Data center energy consumption modeling method, system and equipment based on dynamic coupling of electric power and computing power, and medium

Abstract

The invention belongs to the technical field of data center energy efficiency management, and discloses a data center energy consumption modeling method, system, equipment and medium based on dynamic coupling of electric power and computing power, so as to solve the dynamic coupling modeling problem of the electric power and the computing power. The method comprises the following steps: acquiring and processing power side operation data, computing power side load data and scene association data of a data center to form a multi-dimensional data set; extracting a core coupling factor between the electric power and the computing power, and constructing a time sequence correlation model; generating an energy consumption prediction model adaptive to the current operation scene based on the time sequence correlation model; comparing a prediction result of the energy consumption prediction model with actual energy consumption data, calculating a prediction error, and optimizing and re-evaluating the core coupling factor; and performing multi-scene test verification on the optimized energy consumption prediction model, and outputting an energy consumption modeling result. According to the method, through fusion of dynamic coupling, scene self-adaption, closed-loop iteration and interpretability analysis, the accuracy, robustness and practicability of data center energy consumption prediction are improved.

Description

Technical Field

[0001] This invention belongs to the field of data center energy efficiency management technology, specifically relating to a data center energy consumption modeling method, system, equipment, and medium based on dynamic coupling of power and computing power. Background Technology

[0002] With the continuous advancement of the digital economy, data centers, as the key physical infrastructure supporting computing services, are experiencing rapid expansion in both construction scale and operational numbers, leading to increasingly significant energy consumption issues. The energy consumption of data centers is complex, encompassing not only the energy consumption of the information technology equipment itself but also the energy consumption of critical auxiliary facilities such as cooling systems and power distribution systems. Furthermore, the overall energy consumption is influenced by a combination of factors, including load conditions, environmental conditions, and equipment performance. Achieving accurate energy consumption modeling is a crucial technological foundation for optimizing data center energy efficiency and controlling operating costs, and has become a research focus in this field.

[0003] Currently, common data center energy consumption modeling methods mainly fall into two categories: The first is a single-dimensional modeling method based on power consumption. This method constructs a static energy consumption statistical model by collecting data such as power supply and various energy consumption components. Its limitation lies in focusing only on power-side data and failing to reflect the dynamic impact of computing load fluctuations on energy consumption. Therefore, in scenarios with frequent changes in computing demand, the model's prediction accuracy significantly decreases. The second is a modeling method based on computing demand. This method estimates energy consumption based on computing-side indicators such as server utilization and task load, but it does not fully consider the actual interference of power supply stability and voltage fluctuations on energy consumption, leading to a significant increase in model error when power supply conditions change.

[0004] Furthermore, most existing modeling methods employ fixed parameter systems, lacking the ability to adaptively adjust to different operating scenarios. For example, when facing differentiated scenarios such as peak-valley electricity price transitions, sudden computing power demands, and low-load sleep periods, the correlation between electricity and computing power often changes. Existing models lack corresponding dynamic parameter adjustment mechanisms, resulting in weak applicability under different operating conditions. Simultaneously, most existing models are open-loop modeling structures, lacking closed-loop iterative optimization mechanisms. This prevents continuous correction of model parameters based on actual energy consumption monitoring data, leading to a decline in model prediction accuracy after long-term operation.

[0005] On the other hand, existing methods have relatively limited scenario coverage, mostly designed only for full-load or stable operation. They are insufficiently adaptable to complex operating scenarios such as sudden increases in computing power demand and fluctuations in power supply, making it difficult to effectively match the diverse business needs and operating states of real data centers. These limitations restrict the predictive accuracy and practical application value of existing energy consumption modeling methods, making it difficult to support data centers in achieving refined and intelligent decision-making in resource scheduling and energy consumption management, thus affecting the potential for overall energy efficiency improvement.

[0006] In summary, existing technologies have not yet achieved dynamic coupling modeling of power and computing power, making it difficult to adapt to changes in multiple scenarios, and they lack the ability for parameter self-calibration and model self-evolution. Summary of the Invention

[0007] Based on the aforementioned shortcomings and deficiencies in the existing technology, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the existing technology. In other words, one of the objectives of this invention is to provide a data center energy consumption modeling method, system, device, and medium based on dynamic coupling of power and computing power that meets one or more of the aforementioned requirements, so as to achieve accurate characterization of the dynamic relationship between power and computing power, adaptive modeling for different operating scenarios, and continuous closed-loop optimization of the model, thereby improving the accuracy, robustness, and practical value of data center energy consumption prediction.

[0008] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a data center energy consumption modeling method based on dynamic coupling of power and computing power, comprising the following steps: S1. Collect data on the power side of the data center, the load data on the computing power side, and the scenario-related data, and perform time-series alignment processing on the collected multi-source data to form a time-series aligned multi-dimensional dataset. S2. Extract the core coupling factors between power and computing power from the multi-dimensional dataset, and construct a time-series correlation model with time as the dimension to characterize the dynamic coupling law between power-side parameters and computing power-side parameters. S3. Based on the time-series correlation model, an energy consumption prediction model adapted to the current operating scenario is generated through a parameter adaptive optimization mechanism. S4. Compare the prediction results of the energy consumption prediction model with the actual energy consumption data of the data center, calculate the prediction error, optimize the parameters of the time series correlation model based on the prediction error, and re-evaluate the core coupling factor. S5. Perform multi-scenario testing and verification on the energy consumption prediction model optimized in step S4, and output the energy consumption modeling results.

[0010] As a preferred approach, step S1 involves performing time-series alignment processing on the collected multi-source data, specifically as follows: Timestamp the raw data points from different data sources; Define a standard timestamp sequence and a tolerance delay time window corresponding to each standard timestamp in the standard timestamp sequence; For each standard timestamp and each data source in the standard timestamp sequence, an alignment value is determined based on preset rules, which include, in sequence: If a data point from the current data source exists within the tolerance delay time window corresponding to the current standard timestamp, the value of the data point whose timestamp is closest to the current standard timestamp is selected as the alignment value. If it does not exist, the alignment value is calculated by linear interpolation based on the valid data points of the current data source within the adjacent time window before and after the current standard timestamp; If linear interpolation is not possible, the previous valid alignment value is used to fill in the gaps to generate the alignment value.

[0011] As a preferred approach, step S2 extracts the core coupling factor between power and computing power, specifically: Calculate the Pearson correlation coefficient between each candidate parameter and the total power consumption in the power-side operation data, the computing power-side load data, and the scenario-related data, and screen for linear correlation factors; Calculate the maximum information coefficient between each candidate parameter and the total power consumption, and screen for nonlinear correlation factors; The core coupling factor is determined by combining the linear and nonlinear correlation factors.

[0012] As a preferred embodiment, the temporal correlation model in step S2 is a state-space model, the construction of which includes: Define the system state vector The system state vector This includes the power consumption of IT equipment, the power consumption of the cooling system, and the components of the power supply and distribution system; Constructing state equations ,in, Here is the state transition matrix. For external driving matrix, The input vector contains parameters such as computing load and ambient temperature. This is process noise; Observation equations are constructed based on the system state vector. ,in, The observed total power consumption of the data center. For the observation matrix, To observe noise.

[0013] As a preferred embodiment, the parameter adaptive optimization mechanism in step S3 includes: A preset benchmark library is provided, which includes multiple typical operating scenarios and their corresponding initial parameter sets for the models. Based on the real-time running data characteristics reflected by the multi-dimensional dataset, the scenario most similar to the current running scenario is matched from the benchmark library, and the initial parameters corresponding to the most similar scenario are loaded. The initial parameter set is dynamically adjusted using a Bayesian optimization algorithm to minimize the prediction bias of the time-series correlation model on the validation dataset, thereby generating the energy consumption prediction model.

[0014] As a preferred embodiment, step S4 involves reverse correction of the core coupling factor and the parameters of the time-series correlation model based on the prediction error, including: The rolling prediction error of the energy consumption prediction model is continuously calculated; When the rolling prediction error continuously exceeds the dynamically set threshold, the optimization process is triggered; The optimization process includes re-evaluating and updating the core coupling factor, optimizing and adjusting the parameters of the time series correlation model, and, if the rolling prediction error still exceeds the threshold after parameter adjustment, evaluating and adjusting the structure of the time series correlation model.

[0015] As a preferred option: The multi-scenario testing and verification includes full-load operation scenario, low-load operation scenario, sudden computing power demand scenario, and electricity price switching scenario. The output energy consumption modeling results include energy consumption prediction data for future periods and identification results of key energy consumption nodes.

[0016] Secondly, the present invention provides a data center energy consumption modeling system based on dynamic coupling of power and computing power, used to implement the data center energy consumption modeling method as described in the first aspect, comprising: The multi-dimensional data acquisition module is used to collect data on the power side of the data center, the load data on the computing power side, and the scenario-related data, and to perform time-series alignment processing on the collected multi-source data to form a time-series aligned multi-dimensional dataset. The dynamic coupling analysis module is used to extract the core coupling factors between power and computing power from the multi-dimensional dataset and construct a time-series correlation model with time as the dimension to characterize the dynamic coupling law between power-side parameters and computing power-side parameters. The model dynamic generation module is used to generate an energy consumption prediction model that adapts to the current operating scenario based on the time-series correlation model through a parameter adaptive optimization mechanism. The closed-loop iterative optimization module is used to compare the prediction results of the energy consumption prediction model with the actual energy consumption data of the data center, calculate the prediction error, optimize the parameters of the time series correlation model based on the prediction error, and re-evaluate the core coupling factor. The model output module is used to perform multi-scenario testing and verification on the optimized energy consumption prediction model and output the energy consumption modeling results.

[0017] Thirdly, the present invention provides an electronic device, the computer device including a memory, a processor and a computer program, wherein when the computer program is executed by the processor, it implements the data center energy consumption modeling method as described in the first aspect.

[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the data center energy consumption modeling method as described in the first aspect.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention, through multi-dimensional data acquisition and time-series alignment, and by screening core coupling factors based on correlation analysis, constructs a time-centric time-series correlation model (such as a state-space model). For the first time in energy consumption modeling, it systematically quantifies the dynamic interaction between electricity consumption and computing power demand over time. This overcomes the limitations of existing technologies that focus only on a single dimension (electricity or computing power), enabling the model to maintain high prediction accuracy even in complex scenarios such as drastic fluctuations in computing power demand or changes in power supply conditions.

[0020] 2. This invention introduces a parameter adaptive optimization mechanism. Through steps such as pre-setting a multi-scenario parameter benchmark library, real-time scenario matching, and Bayesian optimization, the model parameters can be dynamically adjusted according to the actual operating conditions of the data center (such as load level, ambient temperature, and electricity price periods). This solves the problem of poor adaptability of existing fixed-parameter models when facing diverse scenarios such as peak-valley electricity price switching, sudden computing power demands, and seasonal temperature changes. It endows the model with intelligent "location-specific" characteristics, greatly expanding its application scope and practical value.

[0021] 3. This invention designs a closed-loop iterative optimization process based on prediction error feedback. By continuously monitoring the deviation between the model's prediction results and actual energy consumption, and automatically triggering the re-evaluation of core coupling factors, re-optimization of model parameters, and even adjustment of the model structure when the error exceeds a dynamic threshold, this mechanism enables the model to have the ability to self-diagnose, self-correct, and self-evolve. This effectively solves the problem of model accuracy "decaying" over time due to equipment aging, business model changes, or infrastructure upgrades in traditional open-loop modeling methods, ensuring the reliability and stability of the model throughout its entire lifecycle in the data center.

[0022] 4. The output of this invention is not limited to energy consumption prediction data, but also includes key energy consumption node identification results obtained based on interpretable analysis methods such as SHAP values. This enables operators to clearly understand the specific contribution of different devices, systems, or business loads to total energy consumption, and accurately locate energy efficiency bottlenecks and abnormal consumption points. This provides direct, actionable data support for data center resource scheduling (such as task migration and server hibernation), infrastructure optimization (such as cooling strategy adjustments), and cost control, driving the evolution of energy efficiency management from extensive to refined and intelligent.

[0023] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0025] Figure 1 This is a flowchart illustrating the data center energy consumption modeling method described in Embodiment 1 of the present invention.

[0026] Figure 2 This is a schematic diagram of the data center energy consumption modeling system described in Embodiment 2 of the present invention.

[0027] Figure 3 This is a structural diagram of the electronic device described in Embodiment 3 of the present invention.

[0028] Icon labels: 300. Electronic devices; 301. Processor; 302. Communication bus; 303. User interface; 304. Network interface; 305. Memory. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0030] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0031] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0032] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0033] The data center energy consumption modeling method described in the embodiments of this specification is applicable to modern data centers of various sizes and types, especially those operating environments with significant fluctuations in computing load, complex and diverse operating scenarios, and refined requirements for energy efficiency management. In these scenarios, the application of the data center energy consumption modeling method aims to break through the limitations of traditional static or single-dimensional models. By predicting overall and component energy consumption in real time and accurately, it provides scientific data support and decision-making basis for dynamic resource scheduling (such as server cluster expansion and contraction, task migration), cooling system optimization and control, peak-valley electricity pricing strategy formulation, and carbon footprint monitoring of data centers, ultimately achieving the goals of reducing operating costs (OPEX) and improving power efficiency (PUE).

[0034] The following is a brief explanation of the data center, core coupling factor, time series correlation model, parameter adaptive optimization mechanism, energy consumption prediction model, and energy consumption modeling results involved in several embodiments of this specification: A data center is a physical location that houses a large number of information technology (IT) devices (such as servers, storage devices, and network equipment) and the infrastructure (including but not limited to cooling systems, power distribution systems, and monitoring and management systems) that provides a stable operating environment for them. Its core function is to centrally process, store, and exchange data and information, providing computing power services to society.

[0035] Core coupling factors refer to a set of parameters that can significantly characterize the dynamic interaction between power consumption and computing power demand, selected from data center power-side operation data, computing power-side load data, and scenario-related data through correlation analysis (such as Pearson correlation coefficient and maximum information coefficient). These factors typically include, but are not limited to: computing power indicators (such as CPU / GPU utilization), environmental impact indicators (such as outdoor wet-bulb temperature), and infrastructure status indicators (such as cooling system efficiency COP).

[0036] A time-series correlation model refers to a model that uses time as a dimension to mathematically describe the dynamic coupling relationship between power-side parameters and computing-side parameters. In this embodiment of the invention, a state-space model or similar form is preferred, using system matrices (such as state transition matrix A, input matrix B, and observation matrix C) to quantify the evolution of this coupling relationship over time, effectively characterizing the system's inertia, delay, and nonlinearity.

[0037] The parameter adaptive optimization mechanism refers to an intelligent process that enables model parameters to automatically adjust based on the real-time operating status of the data center and the external environment. This mechanism typically includes: a pre-built multi-scenario parameter benchmark library (storing initial parameters under typical operating conditions), a real-time scene matching unit (identifying the most similar benchmark scene), and a parameter fine-tuning optimizer (such as a Bayesian optimization-based algorithm). Their common goal is to dynamically adapt the model parameters to the specific current scenario to maintain prediction accuracy.

[0038] The energy consumption prediction model refers to the mathematical model generated by the method of this invention, which can be used to numerically predict the total power consumption and individual power consumption of a data center within a specific future time period. It is an executable computational model or algorithm instance that is highly adapted to the current operating scenario, obtained by adjusting the parameters through a parameter adaptive optimization mechanism based on the aforementioned time-series correlation model.

[0039] Energy consumption modeling results refer to the structured data and insights output after running the energy consumption prediction model and conducting multi-scenario testing and verification. It mainly includes two parts: first, energy consumption prediction data, such as point prediction values ​​and probability confidence intervals at a specified future time granularity; and second, key energy consumption node identification results, which are the devices, systems, or time periods that contribute the most to total energy consumption or exhibit abnormal energy efficiency, as indicated by interpretability analysis (such as SHAP values), providing specific targets for energy-saving optimization.

[0040] Example 1: like Figure 1 As shown, this embodiment provides a data center energy consumption modeling method based on dynamic coupling of power and computing power, including the following steps: Step S1 involves multi-dimensional data acquisition: Simultaneously collecting data from the data center's power supply side, computing power side load, and scenario-related data to construct a complete modeling dataset. The collected data is then preliminarily validated, stored, and time-series aligned to form a time-series aligned multi-dimensional dataset. Step S1 specifically includes the following operations: Step S11, data synchronization acquisition, includes: 1. Power-side data acquisition: Using the Modbus / TCP protocol, total power consumption, branch current, voltage, power factor, and air conditioner operating status (fan frequency, set temperature) are collected from the intelligent cabinet PDU (Power Distribution Unit) and the precision air conditioner controller at a frequency of 1 time / second. The core data acquisition indicators of the power side are shown in Table 1.

[0041] Table 1:

[0042] 2. Computing Power Side Data Collection: Data collection agents deployed in the server operating system collect average CPU utilization (average across all cores), memory usage, and network inbound / outbound traffic at a frequency of once per second. Real-time task queue length, Pod replica count, and other business metrics are collected from the cluster management platform (e.g., Kubernetes) via API. The core data collection metrics and their representations on the computing power side are shown in Table 2 below.

[0043] Table 2:

[0044] 3. Scene-related data acquisition: The supply and return air temperatures of the hot / cold aisle are collected from temperature and humidity sensors in the data center via Modbus / RJ45 protocol; real-time outdoor dry / wet bulb temperatures of the data center location are obtained by accessing public meteorological data APIs. The core data acquisition indicators for scene-related data are shown in Table 3 below.

[0045] Table 3:

[0046] Step S12 Timing Alignment: Raw data streams from different sources with varying transmission delays are uniformly aligned to a series of standard timestamps. superior, ( Define an alignment time window (for time granularity, such as 1 second). (δ represents the tolerance delay, such as 2 seconds). Its position in... Alignment value at time Calculated according to the following rules: Rule 1 (Normal value): If in A data point exists within a window. ,and closest Then the data value of that point is directly taken as... The expression is: ; Rule 2 (Interpolation): If a certain indicator is in There is no data inside, but in the windows before and after it. , Memory in data points ( )and( If the result is negative, then linear interpolation is used for calculation. The calculation formula is: ; Rule 3 (Padding): If none of the above conditions are met (e.g., continuous packet loss), then the previous valid alignment value is used. Perform the filling and add a quality label to the data point. flagmissing The expression is: , .

[0047] Step S2 involves constructing the dynamic coupling relationship: extracting the core coupling factors between power and computing power from the multi-dimensional dataset, and constructing a time-series correlation model to characterize the dynamic coupling pattern between power-side parameters and computing power-side parameters. Step S2 specifically includes the following operations: Step S21: Data Preprocessing The prerequisite for constructing dynamic coupling relationships is to perform targeted preprocessing on the multi-dimensional data collected in step S1 to ensure the validity, consistency, and correlation of the data, laying the foundation for factor extraction and model construction, including: Step S211: Linear correlation analysis: The box plot method is used to remove data exceeding 1.5 times the interquartile range, filtering out extreme outliers in step S1. Missing data are then supplemented using linear interpolation, with the formula as follows: , In the formula, For missing data, This corresponds to the timestamp.

[0048] Step S212: Data normalization processing: Z-score is used to map all indices to the [-1, 1] interval to eliminate the influence of dimensional differences on the coupling analysis. The Z-score standardization formula is: , In the formula, The mean of the data. The standard deviation is denoted as .

[0049] Step S22, quantitative screening of multi-dimensional coupling factors, includes: Step S221: Linear correlation analysis: Calculate the candidate factors in the power side, computing power side, and scene-related data. The degree of linear correlation between power consumption and Pearson correlation coefficient is calculated as follows: , In the formula, and These are the mean of the sequence. The total number of data points, retaining | Factors with a value greater than 0.7 are considered strong linear candidate factors.

[0050] Step S222 Nonlinear Correlation Analysis: To capture complex relationships that linear regression cannot describe (such as periodicity and threshold effects), the MIC is calculated: , In the formula, It is mutual information. It is a grid into which the data space is divided. It is the upper limit of the number of grid divisions (usually taken as...). A MIC value close to 1 indicates a strong nonlinear relationship; factors with MIC > 0.6 are retained.

[0051] Step S223: Factor Determination and Classification comprehensive By combining the MIC value, the core coupling factor set is finally determined, and its physical classification is clarified.

[0052] IT power consumption The direct driving factors are mainly Computing power metrics.

[0053] Cooling power consumption The main driving factor is the heat dissipation of IT equipment, and Directly related. The key modulation factor is... Outdoor wet-bulb temperature directly affects the efficiency (COP value) of the refrigeration unit.

[0054] Power supply and distribution losses The driving factors include total load current and total power. , Related. The modulation factor is mainly related to the load rate, affecting the operating efficiency curves of equipment such as transformers and UPS.

[0055] Step S23 involves constructing a multi-energy flow coupling model based on physical mechanisms and data-driven methods, including: Step S231: State vector definition: Define the state vector of the system at time k. for: , in: The total power consumption of the IT equipment at time k; Let k be the total power consumption of the cooling system at time k; Let K be the total loss of the power supply and distribution system at time k.

[0056] Step S232: Construction of state equations: The system's state equations (describing the dynamic evolution of the system's state): , in, The matrix is ​​the state transition matrix, which describes the mutual influence and inertia between the internal states of the system. , express Its own inertia is usually close to 0 because IT power consumption can change instantaneously. (Core coupling coefficient) represents the driving force of IT power consumption on cooling power consumption; This represents the inertia of the refrigeration system itself. Due to the inertia of the compressor, fan, etc., the current refrigeration power consumption will affect the refrigeration power consumption at the next moment. This value is usually positive and less than 1. and The impact of IT power consumption and cooling power consumption on power distribution losses are respectively represented, because From the total load ( + Caused by, usually ≈ . This represents the inertia of power distribution losses themselves, which is usually very small.

[0057] in, The external driving matrix describes how external inputs directly affect the system state; the input vector... , , This indicates the driving efficiency of computing load (such as CPU utilization) on IT power consumption. For example, how many kW of IT power consumption increases for every 10% increase in CPU utilization. This indicates the modulating effect of outdoor wet-bulb temperature on cooling power consumption. As the temperature increases, the cooling efficiency (COP) decreases, requiring more power to dissipate the same amount of heat. This coefficient should be positive.

[0058] in, For process noise, representing small, unmodeled dynamics.

[0059] Step S233: Construction of observation equations: The observation equation describes how the total power consumption is observed. The relationship between the system's observed values ​​(total power consumption) and the state vector is as follows: , Where C=[1,1,1] is the observation matrix, representing that the total power consumption is the sum of IT power consumption, cooling power consumption, and power supply and distribution power consumption, i.e. , The noise level is used to observe the meter's measurement error.

[0060] Step S234 is based on the N4SID subspace identification algorithm: Using the subspace identification method (N4SID algorithm) and historical data sequences Solve for the optimal system matrix. , , and initial state This algorithm can directly estimate the state-space model representing the dynamic characteristics of the system from the input and output data.

[0061] Step S3 is the dynamic generation of the energy consumption model: Based on a preset multi-scenario parameter benchmark library, an adaptive optimization algorithm is used to match the real-time scenario and dynamically adjust the coupling parameters to generate a personalized energy consumption model.

[0062] Step S31: Construction of a multi-scenario parameter benchmark library: In data centers, different business loads and external environments lead to significant differences in energy consumption characteristics. To build a model that can adapt to various situations, a multi-scenario parameter benchmark library needs to be established first. Step S3 specifically includes the following operations: Step S311 Feature Engineering and Cluster Analysis: Key energy-related features were extracted from historical data, focusing on core metrics such as average CPU utilization (computing load), memory usage, outdoor wet-bulb temperature (affecting cooling efficiency), and time period (weekday / weekend, daytime / night). Operational data from the past year was collected, and feature vectors were extracted from it. For each time point... Define the feature vector: , in, , It is a periodic time encoding.

[0063] The k-means clustering algorithm is applied, with the objective function being: .

[0064] By iteratively allocating and updating the centroid, the operating conditions are divided into typical scenarios. Based on the elbow rule, the optimal number of clusters K=4 is determined, corresponding to four typical scenarios: peak trading day, nighttime batch processing, low load on weekends, and sudden AI training.

[0065] Step S312: Baseline Model Training For historical data in each clustering scenario, the corresponding state-space model is trained using the subspace identification method to obtain the optimal parameter set. Establish a parameter benchmark library, the expression is: .

[0066] Step S32, real-time scene matching and parameter fine-tuning, includes: Step S321 Real-time scene matching: The system calculates the feature vector of the current time window in real time. The weighted Euclidean distance method is used to calculate the distance between the current feature vector and the feature center of each scene in the benchmark library, and scene matching is performed. The calculation formula is as follows: .

[0067] Choose the scene with the shortest distance. That is, the scenario most similar to the current situation, and load the model parameters corresponding to that scenario. As initial parameters.

[0068] Step S322 Bayesian optimization parameter fine-tuning: Since real-time operating conditions cannot be completely identical to historical scenarios, fine-tuning of the initial parameters is necessary. A Bayesian optimization method is employed, aiming to minimize the prediction error of recent data (e.g., the past hour), to optimize the key parameters in the model.

[0069] The optimization objective is to minimize the normalized root mean square error, expressed as: .

[0070] The Bayesian optimization process is as follows: 1. Prior function: The objective function is modeled using a Gaussian process, and its expression is: , in, This is the Matern 5 / 2 kernel function.

[0071] 2. The data acquisition function, using the expected improvement criterion, is expressed as follows: , in, This is the current optimal value.

[0072] 3. Parameter Update: Determine the next evaluation point by maximizing the acquisition function. .

[0073] 4. Iterative optimization: Repeat the evaluation and update process until convergence to obtain the optimal parameters. .

[0074] Step S323 Model Validation and Update: The fine-tuned model needs to pass statistical tests: .

[0075] Simultaneously satisfying the statistical significance test ( p <0.05), ensuring the effectiveness of model improvements. Establish a model version management mechanism and retrain the benchmark library regularly. When the prediction error continuously exceeds the limit, trigger an emergency re-optimization process to ensure the continuous adaptability of the model.

[0076] Step S4 is the model closed-loop iteration: compare the model prediction results with the actual energy consumption data of the data center, and calculate the error value; when the error exceeds the preset threshold, trigger the closed-loop feedback mechanism, re-extract the coupling factors and optimize the model parameters and structure.

[0077] Step S41, the continuous monitoring and intelligent triggering judgment mechanism, includes: Step S411 Multi-index rolling error calculation: After model deployment, a real-time monitoring system is established to calculate various prediction error indicators. The main indicator is the Mean Absolute Percentage Error (MAPE), calculated using a sliding window: , Among them, window size (Data is presented in 1-minute increments, corresponding to a 24-hour period).

[0078] The auxiliary indicator is the root mean square error: Maximum absolute error: .

[0079] Step S412 Dynamic threshold setting algorithm: An adaptive threshold mechanism is employed to avoid the limitations of a fixed threshold. First, the baseline error level is calculated based on a historically strong performance period (e.g., the past 30 days). , .

[0080] Next, set a dynamic threshold: ,in This represents the recent trend term of the error. α This is the trend weighting coefficient.

[0081] Finally, set the trigger condition: when the MAPE exceeds the dynamic threshold for M consecutive time points (e.g., M=6 corresponds to 30 minutes), and RMSE and MaxAE are also abnormal at the same time, the optimization process is triggered.

[0082] Step S42, the deep iterative optimization process, includes: Step S421 Coupling factor re-evaluation mechanism: After optimization is triggered, the system automatically performs factor importance analysis, using random forest or XGBoost to calculate feature importance scores on the latest data: .

[0083] Retain importance score > (e.g., 0.05), Importance score < Factors with a correlation of 0.01 or less and a continuously decreasing correlation with the output are deleted. New factors with a strong correlation with the current residual are selected from the backup factor pool. The rolling correlation coefficient between the factor and energy consumption is calculated using a sliding window to detect abrupt changes in the relationship.

[0084] Step S422: Parameter Structure Linkage Optimization Establish a hierarchical optimization framework, and try different levels from simple to complex: Level 1, Parameter Optimization: With the current model structure fixed, the parameters are optimized using a sequential quadratic programming algorithm, expressed as:

[0085] R(θ) is a regularization term used to prevent overfitting.

[0086] Level Two, Structural Optimization Assessment: If the error still does not meet the target after parameter optimization, initiate structural optimization, try increasing / decreasing the dimension of state variables to adjust complexity, add cross terms or polynomial terms to the state equations to introduce nonlinearity, and introduce time delay terms for key input variables. Time delays are taken into account. The model family is expanded by comparing the state-space model with alternative models such as neural networks and support vector machines.

[0087] Optimize path selection: Select the optimal complexity based on the AIC criterion, expressed as: , in, For the number of parameters, This represents the model likelihood value.

[0088] Step S423 Model Validation and Version Management includes: 1. Cross-validation: The data is divided into training set (70%), validation set (15%), and test set (15%). 2. Statistical significance test: The Diebold-Mariano test is used to compare the prediction accuracy of the old and new models. The expression is: , in, This represents the squared difference of the prediction error.

[0089] 3. Version control: Establish a model version repository to record metadata for each optimization (triggered reason, improvement effect, and modified content). Step S43 Adaptive learning mechanism: For small daily drifts, incremental updates are used to avoid full retraining, expressed as: , Among them, learning rate It adapts to the rate of change of error.

[0090] Significant Change Detection: Key indicators for monitoring infrastructure changes include detecting sudden changes in server counts, identifying cooling system configuration changes, and structural changes in business load patterns. When a significant change is detected, a reinforcement learning process is triggered, involving exploratory training in a simulation environment.

[0091] Step S5 is the model output: the effectiveness of the model is verified through multi-scenario testing, and the energy consumption modeling results are output to provide support for data center energy efficiency optimization.

[0092] Step S51 high-precision energy consumption prediction output includes: 1. Point forecast: Predicted values ​​for the next 24 hours with 15-minute intervals. ; 2. Probability Prediction: Provides 80% and 95% confidence intervals, expressed as: ; 3. Scenario Analysis: Generate multiple sets of prediction results based on different business assumptions (such as peak load, normal load, and low load scenarios).

[0093] Step S52, key energy consumption node identification, includes: 1. Interpretability analysis based on SHAP values: , in, For the model, As input features, Features The SHAP value; 2. Contribution Analysis: Calculate the contribution ratio of each factor to the total power consumption; 3. Anomaly localization: Identifying devices and systems with abnormal fluctuations in contribution levels; 4. Bottleneck identification: Identify the critical nodes with the lowest energy efficiency.

[0094] Example 2: like Figure 2 As shown, this embodiment provides a data center energy consumption modeling system based on dynamic coupling of power and computing power, used to implement the data center energy consumption modeling method described in Embodiment 1, and deployed on physical servers or virtual machine clusters. It consists of a multi-dimensional data acquisition module, a dynamic coupling analysis module, a model dynamic generation module, a closed-loop iterative optimization module, and a model output module.

[0095] Specifically, the multi-dimensional data acquisition module is responsible for providing high-quality input data for the algorithm model. It synchronously collects power-side (total power consumption, PDU current and voltage), computing-side (CPU / memory utilization, task queue), and environmental data (indoor and outdoor temperature and humidity) through a unified clock source, using NTP / PTP protocols to ensure timestamp accuracy. The data verification submodule employs range checks (e.g., non-negative power consumption) and mutation detection to remove outliers. The stream processing engine (Apache Flink) performs time-series alignment at a 1-second granularity, and linear interpolation is used to fill in delayed data. Finally, strictly aligned time-series data is output to InfluxDB, providing a complete dataset for coupled analysis.

[0096] Specifically, the dynamic coupling analysis module quantifies the coupling relationships in the algorithm. Based on the collected time-series data, the correlation analysis submodule calculates the Pearson coefficient and the maximum information coefficient, and filters out core factors strongly correlated with total power consumption (such as CPU utilization and outdoor temperature). The time-series correlation model adopts a state-space equation form, and solves for system matrices A, B, and C through subspace identification (N4SID algorithm). Matrix A encodes inertia effects (such as cooling delay), and matrix B quantifies the input driving strength (such as the influence of temperature on cooling power consumption), ultimately establishing a system model of the form... The dynamic equations capture the transient coupling patterns of power computing power.

[0097] Specifically, the model dynamic generation module enables scene adaptive modeling. A pre-built benchmark library generates initial parameter sets for typical scenarios (such as "high summer load") by clustering historical data using k-means. The real-time scene matching submodule calculates the weighted Euclidean distance between the current features and the scene centroid, selecting the parameters of the most similar scene. The Bayesian optimization engine, starting with the initial parameters, fine-tunes key parameters (such as the temperature coefficient in matrix B) with the goal of minimizing the short-term prediction error (MAPE). The optimization process establishes a Gaussian process surrogate model, guides the search by expecting to improve the acquisition function, and quickly outputs personalized model parameters adapted to the current operating state.

[0098] Specifically, the closed-loop iterative optimization module ensures the model's continued effectiveness. The monitoring submodule calculates the prediction error MAPE in real time for a rolling window (e.g., 1 hour), and the dynamic threshold is automatically adjusted based on historical performance (baseline error + 3 standard deviations). When the error continues to exceed the standard, a factor re-evaluation is triggered: random forest analysis is used to analyze feature importance, eliminating ineffective factors (such as obsolete server indicators) and adding significant factors. Parameter-structure linkage optimization first adjusts parameters through sequential quadratic programming; if the accuracy requirements are still not met, the addition of nonlinear or time-delay terms is evaluated. Finally, the statistical significance of the new model improvement is verified through the Diebold-Mariano test, completing the iterative update.

[0099] Specifically, the model output module is used to test and verify the optimized model in multiple scenarios and output the energy consumption modeling results.

[0100] Example 3: like Figure 3 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0101] The communication bus can be used to enable communication between the various components mentioned above.

[0102] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0103] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0104] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0105] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a modeling application program. The processor can be used to call the modeling application program stored in the memory and execute the steps of the modeling methods mentioned in the foregoing embodiments.

[0106] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0107] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0108] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0109] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0110] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0111] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A data center energy consumption modeling method based on dynamic coupling of power and computing power, characterized in that, Including the following steps: S1. Collect data on the power side of the data center, the load data on the computing power side, and the scenario-related data, and perform time-series alignment processing on the collected multi-source data to form a time-series aligned multi-dimensional dataset. S2. Extract the core coupling factors between power and computing power from the multi-dimensional dataset, and construct a time-series correlation model with time as the dimension to characterize the dynamic coupling law between power-side parameters and computing power-side parameters. S3. Based on the time-series correlation model, an energy consumption prediction model adapted to the current operating scenario is generated through a parameter adaptive optimization mechanism. S4. Compare the prediction results of the energy consumption prediction model with the actual energy consumption data of the data center, calculate the prediction error, optimize the parameters of the time series correlation model based on the prediction error, and re-evaluate the core coupling factor. S5. Perform multi-scenario testing and verification on the energy consumption prediction model optimized in step S4, and output the energy consumption modeling results.

2. The data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1, characterized in that, Step S1 involves performing time-series alignment processing on the collected multi-source data, specifically as follows: Timestamp the raw data points from different data sources; Define a standard timestamp sequence and a tolerance delay time window corresponding to each standard timestamp in the standard timestamp sequence; For each standard timestamp and each data source in the standard timestamp sequence, an alignment value is determined based on preset rules, which include, in sequence: If a data point from the current data source exists within the tolerance delay time window corresponding to the current standard timestamp, the value of the data point whose timestamp is closest to the current standard timestamp is selected as the alignment value. If it does not exist, the alignment value is calculated by linear interpolation based on the valid data points of the current data source within the adjacent time window before and after the current standard timestamp; If linear interpolation is not possible, the previous valid alignment value is used to fill in the gaps to generate the alignment value.

3. The data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1, characterized in that, Step S2 extracts the core coupling factor between power and computing power, specifically: Calculate the Pearson correlation coefficient between each candidate parameter and the total power consumption in the power-side operation data, the computing power-side load data, and the scenario-related data, and screen for linear correlation factors; Calculate the maximum information coefficient between each candidate parameter and the total power consumption, and screen for nonlinear correlation factors; The core coupling factor is determined by combining the linear and nonlinear correlation factors.

4. A data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1 or 3, characterized in that, The temporal correlation model mentioned in step S2 is a state-space model, and its construction includes: Define the system state vector The system state vector This includes the power consumption of IT equipment, the power consumption of the cooling system, and the components of the power supply and distribution system; Constructing state equations ,in, Here is the state transition matrix. For external driving matrix, The input vector contains parameters such as computing load and ambient temperature. This is process noise; Observation equations are constructed based on the system state vector. ,in, The observed total power consumption of the data center. For the observation matrix, To observe noise.

5. The data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1, characterized in that, The parameter adaptive optimization mechanism described in step S3 includes: A preset benchmark library is provided, which includes multiple typical operating scenarios and their corresponding initial parameter sets for the models. Based on the real-time running data characteristics reflected by the multi-dimensional dataset, the scenario most similar to the current running scenario is matched from the benchmark library, and the initial parameters corresponding to the most similar scenario are loaded. The initial parameter set is dynamically adjusted using a Bayesian optimization algorithm to minimize the prediction bias of the time-series correlation model on the validation dataset, thereby generating the energy consumption prediction model.

6. The data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1, characterized in that, Step S4 involves reversely correcting the parameters of the core coupling factor and the time-series correlation model based on the prediction error, including: The rolling prediction error of the energy consumption prediction model is continuously calculated; When the rolling prediction error continuously exceeds the dynamically set threshold, the optimization process is triggered; The optimization process includes re-evaluating and updating the core coupling factor, optimizing and adjusting the parameters of the time series correlation model, and, if the rolling prediction error still exceeds the threshold after parameter adjustment, evaluating and adjusting the structure of the time series correlation model.

7. The data center energy consumption modeling method based on dynamic coupling of power and computing power according to claim 1, characterized in that: The multi-scenario testing and verification includes full-load operation scenario, low-load operation scenario, sudden computing power demand scenario, and electricity price switching scenario. The output energy consumption modeling results include energy consumption prediction data for future periods and identification results of key energy consumption nodes.

8. A data center energy consumption modeling system based on dynamic coupling of power and computing power, characterized in that, The method for implementing data center energy consumption modeling as described in any one of claims 1 to 7 includes: The multi-dimensional data acquisition module is used to collect data on the power side of the data center, the load data on the computing power side, and the scenario-related data, and to perform time-series alignment processing on the collected multi-source data to form a time-series aligned multi-dimensional dataset. The dynamic coupling analysis module is used to extract the core coupling factors between power and computing power from the multi-dimensional dataset and construct a time-series correlation model with time as the dimension to characterize the dynamic coupling law between power-side parameters and computing power-side parameters. The model dynamic generation module is used to generate an energy consumption prediction model that adapts to the current operating scenario based on the time-series correlation model through a parameter adaptive optimization mechanism. The closed-loop iterative optimization module is used to compare the prediction results of the energy consumption prediction model with the actual energy consumption data of the data center, calculate the prediction error, optimize the parameters of the time series correlation model based on the prediction error, and re-evaluate the core coupling factor. The model output module is used to perform multi-scenario testing and verification on the optimized energy consumption prediction model and output the energy consumption modeling results.

9. A computer device, the computer device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by the processor, it implements the data center energy consumption modeling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data center energy consumption modeling method as described in any one of claims 1 to 7.

Images