distribution; determining the load status of each physical device based on various operating status data; determining the charging and discharging power of each physical device; constructing a renewable energy scheduling optimization model with the objective function of minimizing both the data center's operating cost and carbon emission cost; solving the renewable energy scheduling optimization model based on constraints and the probability distribution of power generation scenarios; and obtaining and executing a renewable energy scheduling scheme. This invention achieves safe and stable power supply to the data center.

Description

Technical Field

[0001] This invention relates to the field of energy dispatching and intelligent optimization technology, and in particular to a collaborative optimization method and system for data center energy dispatching. Background Technology

[0002] Data centers are adopting renewable energy sources for power in pursuit of a green transformation, but direct grid connection can easily lead to power instability and resource waste. Therefore, a scientific scheduling plan is necessary to integrate and regulate various resources within the data center to ensure power supply security.

[0003] In existing technologies, renewable energy dispatching methods rely on data center dispatchers to develop dispatching plans based on their personal experience. However, human experience makes it difficult to accurately predict the power demand of physical equipment and the fluctuations in renewable energy availability. Plans based on experience can affect the power supply stability of the data center and make it difficult to guarantee the power supply security of the data center. Summary of the Invention

[0004] This invention provides a collaborative optimization method and system for data center energy scheduling to solve the technical problem of unstable power supply in data centers caused by manual experience-based scheduling, thereby achieving the effect of safe and stable power supply to data centers.

[0005] To address the aforementioned technical problems, this invention provides a collaboratively optimized data center energy scheduling method, comprising: Acquire historical renewable energy generation data and operational status data of selected physical devices in the data center; Feature extraction is performed on the historical renewable energy power generation data, and the feature extraction results are vectorized to obtain the renewable energy power generation feature vector of the data center; Cluster analysis is performed on the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios for the data center, and the probability distribution results of the power generation scenarios are determined based on the typical renewable energy power generation scenarios. Based on the various operational status data, determine the load status of each physical device; Analyze each load condition to obtain the corresponding charging and discharging power of each physical device. A renewable energy scheduling optimization model is constructed with the goal of minimizing the combined operating cost and carbon emission cost of the data center. The constraints of the renewable energy scheduling optimization model are determined based on at least the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device. The renewable energy scheduling optimization model is solved based on the constraints and the probability distribution results of the power generation scenario to obtain and execute a renewable energy scheduling scheme.

[0006] As one preferred embodiment, the step of extracting features from the historical renewable energy power generation data and performing vector transformation on the feature extraction results to obtain the renewable energy power generation feature vector of the data center includes: The historical renewable energy power generation data is cleaned to obtain cleaned renewable energy power generation data; The cleaned renewable energy power generation data is normalized to obtain normalized renewable energy power generation data; Extract the power generation characteristics from the normalized renewable energy power generation data, and perform the vector transformation on the power generation characteristics to obtain the renewable energy power generation characteristic vector of the data center.

[0007] As one preferred embodiment, the clustering analysis of the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios for the data center includes: Based on the elbow method, the feature vector of the renewable energy power generation is analyzed to obtain the number of clusters; Based on the number of clusters and the improved K-means clustering algorithm, the renewable energy power generation feature vectors are clustered to obtain the clustering results of each renewable energy power generation feature vector; The clustering results are processed to extract power generation scenarios, resulting in the typical renewable energy power generation scenarios.

[0008] As one preferred embodiment, determining the probability distribution result of the power generation scenario based on the typical renewable energy power generation scenario includes: Based on the Wasserstein distance, an initial probability statistical analysis was performed on the typical renewable energy power generation scenario to obtain the initial probability results of the typical renewable energy power generation scenario; Based on the Laplace algorithm, the initial probability results are corrected to obtain the probability distribution results of the power generation scenario.

[0009] As one preferred embodiment, the step of analyzing each load condition to obtain the corresponding charging and discharging power of each physical device includes: Based on the load status of each physical device, determine the corresponding power load requirement of that physical device; A power balance analysis is performed on the electrical load demand of each physical device to obtain the corresponding charging and discharging power of the physical device.

[0010] Another aspect of the present invention provides a collaboratively optimized data center energy scheduling system, comprising: The data acquisition module is used to acquire historical renewable energy power generation data and operational status data of selected physical devices in the data center; The feature vector conversion module is used to extract features from the historical renewable energy power generation data and convert the feature extraction results into vectors to obtain the renewable energy power generation feature vector of the data center. The clustering analysis module is used to perform clustering analysis on the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios of the data center, and determine the probability distribution results of the power generation scenarios based on the typical renewable energy power generation scenarios; The load analysis module is used to determine the load status of each physical device based on the various operating status data. The power analysis module is used to analyze each load condition to obtain the charging and discharging power of each physical device. The scheduling scheme generation module is used to construct a renewable energy scheduling optimization model with the objective function of minimizing the combined operating cost and carbon emission cost of the data center. It determines the constraints of the renewable energy scheduling optimization model based at least on the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device. It solves the renewable energy scheduling optimization model based on the constraints and the probability distribution results of the power generation scenario to obtain and execute the renewable energy scheduling scheme.

[0011] As one preferred embodiment, the feature vector transformation module includes: The data cleaning unit is used to clean the historical renewable energy power generation data to obtain cleaned renewable energy power generation data. The normalization processing unit is used to normalize the cleaned renewable energy power generation data to obtain normalized renewable energy power generation data. The vector transformation unit is used to extract the power generation characteristics of the normalized renewable energy power generation data, and perform the vector transformation on the power generation characteristics to obtain the renewable energy power generation characteristic vector of the data center.

[0012] As one preferred embodiment, the clustering analysis module includes: The elbow method analysis unit is used to analyze the feature vector of the renewable energy power generation based on the elbow method to obtain the number of clusters; The clustering analysis unit is used to cluster the renewable energy power generation feature vectors based on the number of clusters and the improved K-means clustering algorithm, and obtain the clustering results of each renewable energy power generation feature vector; The scene extraction unit is used to extract power generation scenarios from all the clustering results to obtain the typical renewable energy power generation scenarios.

[0013] As one preferred embodiment, the clustering analysis module includes: An initial probability statistics unit is used to perform initial probability statistical analysis on the typical renewable energy power generation scenario based on the Wasserstein distance, and obtain the initial probability results of the typical renewable energy power generation scenario. The probability correction unit is used to correct the initial probability result based on the Laplace algorithm to obtain the probability distribution result of the power generation scenario.

[0014] As one preferred embodiment, the power analysis module includes: The load demand analysis unit is used to determine the power load demand of each physical device based on its load status. The power balance analysis unit is used to perform power balance analysis on the power load demand of each physical device to obtain the charging and discharging power of the corresponding physical device.

[0015] Compared to existing technologies, the beneficial effects of this invention are at least one of the following: This invention, through feature extraction and vector transformation of historical renewable energy power generation data, automatically extracts key fluctuation patterns from time-series power generation data and transforms them into standardized mathematical vectors, providing a structured data foundation for subsequent intelligent analysis; This invention, through cluster analysis of power generation feature vectors and determination of the probability distribution results of power generation scenarios, abstracts and quantifies the continuous and complex uncertainty of new energy output into a probability model composed of a finite number of typical scenarios and their probabilities of occurrence, solving the limitation that human experience cannot scientifically describe and quantify fluctuation risks; This invention, by using the probability distribution results of power generation scenarios as key inputs to solve the optimization model, achieves deep integration of the scheduling decision-making process and the probability model of new energy uncertainty, enabling the final generated scheduling scheme to proactively consider the impact of various possible power generation scenarios, thereby significantly improving the adaptability and reliability of the scheme in actual fluctuation environments, and solving the vulnerability problem of performance drop when deterministic optimization methods face prediction deviations. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a collaboratively optimized data center energy scheduling method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a collaboratively optimized data center energy scheduling system in one embodiment of the present invention; Figure label: The module includes: 11. Data acquisition module; 12. Feature vector conversion module; 13. Cluster analysis module; 14. Load analysis module; 15. Power analysis module; and 16. Scheduling scheme generation module. Detailed Implementation

[0017] 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. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] One embodiment of the present invention provides a collaborative optimization method for data center energy scheduling. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown illustrates a flowchart of a collaboratively optimized data center energy scheduling method according to one embodiment of the present invention. The method includes steps S1 to S6: S1. Obtain historical renewable energy power generation data and operational status data of each selected physical device in the data center; S2. Extract features from the historical renewable energy power generation data, and perform vector transformation on the feature extraction results to obtain the renewable energy power generation feature vector of the data center; S3. Perform cluster analysis on the renewable energy power generation feature vector to obtain typical renewable energy power generation scenarios of the data center, and determine the probability distribution results of the power generation scenarios based on the typical renewable energy power generation scenarios; S4. Determine the load status of each physical device based on the various operating status data. S5. Analyze each load condition to obtain the charging and discharging power of each physical device. S6. With the goal of minimizing the combined operating cost and carbon emission cost of the data center, construct a renewable energy scheduling optimization model. Determine the constraints of the renewable energy scheduling optimization model based at least on the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device. Solve the renewable energy scheduling optimization model based on the constraints and the probability distribution results of the power generation scenario to obtain and execute the renewable energy scheduling scheme.

[0022] As a core infrastructure for the development of the digital economy, data centers are experiencing ever-increasing computing power demands and escalating energy consumption. To reduce operating costs, renewable energy sources such as wind and solar power are being introduced into data centers for power supply upgrades. However, wind and solar power generation is significantly affected by weather conditions and seasonal variations, exhibiting marked randomness, volatility, and intermittency. Directly connecting such renewable energy sources to the grid to power data centers can easily lead to large fluctuations in power supply, resulting not only in low renewable energy absorption rates and severe waste of energy resources, but also impacting the stability of the data center's power system. This can even cause operational failures in core hardware such as servers and energy storage devices due to voltage or power instability, directly affecting the continuous and reliable operation of data center services. Therefore, a scientific and intelligent scheduling scheme is essential to integrate and dynamically regulate various resources within the data center, including computing load, energy storage devices, and power procurement, in order to improve the efficiency of renewable energy utilization while ensuring the power supply security and business stability of the data center.

[0023] Furthermore, in step S1, to solve the problem that manual scheduling cannot accurately match the renewable energy power generation with the equipment power consumption demand and ensure the scientificity and rationality of the scheduling plan from the data source. Specifically, it is achieved by using the data acquisition module supporting the data center. Preferably, for historical renewable energy power generation data, the time-series data of the full-cycle power generation of renewable energy such as wind power and photovoltaic power accessed by the data center is collected. At the same time, the real-time operation status of the core physical equipment participating in energy scheduling within the data center is collected and historical data is summarized. During the collection process, the time alignment of the data is completed synchronously to ensure the integrity and consistency of the data, laying a foundation for subsequent data processing. The core benefit of doing this is that on the one hand, it can fully master the output law and fluctuation characteristics of renewable energy, providing a data basis for the construction of subsequent typical power generation scenarios. On the other hand, it can accurately obtain the operation status of physical equipment, clarify the power consumption load basis and regulation ability of the equipment, realize the collaborative collection of energy-side data and load-side data, enable the subsequent scheduling optimization model to be constructed based on actual data, greatly improving the adaptability and feasibility of the scheduling plan. At the same time, it also provides accurate equipment parameters and operation basis for subsequent power balance analysis and charge-discharge power calculation.

[0024] Among them, the selected physical equipment refers to the core hardware equipment that constitutes the energy supply, computing power bearing, and energy storage system of the data center and directly participates in renewable energy scheduling and energy resource collaborative optimization. It is the key carrier for the data center to achieve power consumption, energy regulation, and business operation. Its operation status directly determines the power consumption load demand and energy regulation ability of the data center. The overall energy scheduling demand of the data center is realized by regulating the operation status of these selected physical equipment; the selected physical equipment in the present invention specifically includes three categories. The first category is the equipment supporting renewable energy power generation, that is, the power generation equipment such as wind power generators, photovoltaic inverters, and photovoltaic panel groups accessed by the data center, which are used to convert wind energy and light energy into electrical energy and are the source carriers of renewable energy power generation data. The second category is energy storage equipment, including electrochemical energy storage batteries, energy storage converters, etc., which have the ability to store and release electrical energy and are the core equipment for realizing energy regulation in the data center and suppressing the output fluctuations of renewable energy. The third category is the core computing power equipment, that is, the server clusters, switches, etc. in the data center that undertake computing power services and are the main consumption carriers of the power consumption load in the data center. In addition, it also includes key equipment such as the refrigeration supporting control equipment in the data center that directly affects the power consumption load.

[0025] Furthermore, in step S2, to transform the unstructured, time-series raw data of renewable energy power generation into standardized, structured mathematical feature vectors that can be processed by intelligent algorithms such as cluster analysis, and to extract core information such as the fluctuation patterns, time-specific characteristics, and amplitude variation characteristics of wind and solar power output contained in the data, the following steps are taken: data cleaning, normalization, feature extraction, and vector transformation. First, outlier removal is performed on the historical renewable energy power generation data to eliminate data noise and invalid information. Then, the cleaned data is normalized to map power generation data with different dimensions and amplitudes to a unified numerical range, avoiding the impact of data scale differences on feature extraction results. Next, core features reflecting the characteristics of renewable energy power generation are extracted from the normalized data. Finally, principal component analysis (PCA) is performed on the extracted feature data to obtain the renewable energy power generation feature vector. The advantages of this approach are threefold: First, data cleaning and normalization ensure the accuracy and effectiveness of feature extraction, eliminating the influence of interfering data on subsequent analysis. Second, feature extraction achieves dimensionality reduction and information condensation of the original time-series data, preserving the key patterns of power generation output while discarding redundant information. Third, vector transformation converts abstract power generation characteristics into quantified mathematical vectors, making them compatible with the operational logic of the subsequently improved K-means clustering algorithm. This allows for direct use in cluster analysis, significantly improving the efficiency and accuracy of subsequent typical scenario construction, while also providing a standardized data carrier for calculating the probability distribution of power generation scenarios.

[0026] Furthermore, in step S3, in order to classify and integrate the high-dimensional, discrete renewable energy power generation feature vectors, extract representative typical power generation scenarios, and quantify the occurrence probability of each scenario, the complex and continuous uncertainty of new energy output is transformed into a finite and quantifiable probabilistic scenario model. This solves the technical problem that the original feature vectors have high dimensionality, are difficult to analyze, and cannot directly provide probabilistic inputs to the scheduling model. This allows subsequent scheduling optimization to proactively consider all possible situations of new energy output, fundamentally avoiding the drawbacks of human experience being unable to quantify fluctuation risks.

[0027] Specifically, firstly, the optimal number of clusters is determined by combining the elbow method with the silhouette coefficient. For the preprocessed renewable energy power generation feature vectors, a reasonable candidate interval for the number of clusters is preset. For each cluster within the interval, the sum of squares within the cluster is calculated and an elbow curve is plotted. The number of clusters corresponding to the inflection point where the rate of decline of the curve decreases from steep to gentle is selected as the optimal value of the elbow method. Then, the silhouette coefficient corresponding to each cluster within the candidate interval is calculated. The value with the silhouette coefficient closest to 1 and matching the optimal value of the elbow method is selected as the comprehensive optimal number of clusters k. Secondly, based on this number of clusters k, an improved K-means clustering algorithm is used. Feature vectors with the same number as the optimal number of clusters are randomly selected from the feature vector sample set as the initial cluster centers. Then, the Euclidean algorithm is calculated for each feature vector in the sample set and all the initial cluster centers. The Wasserstein distance is used to divide each feature vector into sample clusters corresponding to the cluster centers with the smallest Euclidean distance. After the initial division, the mean of each dimension of all feature vectors in each sample cluster is calculated. This mean vector is used as the new cluster center to replace the original center. The operation of distance calculation, sample division, and center update is repeated until the absolute value of the numerical deviation of each dimension of the cluster centers in two adjacent iterations is less than a preset threshold, and the clustering results converge. At this time, the final cluster centers of each sample cluster are extracted and used as typical renewable energy power generation scenarios that can reflect the core fluctuation characteristics of new energy output. Subsequently, the initial probability statistical analysis of typical power generation scenarios is performed based on the Wasserstein distance to obtain the initial probability results. Then, the initial probability is corrected by the Laplace algorithm to finally determine the accurate probability distribution results of the power generation scenarios. The advantages of this approach are threefold. First, the improved K-means clustering combined with the elbow method ensures the representativeness and discriminative power of typical scenarios, preserving the core fluctuation characteristics of renewable energy output while significantly reducing data dimensionality and analytical complexity, allowing the scheduling model to adapt efficiently. Second, the probability determination method based on Wasserstein distance combined with the Laplace algorithm accurately quantifies the probability of occurrence of each typical scenario, effectively avoiding the bias problem of probability statistics, and making the probability distribution results of power generation scenarios more consistent with actual operating conditions. Third, transforming the uncertainty of renewable energy output into a probabilistic set of typical scenarios provides standardized and probabilistic key inputs for subsequent scheduling optimization models, enabling scheduling schemes to be collaboratively optimized based on the probabilities of each scenario. This significantly improves the adaptability and robustness of scheduling schemes in the face of actual fluctuations in renewable energy, achieving precise matching between renewable energy output and data center power load, while ensuring the stability of data center power supply and the efficiency of renewable energy absorption.

[0028] Furthermore, in step S4, precise equipment load data is provided for subsequent charging and discharging power calculations and scheduling optimization model constraint settings. Simultaneously, a refined breakdown of the overall power load of the data center is achieved, enabling energy scheduling strategies to accurately match the actual load demands of each device. Specifically, this is achieved by classifying, analyzing, quantifying, and mapping the collected operating status data of each selected physical device. First, the operating status data is classified according to device type. Core data such as CPU utilization, memory usage, task processing volume, and operating power are extracted from server clusters. Core data such as state of charge, real-time charging and discharging power, and equipment operating efficiency are extracted from energy storage devices. Core data such as real-time output power and equipment operating load rate are extracted from renewable energy power generation equipment. Then, the actual load rate is calculated by combining server operating power with rated power. The energy storage load level is determined by combining the state of charge of energy storage devices with rated capacity. The power generation load is determined by combining the real-time output power of power generation equipment with rated output. Finally, the calculated quantitative indicators are normalized to obtain complete load information for each physical device, including load type, load value, load fluctuation characteristics, and load threshold.

[0029] Furthermore, in step S5, considering the differences in load types among selected physical devices such as renewable energy power generation equipment, energy storage equipment, and computing core equipment within the data center, a classification analysis and step-by-step calculation approach is adopted to sequentially complete the derivation of electricity load demand and power balance analysis, ultimately obtaining the charging and discharging power values ​​and power constraint ranges for each physical device to meet scheduling requirements. Specifically, firstly, for power generation equipment, the actual output power demand is determined based on its output load rate; for computing core equipment, the minimum power supply demand required for stable operation is determined based on its operating load rate; and for energy storage equipment, its charging and discharging power regulation requirements are determined based on its state of charge (SOC) load level. Secondly, a neural network model is used to perform power balance analysis on the electricity load demand of each physical device to obtain the corresponding charging and discharging power of the physical device.

[0030] The technical principle of the neural network model for power balance analysis is as follows: It uses the power load demand of each physical device, the inherent rated parameters of the device, the operational safety constraint thresholds, and the energy supply and demand gap within the data center scheduling cycle as multi-dimensional input features. The model's temporal convolutional layers mine the temporal correlation patterns of load power and output power at different times. An attention mechanism is used to assign high weights to key features affecting power balance (such as energy storage SOC level and renewable energy output fluctuation amplitude). Then, a fully connected layer maps the features to the solution space of each device's charging and discharging power. With real-time energy supply and demand balance of the data center as the core optimization objective, and with upper and lower limits of device charging and discharging power and power change rate as constraints, the model parameters are iteratively optimized through backpropagation. This ensures that the charging and discharging power values ​​of each physical device output by the model satisfy both the operational safety constraints of individual devices and achieve overall power supply and demand matching for the data center. Finally, it accurately outputs charging and discharging power values ​​and constraint ranges that adapt to scheduling requirements.

[0031] Furthermore, in step S6, a quantitative optimization model that takes into account both economic and low-carbon goals is constructed. This model integrates the probabilistic characteristics of new energy output, the power constraints of physical equipment, and the energy supply and demand patterns of data centers into a unified decision-making framework. This addresses the problems of unstable power supply and resource waste caused by the lack of quantitative goals and precise constraints in existing manual scheduling. The model enables the scheduling scheme to achieve the optimal balance between operating costs and carbon emission costs while ensuring safe equipment operation and energy supply and demand balance. At the same time, the probabilistic scenarios and equipment constraints are transformed into executable scheduling strategies through model solving, fundamentally improving the scientific nature, accuracy, and operability of renewable energy scheduling in data centers.

[0032] Preferably, the core objective function is to minimize the sum of grid power purchase cost, equipment operation and maintenance cost, energy storage loss cost, and carbon emission cost throughout the entire data center scheduling cycle. This objective function includes multiple cost factors, and the weights of these cost factors are dynamically adjusted according to actual scheduling needs. Next, based on data center energy balance constraints and the charging and discharging power constraints of each physical device, and combined with multiple dimensions such as equipment operation safety and task execution requirements, a complete scheduling optimization model constraint system is constructed. Finally, the probability distribution results of power generation scenarios are used as model input, and a combination of column and constraint generation algorithms is employed to solve the model. This decomposes the complex multi-scenario optimization problem into a main problem and sub-problems, iteratively solving them until convergence to the optimal solution. The solution is then transformed into a renewable energy scheduling scheme that includes power procurement, energy storage charging and discharging, and equipment load adjustment, and is then sent to the data center control system for execution.

[0033] Specifically, the formula for calculating the objective function is as follows: in, This indicates the amount of electricity purchased from the power grid at different times. Indicates carbon emission factor, M(t) represents the carbon price; M(t) represents the task migration amount in each time period; N(t) represents the task delay amount in each time period. and Indicates the charging and discharging power of the energy storage system; , This represents the migration cost coefficient, the delay execution penalty cost coefficient, and the energy storage loss cost coefficient.

[0034] Preferably, the constraints of the scheduling optimization model are a multi-dimensional system, which includes six core categories, specifically: 1. Energy balance constraint is the core fundamental constraint, requiring that the sum of the power purchased by the data center grid, the power output of renewable energy, and the power discharge of energy storage in each time period within the scheduling cycle be equal to the sum of the total load power of the data center and the charging power of energy storage, so as to ensure real-time balance between energy supply and demand.

[0035] 2. Physical equipment charging and discharging power constraints: Based on the rated parameters and real-time operating status of each device, limit the maximum / minimum charging and discharging power and power adjustment range of energy storage devices, power generation supporting equipment, etc., and at the same time clarify the lower limit of the power supply of the core computing equipment to avoid the equipment from operating beyond its power.

[0036] 3. Equipment operation safety constraints, including equipment power change rate constraints, limit the maximum fluctuation range of equipment operating power in adjacent time periods to avoid damage to the equipment from frequent start-stop. At the same time, safety threshold constraints such as operating temperature and state of charge are set according to the equipment technical specifications.

[0037] in, This indicates that a constrained device is within a certain time period. t Operating power The maximum allowable power variation of the equipment within adjacent time periods.

[0038] 4. State of Charge (SOC) constraint for energy storage devices: This constraint limits the state of charge of energy storage devices to between preset minimum and maximum thresholds at any time during the scheduling cycle, ensuring the physical safety of the energy storage devices while retaining a certain adjustment margin.

[0039] Assume the energy storage device is in time t The state of charge is SOC t ,but: And satisfy: in, , The charging power and discharging power of the energy storage device during time period t; , The charging and discharging efficiency of energy storage devices; , The minimum and maximum states of charge allowed for energy storage devices.

[0040] 5. Task execution constraints: For computing power tasks carried by data center servers, set completion time constraints for migrated and delayed tasks to ensure the real-time performance of high-priority tasks, while limiting the upper and lower limits of server load to avoid overload or inefficient operation.

[0041] Let the first i The execution power of each deferred task is p i,t The total computational load required for the task is D i The latest completion time is Then we have: Let the first j A portable task on the server m ,time t The computing resources on the platform are allocated as follows: x j,m,t The total computational cost required for the transferable task is D j Then we have: 6. Constraints on power purchase and renewable energy utilization: In conjunction with electricity market rules, set the adjustment range of power purchase capacity of the power grid, and at the same time, set the minimum consumption ratio constraint of renewable energy output based on the renewable energy consumption target of data centers, so as to improve the utilization efficiency of renewable energy.

[0042] Another embodiment of the present invention provides a collaboratively optimized data center energy scheduling system; for details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown illustrates the structure of a collaboratively optimized data center energy scheduling system according to one embodiment of the present invention. The system includes: The data acquisition module 11 is used to acquire historical renewable energy power generation data of the data center and operating status data of each selected physical device; The feature vector conversion module 12 is used to extract features from the historical renewable energy power generation data and convert the feature extraction results into vectors to obtain the renewable energy power generation feature vector of the data center. Clustering analysis module 13 is used to perform clustering analysis on the renewable energy power generation feature vector to obtain typical renewable energy power generation scenarios of the data center, and determine the probability distribution results of the power generation scenarios based on the typical renewable energy power generation scenarios; The load analysis module 14 is used to determine the load status of each physical device based on the various operating status data. The power analysis module 15 is used to analyze each load condition to obtain the charging and discharging power of each physical device. The scheduling scheme generation module 16 is used to construct a renewable energy scheduling optimization model with the objective function of minimizing the combined operating cost and carbon emission cost of the data center, and to determine the constraints of the renewable energy scheduling optimization model based at least on the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device, and to solve the renewable energy scheduling optimization model based on the constraints and the probability distribution results of the power generation scenario, so as to obtain and execute the renewable energy scheduling scheme.

[0043] Furthermore, in the above embodiments, the feature vector transformation module includes: The data cleaning unit is used to clean the historical renewable energy power generation data to obtain cleaned renewable energy power generation data. The normalization processing unit is used to normalize the cleaned renewable energy power generation data to obtain normalized renewable energy power generation data. The vector transformation unit is used to extract the power generation characteristics of the normalized renewable energy power generation data, and perform the vector transformation on the power generation characteristics to obtain the renewable energy power generation characteristic vector of the data center.

[0044] Furthermore, in the above embodiments, the clustering analysis module includes: The elbow method analysis unit is used to analyze the feature vector of the renewable energy power generation based on the elbow method to obtain the number of clusters; The clustering analysis unit is used to cluster the renewable energy power generation feature vectors based on the number of clusters and the improved K-means clustering algorithm, and obtain the clustering results of each renewable energy power generation feature vector; The scene extraction unit is used to extract power generation scenarios from all the clustering results to obtain the typical renewable energy power generation scenarios.

[0045] Furthermore, in the above embodiments, the clustering analysis module includes: An initial probability statistics unit is used to perform initial probability statistical analysis on the typical renewable energy power generation scenario based on the Wasserstein distance, and obtain the initial probability results of the typical renewable energy power generation scenario. The probability correction unit is used to correct the initial probability result based on the Laplace algorithm to obtain the probability distribution result of the power generation scenario.

[0046] Furthermore, in the above embodiments, the power analysis module includes: The load demand analysis unit is used to determine the power load demand of each physical device based on its load status. The power balance analysis unit is used to perform power balance analysis on the power load demand of each physical device to obtain the charging and discharging power of the corresponding physical device.

[0047] Compared to existing technologies, the beneficial effects of this invention are at least one of the following: This invention, through feature extraction and vector transformation of historical renewable energy power generation data, automatically extracts key fluctuation patterns from time-series power generation data and transforms them into standardized mathematical vectors, providing a structured data foundation for subsequent intelligent analysis; This invention, through cluster analysis of power generation feature vectors and determination of the probability distribution results of power generation scenarios, abstracts and quantifies the continuous and complex uncertainty of new energy output into a probability model composed of a finite number of typical scenarios and their probabilities of occurrence, solving the limitation that human experience cannot scientifically describe and quantify fluctuation risks; This invention, by using the probability distribution results of power generation scenarios as key inputs to solve the optimization model, achieves deep integration of the scheduling decision-making process and the probability model of new energy uncertainty, enabling the final generated scheduling scheme to proactively consider the impact of various possible power generation scenarios, thereby significantly improving the adaptability and reliability of the scheme in actual fluctuation environments, and solving the vulnerability problem of performance drop when deterministic optimization methods face prediction deviations.

[0048] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A collaborative optimization method for data center energy scheduling, characterized in that, include: Acquire historical renewable energy generation data and operational status data of selected physical devices in the data center; Feature extraction is performed on the historical renewable energy power generation data, and the feature extraction results are vectorized to obtain the renewable energy power generation feature vector of the data center; Cluster analysis is performed on the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios for the data center, and the probability distribution results of the power generation scenarios are determined based on the typical renewable energy power generation scenarios. Based on the various operational status data, determine the load status of each physical device; Analyze each load condition to obtain the corresponding charging and discharging power of each physical device. A renewable energy scheduling optimization model is constructed with the goal of minimizing the combined operating cost and carbon emission cost of the data center. The constraints of the renewable energy scheduling optimization model are determined based on at least the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device. The renewable energy scheduling optimization model is solved based on the constraints and the probability distribution results of the power generation scenario to obtain and execute a renewable energy scheduling scheme.

2. The collaborative optimization data center energy scheduling method as described in claim 1, characterized in that, The step of extracting features from the historical renewable energy power generation data and performing vector transformation on the extracted features to obtain the renewable energy power generation feature vector of the data center includes: The historical renewable energy power generation data is cleaned to obtain cleaned renewable energy power generation data; The cleaned renewable energy power generation data is normalized to obtain normalized renewable energy power generation data; Extract the power generation characteristics from the normalized renewable energy power generation data, and perform the vector transformation on the power generation characteristics to obtain the renewable energy power generation characteristic vector of the data center.

3. The collaborative optimization data center energy scheduling method as described in claim 1, characterized in that, The clustering analysis of the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios for the data center includes: Based on the elbow method, the feature vector of the renewable energy power generation is analyzed to obtain the number of clusters; Based on the number of clusters and the improved K-means clustering algorithm, the renewable energy power generation feature vectors are clustered to obtain the clustering results of each renewable energy power generation feature vector; The clustering results are processed to extract power generation scenarios, resulting in the typical renewable energy power generation scenarios.

4. The collaborative optimization data center energy scheduling method as described in claim 1, characterized in that, The determination of the probability distribution results of the power generation scenario based on the typical renewable energy power generation scenario includes: Based on the Wasserstein distance, an initial probability statistical analysis was performed on the typical renewable energy power generation scenario to obtain the initial probability results of the typical renewable energy power generation scenario; Based on the Laplace algorithm, the initial probability results are corrected to obtain the probability distribution results of the power generation scenario.

5. The collaborative optimization data center energy scheduling method as described in claim 1, characterized in that, The step of analyzing each load condition to obtain the corresponding charging and discharging power of each physical device includes: Based on the load status of each physical device, determine the corresponding power load requirement of that physical device; A power balance analysis is performed on the electrical load demand of each physical device to obtain the corresponding charging and discharging power of the physical device.

6. A collaboratively optimized data center energy scheduling system, characterized in that, include: The data acquisition module is used to acquire historical renewable energy power generation data and operational status data of selected physical devices in the data center; The feature vector conversion module is used to extract features from the historical renewable energy power generation data and convert the feature extraction results into vectors to obtain the renewable energy power generation feature vector of the data center. The clustering analysis module is used to perform clustering analysis on the renewable energy power generation feature vectors to obtain typical renewable energy power generation scenarios of the data center, and determine the probability distribution results of the power generation scenarios based on the typical renewable energy power generation scenarios; The load analysis module is used to determine the load status of each physical device based on the various operating status data. The power analysis module is used to analyze each load condition to obtain the charging and discharging power of each physical device. The scheduling scheme generation module is used to construct a renewable energy scheduling optimization model with the objective function of minimizing the combined operating cost and carbon emission cost of the data center. It determines the constraints of the renewable energy scheduling optimization model based at least on the energy balance constraints of the data center and the charging and discharging power corresponding to each physical device. It solves the renewable energy scheduling optimization model based on the constraints and the probability distribution results of the power generation scenario to obtain and execute the renewable energy scheduling scheme.

7. The collaborative optimization data center energy scheduling system as described in claim 6, characterized in that, The feature vector transformation module includes: The data cleaning unit is used to clean the historical renewable energy power generation data to obtain cleaned renewable energy power generation data. The normalization processing unit is used to normalize the cleaned renewable energy power generation data to obtain normalized renewable energy power generation data. The vector transformation unit is used to extract the power generation characteristics of the normalized renewable energy power generation data, and perform the vector transformation on the power generation characteristics to obtain the renewable energy power generation characteristic vector of the data center.

8. The collaborative optimization data center energy scheduling system as described in claim 6, characterized in that, The clustering analysis module includes: The elbow method analysis unit is used to analyze the feature vector of the renewable energy power generation based on the elbow method to obtain the number of clusters; The clustering analysis unit is used to cluster the renewable energy power generation feature vectors based on the number of clusters and the improved K-means clustering algorithm, and obtain the clustering results of each renewable energy power generation feature vector; The scene extraction unit is used to extract power generation scenarios from all the clustering results to obtain the typical renewable energy power generation scenarios.

9. A collaboratively optimized data center energy scheduling system as described in claim 6, characterized in that, The clustering analysis module includes: An initial probability statistics unit is used to perform initial probability statistical analysis on the typical renewable energy power generation scenario based on the Wasserstein distance, and obtain the initial probability results of the typical renewable energy power generation scenario. The probability correction unit is used to correct the initial probability result based on the Laplace algorithm to obtain the probability distribution result of the power generation scenario.

10. A collaboratively optimized data center energy scheduling system as described in claim 6, characterized in that, The power analysis module includes: The load demand analysis unit is used to determine the power load demand of each physical device based on its load status. The power balance analysis unit is used to perform power balance analysis on the power load demand of each physical device to obtain the charging and discharging power of the corresponding physical device.

Images