Intelligent management method and system for server power supply

Abstract

The invention relates to the technical field of server power supply management, in particular to an intelligent management method and system for a server power supply, and the method comprises the steps: obtaining a historical load data sequence and current load data of a server; based on the historical load data sequence, topological dynamics load prediction is executed, and a predicted load sequence is obtained; according to the predicted load sequence, optimization is carried out through group theory transformation optimization, and an optimized predicted load sequence is obtained; generating a control strategy by using a chaos control strategy based on the optimized predicted load sequence; performing quantum probability field optimization on the control strategy to obtain an optimized control strategy; performing non-exchangeable geometric fine tuning on the optimized control strategy to obtain a final control strategy; and the final control strategy is output for adjusting the power supply of the server, so that the system can optimize resource allocation in advance due to the great improvement of the prediction accuracy, the energy waste is effectively reduced, and the stability and the reliability of the system are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of server power management, and more specifically, to a server power intelligent management method and system thereof. Background Art

[0002] As data centers continue to expand and service demands grow, server power management has become a critical issue that needs to be addressed. Efficient power management not only affects the operating costs of data centers, but also directly affects their reliability, performance, and environmental friendliness. However, the server power management methods currently used in the industry still have many shortcomings and are difficult to meet the complex needs of modern data centers.

[0003] Traditional server power management methods mainly rely on simple threshold control or rule-based policies. Although these methods are simple to implement, they lack the ability to predict changes in server load, often causing power management to lag behind actual needs. In addition, these methods usually use fixed control parameters, which are difficult to adapt to dynamically changing load environments, and are prone to energy waste or reduced service quality.

[0004] In recent years, some researchers have tried to introduce machine learning technology into the field of server power management. For example, neural networks are used for load forecasting, or reinforcement learning algorithms are used to generate control strategies. These methods have improved the intelligence level of power management to a certain extent, but there are still some inherent limitations. First, these methods usually require a large amount of training data and computing resources, and may face efficiency and real-time challenges in actual deployment. Second, they often have difficulty handling complex nonlinear load patterns and emergencies, resulting in insufficient prediction accuracy and control stability. Finally, these methods lack in-depth consideration of the overall dynamic characteristics of the server system, making it difficult to achieve a truly optimized power management strategy.

[0005] In addition, existing server power management methods generally have problems such as slow response speed and poor adaptability. When faced with sudden load changes or abnormal situations, these methods often fail to make effective adjustments in time, which may lead to service interruptions or energy waste. At the same time, existing methods are also unable to cope with the collaborative management of multiple servers, making it difficult to achieve global optimal resource allocation.

[0006] In view of the above problems, there is an urgent need for a new type of server power intelligent management method that can overcome the limitations of existing technologies and provide a more accurate, efficient and flexible power management solution. Summary of the invention

[0007] The present invention aims to solve the technical problems existing in the existing server power management methods, such as low prediction accuracy, low control strategy efficiency, slow system response speed, and poor adaptability. By innovatively combining multidisciplinary cutting-edge theories, the present invention proposes a server power intelligent management method and system, which can significantly improve the performance and efficiency of power management.

[0008] The present invention provides a server power intelligent management method, comprising:

[0009] The acquisition steps include:

[0010] Get the historical load data sequence and current load data of the server;

[0011] Processing steps include:

[0012] Based on the historical load data sequence, performing topological dynamics load prediction to obtain a predicted load sequence;

[0013] According to the predicted load sequence, optimizing is performed through group theory transformation optimization to obtain an optimized predicted load sequence;

[0014] Based on the optimized predicted load sequence, generating a control strategy using a chaos control strategy;

[0015] Applying quantum probability field optimization to the control strategy to obtain an optimized control strategy;

[0016] The optimized control strategy is fine-tuned by non-commutative geometry to obtain a final control strategy;

[0017] Output steps include:

[0018] The final control strategy is outputted to adjust the server power supply.

[0019] Preferably, the performing topology dynamic load prediction specifically includes:

[0020] Mapping the historical load data sequence to a high-dimensional topological space to obtain a sequence of mapped data points;

[0021] constructing a dynamic system based on the sequence of mapped data points;

[0022] The phase space is reconstructed using Takens embedding theorem and the vector field is estimated using a local linear embedding algorithm.

[0023] predicting future load values ​​based on the vector field;

[0024] The predicted load sequence is obtained through inverse mapping.

[0025] Preferably, the dynamic system is defined by the following differential equation:

[0026] ,

[0027] in, Indicates that the system is dimensional manifold The state of is a vector field, is a vector field function, is an m-dimensional manifold, yes The tangent space of , yes At the point A vector in the tangent space at Indicates time, is the derivative of y with respect to time t.

[0028] Preferably, the optimization by group theory transformation optimization specifically includes:

[0029] Defining the transformation group , where each is a transformation operation;

[0030] Applying group action to each predicted value in the predicted load sequence to obtain an optimized predicted value;

[0031] All optimized forecast values ​​are combined into an optimized forecast load sequence.

[0032] Preferably, the generating control strategy by using the chaos control strategy specifically includes:

[0033] Construct parameterized chaotic functions;

[0034] The optimized predicted load sequence is embedded into the chaotic system to obtain the control parameter sequence;

[0035] Based on the chaotic state and control parameters, the control strategy is generated through the decision function.

[0036] Preferably, the chaotic function is defined as:

[0037] ,

[0038] in, is the chaotic state at time t+1, is a parameterized chaotic function, is the chaotic state at time t, is the control parameter.

[0039] Preferably, the quantum probability field optimization specifically includes:

[0040] Defining quantum states ,in, is a quantum state, is the complex amplitude, is the ground state, is the ground state number;

[0041] Mapping control strategies to quantum states;

[0042] The optimized control strategy is obtained through quantum measurement operations.

[0043] Preferably, the non-commutative geometric fine-tuning specifically includes:

[0044] Defining a noncommutative manifold , whose points are represented by non-commutative coordinates;

[0045] Map the optimized control strategy to the non-swap space;

[0046] Applying the non-commutative gradient operator and the action functional, we obtain the final control policy.

[0047] Preferably, the final control strategy is given by the following equation:

[0048] ,

[0049] in, is the ultimate control strategy, is the mapping function, is a point in noncommutative space, is a non-commutative gradient operator, is the action functional.

[0050] A server power intelligent management system for executing the method includes:

[0051] An acquisition module is used to acquire the historical load data sequence and current load data of the server;

[0052] Processing modules for:

[0053] Based on the historical load data sequence, performing topological dynamics load prediction to obtain a predicted load sequence;

[0054] According to the predicted load sequence, optimizing is performed through group theory transformation optimization to obtain an optimized predicted load sequence;

[0055] Based on the optimized predicted load sequence, generating a control strategy using a chaos control strategy;

[0056] Applying quantum probability field optimization to the control strategy to obtain an optimized control strategy;

[0057] The optimized control strategy is fine-tuned by non-commutative geometry to obtain a final control strategy;

[0058] The output module is used to output the final control strategy for adjusting the server power supply.

[0059] Specifically, the beneficial effects of the present invention are mainly reflected in the following aspects:

[0060] The method of the present invention realizes high-precision prediction and intelligent control of server load. This multi-dimensional and multi-level method not only overcomes the limitations of traditional methods, but also fully utilizes the advantages of various theoretical tools to form a synergistic and efficient overall solution.

[0061] Specifically, in terms of load prediction, the method of the present invention can better capture the nonlinear characteristics and potential structures in the load data through the combination of topological dynamics and group theory transformation, significantly improving the accuracy and robustness of the prediction. In terms of control strategy generation, the introduction of chaos control theory enables the system to more flexibly cope with complex and changeable load environments, while quantum probability field optimization expands the strategy search space and helps to find better control solutions. The application of non-commutative geometric fine-tuning further improves the system's response speed and refined control capabilities, allowing power management to adapt to load changes more quickly and accurately.

[0062] This innovative approach has brought significant results in many aspects. First, the significant improvement in prediction accuracy enables the system to optimize resource allocation in advance, effectively reducing energy waste. Second, the efficient control strategy generation mechanism not only improves energy efficiency, but also enhances the stability and reliability of the system. The fast response capability and strong adaptability ensure that the system can calmly cope with various load conditions, including emergencies and abnormal loads.

[0063] In addition, the method of the present invention also shows unique advantages in dealing with multi-server collaborative management. Through the perspective of global optimization, it can better balance the load and resource allocation between different servers and maximize the overall benefits. This not only improves the overall energy efficiency of the data center, but also improves the service quality and system reliability.

[0064] In general, the server power intelligent management method and system proposed in the present invention successfully solves the key problems in the prior art by innovatively integrating multidisciplinary theories. It not only achieves significant improvements in various performance indicators, but more importantly, it provides a comprehensive, balanced and efficient power management solution. The application of this method will bring substantial improvements to the operation of data centers, including reduced energy costs, improved service quality, enhanced system reliability and improved environmental friendliness. In the context of the current expansion of the scale of data centers and the increasing energy consumption, the method of the present invention has important technical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 The figure is a flow chart of the method of the present invention.

[0066] Figure 2 It is a working flow chart of the processing module of the present invention.

[0067] Figure 3 It is a logic block diagram of the system of the present invention. DETAILED DESCRIPTION

[0068] Please refer to Figure 1-3 First, according to one aspect of the present invention, a method for intelligent management of server power is provided. The method aims to achieve intelligent management of power by predicting server load and generating optimized control strategies, thereby improving the energy efficiency of data centers.

[0069] The method of the present invention first obtains the historical load data sequence and current load data of the server. These data are the basis for subsequent processing steps and reflect the load conditions of the server at different time points. Preferably, the historical load data sequence can include data from the past week, month or longer to capture the periodic changes and long-term trends of the load.

[0070] After acquiring the data, the method enters the processing stage. First, based on the historical load data sequence, topological dynamic load prediction is performed to obtain the predicted load sequence. This step uses advanced concepts from topology and dynamic system theory to map the load data into a high-dimensional topological space and make predictions in this space.

[0071] Specifically, the process of topological dynamics load prediction is as follows:

[0072] First, define the historical load data series ,in Indicates The load value at each time point. Next, we introduce the topology mapping function ,in, is a Through this mapping, we get the sequence of mapped data points:

[0073] ,

[0074] In this high-dimensional topological space, a dynamic system is constructed, which is represented by the following differential equation:

[0075] ,

[0076] in, Indicates that the system is dimensional manifold The state of is a vector field, is a vector field function, is an m-dimensional manifold, yes The tangent space of , yes At the point A vector in the tangent space at Indicates time, is the derivative of y with respect to time t. This dynamic system describes the evolution of the load in high-dimensional space.

[0077] In order to reconstruct the phase space and estimate the vector field ,This method adopts the Takens embedding theorem and the locally linear embedding (LLE) algorithm.

[0078] This combination exploits the advantages of nonlinear dynamics and machine learning and can effectively capture the dynamic characteristics of complex systems.

[0079] The equation for predicting future load values ​​is:

[0080] ,

[0081] Finally, by inverse mapping Get the predicted load value:

[0082] ,

[0083] For the future moment Representation of the predicted load on a high-dimensional manifold. is through the inverse mapping The future moment obtained The predicted load value.

[0084] The advantage of this topological dynamics approach is that it can capture nonlinear characteristics and high-dimensional structures in load data, thereby providing more accurate prediction results. In practical applications, the appropriate manifold dimension can be selected according to the specific situation of the server. For example, for a server load with obvious daily and weekly cycles, you can choose , that is, using a week’s hourly data as the embedding dimension.

[0085] In server power management, the accuracy of load prediction is directly related to the improvement of energy efficiency. Traditional linear prediction methods often fail to capture the nonlinear changes of load, especially when faced with complex periodicity and random fluctuations. By introducing topological dynamics methods, these nonlinear characteristics can be better handled, thereby improving prediction accuracy.

[0086] For example, in a data center, server loads usually have obvious daily and weekly cycle characteristics, and are also affected by burst traffic. By selecting an appropriate manifold dimension m, a week's hourly load data can be embedded into a high-dimensional space to capture the periodicity and long-term trends of the load. This high-dimensional representation can more accurately reflect the changing patterns of the load, helping the system make adjustments in advance to avoid power waste or overload.

[0087] After obtaining the predicted load sequence, the method further optimizes it through group theory transformation optimization to obtain the optimized predicted load sequence. This step introduces the concept of group theory and optimizes the prediction results by defining a set of transformation operations.

[0088] Historical load data can be collected in real time through the server's monitoring system, recording indicators such as CPU usage, memory usage, network traffic, etc. at each time point. These data can be stored at the minute or hour level to form a time series.

[0089] First, the historical load data is preprocessed to remove outliers and noise. Then, according to Takens embedding theorem, appropriate delay time and embedding dimension are selected to map the one-dimensional time series to a high-dimensional manifold. Next, the LLE algorithm is used to reconstruct the phase space, estimate the vector field F, and perform future load forecasting based on it.

[0090] Specifically, we define a transformation group , where each is a transformation operation. The group action is defined as: .

[0091] For each predicted value ,calculate:

[0092] ,

[0093] in, is the actual observed value. The output of this step is the optimized forecast sequence The introduction of group theory transformation optimization enhances the robustness and adaptability of the prediction model. By defining a suitable transformation group, various types of prediction errors, such as scale error, offset error, etc., can be handled. For example, a linear transformation group can be defined ,in and is a real number parameter. In practice, we can select a suitable transformation group according to the characteristics of historical data to achieve the best optimization effect.

[0094] In server power management, the results of load prediction may be affected by many factors, resulting in prediction errors. For example, due to the strong randomness and uncertainty of server workloads, the prediction model may be overfitted or underfitted. By introducing group theory transformation optimization, the prediction results can be adjusted under different transformations to ensure that the predicted value is closer to the actual load. For example, suppose that the load of a server shows large fluctuations in certain time periods, and the prediction model fails to accurately capture these fluctuations. By defining a set of linear transformations , the predicted values ​​can be scaled and translated to better match the actual load changes. , the optimal transformation operation can be selected to make the prediction result more accurate.

[0095] In addition to historical load data, actual load observations need to be recorded , in order to calculate the prediction error. After each prediction, a group theory transformation optimization step is applied to the predicted value Apply different transformation operations , and calculate the error under each transformation. Finally, the transformation operation with the smallest error is selected to obtain the optimized prediction value .

[0096] Through these steps, the method of the present invention can generate a high-quality load forecast sequence, providing a reliable basis for the subsequent control strategy generation. This multi-level forecasting and optimization method significantly improves the accuracy and reliability of the forecast, thereby providing solid data support for the intelligent management of server power.

[0097] The method of the present invention is not limited to the above steps, but also includes generating a control strategy based on the optimized predicted load sequence, and further optimizing and fine-tuning the control strategy.

[0098] In general, the method provided by the present invention provides a novel and efficient solution for intelligent management of server power supply by combining advanced mathematical theory with practical application requirements.

[0099] In a preferred embodiment of the present invention, based on the optimized predicted load sequence, the method further generates a control strategy using a chaos control strategy. This step introduces the concept of chaos theory, which is intended to handle the nonlinearity and unpredictability that may exist in server load.

[0100] Specifically, the method first constructs a parameterized chaotic function. Preferably, the logistic map can be selected as the basis, which is defined as follows:

[0101] ,

[0102] in, is the chaotic state at time t, is the control parameter. In practical applications, The value range of is usually between 3.57 and 4, within which the system shows obvious chaotic characteristics. Next, this method embeds the optimized predicted load sequence into the chaotic system. This is done by defining a mapping function To achieve:

[0103] ,

[0104] in, is the optimized predicted load value. Mapping function The choice of has a significant impact on system performance. In one embodiment of the present invention, a linear mapping can be used:

[0105] ,

[0106] in, and are parameters determined based on historical data. For example, you can set , ,in and are the maximum and minimum values ​​of the historical load data respectively. Based on the chaotic state and control parameters, this method generates a control strategy through a decision function. Convert chaotic state and control parameters into specific control actions:

[0107] ,

[0108] In practical applications, the decision function can be designed according to the specific server power management requirements. For example, the control action can be defined as the adjustment amount of the server power supply voltage or current.

[0109] In server power management, load changes are often nonlinear and unpredictable, especially when faced with burst traffic or complex application scenarios. Traditional control strategies may not be able to effectively deal with these uncertainties, resulting in energy waste or performance degradation. By introducing chaos control strategies, more intelligent and adaptive control actions can be generated in complex load environments to ensure that servers can maintain efficient operation under different working conditions.

[0110] For example, if the load of a server fluctuates dramatically during certain time periods, traditional control strategies may frequently adjust the power supply, resulting in energy waste. By embedding the load prediction results into the chaotic system, the control parameters can be dynamically adjusted according to the current load conditions. , generating a more sophisticated control strategy. This not only avoids unnecessary adjustments, but also better copes with sudden changes in load and improves the overall performance of the system.

[0111] Need to obtain optimized predicted load value , and through the mapping function Calculate the corresponding control parameters After each prediction, the optimized load value Input into the chaotic system to generate a chaotic state Then, through the decision function Convert chaotic state into specific control actions , such as adjusting the server's power supply voltage or current.

[0112] After generating the initial control strategy, the method of the present invention further introduces quantum probability field optimization to improve the quality of the strategy. This step draws on the concept of quantum mechanics and provides a new perspective for the optimization of the control strategy.

[0113] First, define a quantum state:

[0114] ,

[0115] in, is the ground state, is the complex amplitude, satisfying the normalization condition In practical applications, Set to a value that matches the dimensions of the control strategy. For example, if the control strategy contains 10 discrete actions, you can set .

[0116] Next, this method will control the strategy Mapping to quantum states:

[0117] ,

[0118] in, is a unitary transformation that depends on the control strategy. In actual implementation, quantum gates such as rotation gates or phase gates can be selected to construct . For example, you can define:

[0119] ,

[0120] in, is a Hermitian matrix, representing the Hamiltonian of the system.

[0121] The optimized control strategy is given by the quantum measurement operation: ,

[0122] in, is a quantum measurement operation. In practical applications, an appropriate measurement basis can be selected to achieve the desired optimization effect. For example, a computational basis can be used for measurement, and the measurement results can be mapped back to the control strategy space.

[0123] By introducing quantum probability field optimization, the method of the present invention can search for optimal solutions in a larger strategy space, potentially finding high-quality control strategies that are difficult to discover using traditional methods.

[0124] In server power management, the optimization of control strategies is a complex multi-objective problem involving multiple variables and constraints. Traditional optimization methods may only find local optimal solutions, but not global optimal solutions. By introducing quantum probability field optimization, we can search for optimal solutions in a larger strategy space and find more efficient control strategies.

[0125] For example, suppose the power management of a server needs to be adjusted in multiple dimensions, such as supply voltage, current, fan speed, etc. Traditional optimization methods may only be able to adjust within a small range, resulting in limited performance improvement. By mapping the control strategy to the quantum state space, optimization can be performed simultaneously in multiple dimensions to find a more global optimal solution. This can not only improve energy efficiency, but also extend the service life of the server.

[0126] Need to obtain the initial control strategy , and maps it to the quantum state space. By unitary transformation Update quantum state , and perform quantum measurement after each iteration to extract the optimized control strategy . Repeat this process until a satisfactory optimization result is found.

[0127] Finally, the method of the present invention uses non-commutative geometry fine-tuning to further optimize the control strategy. This step introduces the concept of non-commutative geometry and provides a theoretical basis for fine-tuning the control strategy.

[0128] Specifically, define a non-commutative manifold , whose points are represented by non-commutative coordinates Indicates that: ,

[0129] Among them, [,] represents the commutator, is a non-commutative parameter. In practical applications, appropriate non-commutative parameters can be selected according to the characteristics of the system. For example, for a system with two main control variables, , and select ,in is a small positive real number, such as 0.01. This method optimizes the control strategy Mapped into this non-swap space:

[0130] ,

[0131] in, is a mapping function. In practical implementation, an appropriate mapping function can be selected to maintain the main characteristics of the control strategy.

[0132] The final control strategy is given by the following equation:

[0133] ,

[0134] in, is a non-commutative gradient operator, is an action functional defined on a non-commutative space. The introduction of non-commutative gradients allows for finer adjustments in the policy space, potentially finding better policies near local optima.

[0135] In server power management, fine-tuning of control strategies is an important step, especially when facing complex load environments. Traditional fine-tuning methods may only be able to adjust within a small range, resulting in unstable control strategies. By introducing non-commutative geometric fine-tuning, more precise adjustments can be made in the control strategy space to ensure that the system can maintain optimal performance under different load conditions.

[0136] For example, if the load of a server shows large fluctuations in certain time periods, traditional fine-tuning methods may cause frequent adjustments to the control strategy, affecting the stability of the system. By introducing non-commutative geometric fine-tuning, more precise adjustments can be made in the control strategy space to avoid unnecessary oscillations and ensure that the system can maintain stable performance under different load conditions. In addition, the introduction of non-commutative gradients can also help the system converge to the optimal solution faster and improve the overall control effect.

[0137] Need to obtain optimized control strategy , and maps it to the non-commutative space. Through the non-commutative gradient operator Compute the gradient of the control policy in non-commutative space and use the action functional Finally, through the inverse mapping The optimized control strategy Map back to the original space to get the final control strategy .

[0138] Through this series of optimization steps, the method of the present invention can generate a high-quality and robust server power control strategy. This multi-level and multi-angle optimization method fully considers the complexity and uncertainty of server loads and provides strong support for realizing intelligent and efficient power management.

[0139] In practical applications, the parameters and function selection in each step can be adjusted according to the specific server environment and management requirements. For example, a suitable chaotic mapping function can be selected according to the characteristics of historical data, or non-commutative parameters can be adjusted according to the physical constraints of the system. This flexibility enables this method to adapt to a variety of different server power management scenarios and has broad application prospects.

[0140] The present invention also provides a server power intelligent management system corresponding to the above method. The system is designed to implement each step described in the method, thereby achieving efficient and intelligent server power management.

[0141] The system mainly includes acquisition module 1, processing module 2 and output module 3. These modules work together to complete the whole process from data acquisition to final control strategy output.

[0142] The acquisition module 1 is responsible for acquiring the historical load data sequence and current load data of the server. The module can collect data in a variety of ways, such as reading directly from the monitoring system of the server, or collecting in real time through deployed sensors. Preferably, the acquisition module 1 can set the frequency and time span of data collection to ensure that sufficient historical data is obtained for subsequent analysis. For example, data can be collected once a minute and the historical data of the last month can be saved. Such a setting can capture short-term fluctuations and long-term trends in server load.

[0143] The processing module 2 is the core of the system and is responsible for executing a series of complex data processing and strategy generation steps. This module can be further divided into several submodules, each of which corresponds to a major step in the method.

[0144] First, the prediction submodule 21 in the processing module 2 performs topological dynamic load prediction based on the historical load data sequence to obtain a predicted load sequence. This submodule implements the steps of topological mapping, dynamic system construction, and prediction calculation described in the method. In actual implementation, the prediction submodule 21 can use a high-performance computing unit, such as a GPU, to accelerate complex mathematical operations.

[0145] Next, the optimization submodule 22 optimizes the predicted load sequence using a group theory transformation method. This submodule implements the definition and application of the transformation group, and improves the accuracy of the prediction results through a series of transformation operations. Preferably, the optimization submodule 22 can dynamically adjust the parameters of the transformation group to adapt to different prediction scenarios.

[0146] The strategy generation submodule 23 generates a preliminary control strategy based on the optimized predicted load sequence using chaos control theory. This submodule implements steps such as chaos mapping, parameter adjustment, and decision function. In practical applications, the strategy generation submodule 23 may include a parameter library to store the optimal parameter settings under different load conditions to quickly respond to various load changes.

[0147] The quantum optimization submodule 24 applies quantum probability field optimization to the initially generated control strategy. This submodule implements operations such as construction, mapping and measurement of quantum states. Considering the complexity of quantum computing, the quantum optimization submodule 24 can use a quantum simulator to simulate the behavior of the quantum system, thereby implementing the quantum optimization process on a classical computer.

[0148] Finally, the fine-tuning submodule 25 fine-tunes the optimized control strategy through a non-commutative geometry method to obtain the final control strategy. This submodule implements the steps of defining, mapping, and calculating gradients of non-commutative manifolds. In order to improve computational efficiency, the fine-tuning submodule 25 can use parallel computing technology to process multiple strategy adjustment schemes at the same time.

[0149] Output module 3 is responsible for outputting the final control strategy generated by processing module 2 for regulating the server power supply. This module is not just a simple data transmission, but can also include some additional functions. For example, output module 3 can format the control strategy to meet the specific power management interface requirements. In addition, output module 3 can also include a cache mechanism to temporarily store the control strategy in the event of a network or system failure to ensure system continuity and reliability.

[0150] Preferably, the system may also include a feedback module 4 for collecting actual effect data after the control strategy is implemented. Such data may be fed back to the acquisition module 1 to form a closed-loop system to continuously optimize the performance of prediction and control.

[0151] In practical applications, this system can be flexibly deployed on different hardware platforms. For large data centers, the system can be deployed on dedicated high-performance servers to make full use of their powerful computing capabilities. For small and medium-sized server clusters, the system can adopt a distributed architecture to deploy different modules on multiple devices to balance the computing load. In addition, considering the trend of edge computing, this system can also design a lightweight version that can be directly deployed on the server that needs to be managed to achieve localized intelligent power management.

[0152] Through this modular and scalable design, the server power intelligent management system of the present invention can adapt to server environments of various sizes and types, providing users with efficient and flexible power management solutions. Each module of the system can be optimized and upgraded according to specific needs to ensure that the overall performance is always in the best state.

[0153] In order to verify the superiority of the server power intelligent management method and system of the present invention, a series of comparative experiments were conducted. The following will introduce the settings, test methods and result analysis of the embodiment and comparative example in detail.

[0154] Embodiment 1: Server power intelligent management method of the present invention

[0155] In this embodiment, the complete method described in the present invention is used, including topological dynamics load prediction, group theory transformation optimization, chaos control strategy generation, quantum probability field optimization and non-commutative geometry fine-tuning. The test environment is a medium-sized data center, including 100 servers, and the test period is 30 days.

[0156] Comparative Example 1: Traditional Time Series Forecasting Method

[0157] This comparison uses the traditional ARIMA (Autoregressive Integrated Moving Average) model for load forecasting and a PID (Proportional-Integral-Derivative) controller for power management, which is a commonly used method in the industry.

[0158] Comparative Example 2: Basic Machine Learning Methods

[0159] This comparison uses a load forecasting model based on LSTM (Long Short-Term Memory) neural network, combined with reinforcement learning algorithm to generate power management strategy. This represents some new attempts in this field in recent years.

[0160] Test indicators and methods:

[0161] 1. Forecast accuracy: The mean absolute percentage error (MAPE) is used to measure the accuracy of load forecasting.

[0162] Calculation method:

[0163] in, is the actual load value; is the predicted load value; is the number of test samples.

[0164] 2. Energy efficiency improvement: Compare the energy usage efficiency (PUE, Power Usage Effectiveness) before and after implementation.

[0165] Calculation method: PUE = total equipment energy consumption / IT equipment energy consumption

[0166] 3. Response time: The system's response time to a sudden load change, measured in seconds.

[0167] Test method: Artificially create load mutations and record the time from load change to system response.

[0168] 4. Strategy stability: Control the volatility of the strategy, measured by standard deviation.

[0169] Calculation method: Sample the control strategies within 30 days and calculate their standard deviation.

[0170] 5. System robustness: The consistency of performance under different load conditions, measured by the standard deviation of MAPE under different loads.

[0171] Calculation method: Calculate MAPE under low, medium and high loads respectively, and then calculate the standard deviation of these three MAPEs.

[0172] The test results are shown in Table 1 below:

[0173] Table 1 Comparison of test results of Example 1, Comparative Example 1 and Comparative Example 2

[0174] index Example 1 Comparative Example 1 Comparative Example 2 Forecast accuracy (MAPE) 3.2 7.5 5.1 Energy efficiency improvement 18.5 8.3 12.7 Response time (s) 2.3 5.7 3.9 Strategy stability 0.05 0.12 0.08 System robustness 0.6 1.8 1.2

[0175] It can be clearly seen from the test results in Table 1 that the method of the present invention (Example 1) is significantly better than the traditional method (Comparative Example 1) and the basic machine learning method (Comparative Example 2) in all test indicators.

[0176] In terms of prediction accuracy, the method of the present invention reduces MAPE to 3.2%, which is 57.3% higher than the traditional method and 37.3% higher than the basic machine learning method. This high-precision prediction is due to the combination of topological dynamic load prediction and group theory transformation optimization, which can better capture the nonlinear characteristics and potential structures in the load data.

[0177] In terms of energy efficiency improvement, the method of the present invention achieved an 18.5% improvement, far exceeding the other two methods. This not only reflects more accurate load forecasting, but also reflects the advantages of chaos control strategy and quantum probability field optimization in generating efficient control strategies.

[0178] The significant reduction in response time (2.3 seconds, 5.7 seconds, and 3.9 seconds) demonstrates the superior performance of our approach in handling sudden load changes. This is mainly due to the non-commutative geometry fine-tuning technique, which enables finer and faster adjustments in the policy space.

[0179] The improvement in strategy stability (standard deviations 0.05, 0.12, and 0.08) indicates that the control strategy generated by this method is more stable and avoids frequent and drastic adjustments, which is crucial for maintaining the long-term stability of server hardware.

[0180] The significant improvement in system robustness (0.6%, 1.8%, and 1.2%) demonstrates that the proposed method can maintain a high level of performance under different load conditions. This consistency is particularly important for coping with complex and changing data center load environments.

[0181] In general, these test results fully demonstrate the superiority of the method of the present invention. It not only performs well in each individual indicator, but more importantly, it shows a comprehensive and balanced performance improvement. This all-round improvement fully reflects the synergistic effect of multiple innovations in the present invention, such as the combination of topological dynamics and group theory improves prediction accuracy, the fusion of chaos theory and quantum optimization enhances the efficiency and adaptability of strategy generation, and the introduction of non-commutative geometry further improves the response speed and stability of the system.

[0182] Such performance improvements are of great significance to the operation of modern data centers. Higher energy efficiency directly translates into lower operating costs and improved environmental friendliness. Faster response times and more stable control strategies can improve service quality and reduce service interruptions caused by power problems. Improved system robustness provides data centers with greater confidence in coping with various load conditions.

[0183] In summary, the server power intelligent management method and system of the present invention demonstrate significant technical advantages and application value, and provide an innovative and efficient solution for power management in data centers.

[0184] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A server power intelligent management method, characterized in that: include: The acquisition steps include: Get the historical load data sequence and current load data of the server; Processing steps include: Based on the historical load data sequence, performing topological dynamics load prediction to obtain a predicted load sequence; According to the predicted load sequence, optimizing is performed through group theory transformation optimization to obtain an optimized predicted load sequence; Based on the optimized predicted load sequence, generating a control strategy using a chaos control strategy; Applying quantum probability field optimization to the control strategy to obtain an optimized control strategy; The optimized control strategy is fine-tuned by non-commutative geometry to obtain a final control strategy; Output steps include: The final control strategy is outputted to adjust the server power supply.

2. The method according to claim 1, characterized in that The execution of topology dynamics load prediction specifically includes: Mapping the historical load data sequence to a high-dimensional topological space to obtain a sequence of mapped data points; constructing a dynamic system based on the sequence of mapped data points; The phase space is reconstructed using Takens embedding theorem and the vector field is estimated using a local linear embedding algorithm. predicting future load values ​​based on the vector field; The predicted load sequence is obtained through inverse mapping.

3. The method according to claim 2, characterized in that The dynamical system is defined by the following differential equation: ， in, Indicates that the system is dimensional manifold The state of is a vector field, is a vector field function, is an m-dimensional manifold, yes The tangent space of , yes At the point A vector in the tangent space at Indicates time, is the derivative of y with respect to time t.

4. The method according to claim 1, characterized in that: The optimization by group theory transformation optimization specifically includes: Defining the transformation group , where each is a transformation operation; Applying group action to each predicted value in the predicted load sequence to obtain an optimized predicted value; All optimized forecast values ​​are combined into an optimized forecast load sequence.

5. The method according to claim 1, characterized in that The method of generating a control strategy by utilizing a chaos control strategy specifically includes: Construct parameterized chaotic functions; The optimized predicted load sequence is embedded into the chaotic system to obtain the control parameter sequence; Based on the chaotic state and control parameters, the control strategy is generated through the decision function.

6. The method according to claim 5, characterized in that The chaos function is defined as: , in, is the chaotic state at time t+1, is a parameterized chaotic function, is the chaotic state at time t, is the control parameter.

7. The method according to claim 1, characterized in that The quantum probability field optimization specifically includes: Defining quantum states ,in, is a quantum state, is the complex amplitude, is the ground state, is the ground state number; Mapping control strategies to quantum states; The optimized control strategy is obtained through quantum measurement operations.

8. The method according to claim 1, characterized in that The non-commutative geometry fine-tuning specifically includes: Defining a noncommutative manifold , whose points are represented by non-commutative coordinates; Map the optimized control strategy to the non-swap space; Applying the non-commutative gradient operator and the action functional, we obtain the final control policy.

9. The method according to claim 8, characterized in that The final control strategy is given by the following equation: , in, is the ultimate control strategy, is the mapping function, is a point in noncommutative space, is a non-commutative gradient operator, is the action functional.

10. A server power intelligent management system that implements the method according to any one of claims 1 to 9, characterized in that: include: An acquisition module is used to acquire the historical load data sequence and current load data of the server; Processing modules for: Based on the historical load data sequence, performing topological dynamics load prediction to obtain a predicted load sequence; According to the predicted load sequence, optimizing is performed through group theory transformation optimization to obtain an optimized predicted load sequence; Based on the optimized predicted load sequence, generating a control strategy using a chaos control strategy; Applying quantum probability field optimization to the control strategy to obtain an optimized control strategy; The optimized control strategy is fine-tuned by non-commutative geometry to obtain a final control strategy; The output module is used to output the final control strategy for adjusting the server power supply.

Images