Server cluster intelligent dynamic heat dissipation regulation method and system

Abstract

The application discloses a server cluster intelligent dynamic heat dissipation regulation method and system, relates to the technical field of data center heat dissipation control, and discloses the server cluster intelligent dynamic heat dissipation regulation method and system, which obtains server cluster operation state and power supply system state data, calculates heat buffer capacity and impedance characteristics, generates optimized power adjustment instructions and offsets heat dissipation equipment adjustment time points, realizes heat dissipation regulation while reducing power mutation, can effectively inhibit the impact of power mutation on the power supply system, improves data center power stability, and reduces resonance risk.

Description

Technical Field

[0001] This application relates to the field of data center heat dissipation control technology, and in particular to a method and system for intelligent dynamic heat dissipation control of server clusters. Background Technology

[0002] With the rapid evolution of cloud computing and artificial intelligence technologies, the computing density of data center server clusters continues to rise, and the energy consumption of cooling systems has exceeded 40% of the total energy consumption. Current mainstream cooling control strategies generally adopt a temperature threshold-based feedback control mechanism. When the server temperature exceeds a preset threshold, cooling devices such as fans or liquid cooling pumps immediately start to increase cooling power. This control method has technical flaws: in scenarios with sudden changes in server load, the cooling system is forced to implement rapid power adjustments, leading to a sharp increase in the power change rate of the cooling devices, with measured peak values ​​reaching 200 watts per second. This drastic power fluctuation strongly impacts the power supply system, causing voltage fluctuations and even triggering protection mechanisms, seriously threatening the power stability of the data center. Especially when multiple cooling devices adjust their power simultaneously, the cumulative effect of their power changes further amplifies the interference, inducing the risk of power supply system resonance and weakening operational reliability. Existing technical solutions only focus on basic cooling functions and fail to consider the stability requirements of the power supply system. They neither effectively utilize the thermal inertia characteristics of server components to buffer changes in heat load nor deeply analyze the impedance frequency characteristics of the power supply system to avoid resonance regions, resulting in the cooling control process facing long-term technical bottlenecks of severe power fluctuations and significant power supply risks.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method and system for intelligent dynamic heat dissipation control of server clusters, which aims to improve the power stability of data centers and reduce the risk of resonance.

[0005] To achieve the above objectives, this application proposes a method for intelligent dynamic heat dissipation control of a server cluster, the method comprising: Obtain the server cluster's operating status data and power supply system status data, including server temperature data, computing load data, and heat dissipation equipment operating data; Based on the operational status data, the thermal buffer capacity data of the server cluster is calculated using a pre-trained thermal inertia analysis model; the thermal buffer capacity data is used to represent the additional thermal load that the server can withstand before the temperature reaches a preset safety threshold. Based on the power supply system status data, the impedance frequency characteristics of the power supply system are analyzed to generate power supply system impedance characteristic data. Based on the thermal buffer capacity data and the power supply system impedance characteristic data, a first power adjustment command data is generated through a power smoothing optimization algorithm; the first power adjustment command data is used to control the power adjustment process to avoid the resonant frequency region of the power supply system. Based on the first power adjustment command data and the heat dissipation device operation data, a second power adjustment command sequence data for multiple heat dissipation devices is generated through a phase-shifted cooperative scheduling algorithm; the second power adjustment command sequence data is used to control the power adjustment time points of multiple heat dissipation devices to be staggered. The second power adjustment command sequence data is sent to the corresponding heat dissipation device for execution, thereby completing the heat dissipation regulation.

[0006] In one embodiment, the step of calculating the hot buffer capacity data of the server cluster based on the operational status data using a pre-trained thermal inertia analysis model includes: The current thermal capacity status data of each server component is obtained from the server temperature data through thermal capacity characteristic analysis; The load change trend data within a preset first time period is predicted from the calculated load data, and the load change trend data includes the load growth rate and load fluctuation characteristics. Based on the current thermal capacity status data and the load change trend data, the remaining time for each server component to reach the preset safe temperature threshold is calculated using a thermal buffer time window algorithm. Based on the remaining time data and the preset safe temperature threshold data, the thermal buffer capacity quantization data of each server component is obtained through the thermal buffer quantization algorithm. The thermal buffer capacity quantization data is used to represent the tolerable thermal load increment and the corresponding buffer time window. By integrating the quantified hot buffer capacity data of multiple server components, the overall hot buffer capacity data of the server cluster is obtained.

[0007] In one embodiment, the step of calculating the remaining time for each server component to reach a preset safe temperature threshold using a thermal buffer time window algorithm, based on the current thermal capacity status data and the load change trend data, includes: Calculate the heat capacity data of the server component at the current temperature based on the current heat capacity status data; Predict the heat load increment data at different future time points based on the load change trend data; A heat balance equation is established based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server components. Based on the temperature rise rate data and the current temperature data of the server components, predict the time point when the temperature of the server components will reach the preset safe temperature threshold. Calculate the time difference between the current time point and the data at the specified time point to obtain the remaining time data.

[0008] In one embodiment, the step of establishing a heat balance equation based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server components includes: Extract the material thermal capacity parameters, geometric dimension parameters, and current temperature distribution data of the server components from the current thermal capacity status data; Calculate the heat generation rate data inside the server components based on the heat load increment data; Calculate the thermal capacity coefficient of the server component based on the material heat capacity parameter data and the geometric dimension parameter data; The heat generation rate data, the heat capacity coefficient, and the current temperature distribution data are input into the heat conduction equation to solve for the temperature change response data of the server components. Extract the temperature change curve data over time from the temperature change response data, and calculate the temperature rise rate data.

[0009] In one embodiment, the step of analyzing the impedance frequency characteristics of the power supply system and generating power supply system impedance characteristic data based on the power supply system state data includes: Calculate the basic impedance spectrum data of the power supply system based on the voltage and current data in the power supply system status data; Multiple test power disturbance signals of different frequencies are injected into the power supply system, while the voltage response signal data of the power supply system is monitored and recorded. Based on the test power disturbance signal and the voltage response signal data, the actual impedance values ​​of the power supply system at each frequency point are obtained through the impedance frequency scanning algorithm. Analyze the actual impedance values ​​to identify frequency points where the impedance values ​​are lower than a preset first threshold, and obtain strong resonant frequency point data; Analyze the actual impedance data to identify frequency ranges where the impedance value changes more than a preset second threshold within a preset frequency range, and obtain high-sensitivity frequency range data; By integrating the strong resonant frequency data, the high-sensitivity frequency band data, and the basic impedance spectrum data, the impedance characteristic data of the power supply system is obtained.

[0010] In one embodiment, the step of analyzing the actual impedance data, identifying frequency points where the impedance value is lower than a preset first threshold, and obtaining strong resonant frequency point data includes: Frequency domain analysis was performed on the actual impedance numerical data to obtain impedance-frequency characteristic curve data; On the impedance-frequency characteristic curve data, frequency points with impedance amplitudes lower than a preset amplitude threshold are identified to obtain frequency point set data; From the set of frequency points, frequency points whose phase angles are close to the preset resonant phase angle are selected to obtain preliminary resonant frequency point data; The stability of the preliminary resonant frequency data is verified by means of re-injecting the test signal at the corresponding frequency and monitoring the voltage amplification factor. Based on the stability verification results, the frequency points that passed the verification were determined as the final strong resonant frequency point data.

[0011] In one embodiment, the step of generating first power adjustment command data based on the hot buffer capacity data and the power supply system impedance characteristic data using a power smoothing optimization algorithm includes: Extract hot buffer time window data and hot buffer capacity limit data from the hot buffer capacity data; Extract the resonant frequency range that must be avoided and the allowable adjustment frequency range from the impedance characteristic data of the power supply system; The time constraint data for power adjustment is determined based on the hot buffer time window data and the hot buffer capacity limit data. The frequency constraint data for power adjustment is determined based on the resonant frequency range data and the allowable adjustment frequency range data. The time constraint data and the frequency constraint data are input into the dual constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy the thermal buffer constraint and impedance constraint. The scheme with the best smoothness is selected from the set of candidate power adjustment schemes to obtain the first power adjustment command data.

[0012] In one embodiment, the step of inputting the time constraint data and the frequency constraint data into a dual-constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy the heat buffer constraint and impedance constraint includes: A hot buffer constraint function is constructed based on the time constraint data. The hot buffer constraint function is used to constrain the power adjustment process to be completed within the hot buffer time window and not exceed the hot buffer capacity limit. An impedance constraint function is constructed based on the frequency constraint data. The impedance constraint function is used to constrain the frequency components of power adjustment to avoid the resonant frequency range and to be mainly distributed within the allowable adjustment frequency range. A mathematical model for the optimization problem is established based on the heat buffer constraint function and the impedance constraint function. The model takes the power adjustment smoothness as the optimization objective and the heat buffer constraint function and the impedance constraint function as the constraint conditions. The mathematical model data of the optimization problem is solved using a sequential quadratic programming algorithm to obtain preliminary optimization solution data that satisfies the dual constraints; The feasibility of the preliminary optimized solution data is verified and the boundary conditions are satisfied again to obtain the set of candidate power adjustment schemes.

[0013] In one embodiment, the step of generating a second power adjustment command sequence data for multiple heat dissipation devices based on the first power adjustment command data and the heat dissipation device operation data using a phase-shifting cooperative scheduling algorithm includes: Extract dynamic response characteristic data of multiple heat dissipation devices from the heat dissipation device operation data. The dynamic response characteristic data includes device start-up delay time, power rise time, steady state hold time, and shutdown decay time. The total power adjustment task data is obtained by decomposing the first power adjustment instruction data. The total power adjustment task data includes the overall power adjustment range, adjustment start time, and adjustment duration. Based on the dynamic response characteristic data and the total power adjustment task data, sub-task data suitable for the characteristics of each heat dissipation device is obtained through a response characteristic matching algorithm; Based on the time parameters in the dynamic response characteristic data and the sub-task data, the optimal execution time point data of each sub-task is calculated by the timing misalignment optimization algorithm. The optimal execution time point data is used to control the peak power adjustment times of different heat dissipation devices to be staggered. The subtask data and the optimal execution time point data are combined to obtain the second power adjustment instruction sequence data.

[0014] Furthermore, to achieve the above objectives, this application also proposes a server cluster intelligent dynamic heat dissipation control system, which includes: a memory, a processor, and a server cluster intelligent dynamic heat dissipation control program stored in the memory and executable on the processor. The server cluster intelligent dynamic heat dissipation control program is configured to implement the steps of the server cluster intelligent dynamic heat dissipation control method.

[0015] The intelligent dynamic heat dissipation control method and system for server clusters proposed in this application obtains server cluster operating status and power supply system status data, calculates thermal buffer capacity and impedance characteristics, generates optimized power adjustment commands and staggers the adjustment time of heat dissipation equipment, thereby achieving heat dissipation control while reducing power surges. This can effectively suppress the impact of power surges on the power supply system, improve the power stability of data centers, and reduce resonance risks. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the intelligent dynamic heat dissipation control method for server clusters in this application. Figure 2 This is a schematic diagram of a structural embodiment of the intelligent dynamic heat dissipation control system for server clusters in this application.

[0019] Explanation of icon numbers: 10. Memory; 20. Processor.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] In existing technologies, server cluster heat dissipation control methods primarily rely on temperature threshold feedback mechanisms. When server load changes drastically, the cooling system needs to respond quickly to cope with heat load fluctuations, leading to an increased power change rate of the cooling equipment. This power surge can impact the power supply system, causing voltage fluctuations and even triggering the power supply system's protection mechanisms, thereby affecting the power stability of the entire data center. Furthermore, when multiple cooling devices adjust their power simultaneously, their power changes may overlap, amplifying the interference to the power supply system, creating a risk of power supply system resonance, and reducing the reliability of data center operation. Existing technologies fail to fully utilize the thermal inertia characteristics of the servers themselves and do not consider the impedance frequency characteristics of the power supply system, resulting in technical bottlenecks in the heat dissipation control process, such as large power fluctuations and poor power supply stability.

[0024] Based on this, embodiments of this application provide a method for intelligent dynamic heat dissipation control of a server cluster, referring to... Figure 1 The intelligent dynamic heat dissipation control method for server clusters includes steps S100 to S600, wherein: Step S100: Obtain the operating status data of the server cluster and the power supply system status data. The operating status data includes server temperature data, computing load data, and heat dissipation equipment operating data. Step S200: Based on the operating status data, calculate the thermal buffer capacity data of the server cluster using a pre-trained thermal inertia analysis model; the thermal buffer capacity data is used to represent the additional thermal load that the server can withstand before the temperature reaches a preset safety threshold. Step S300: Based on the power supply system status data, analyze the impedance frequency characteristics of the power supply system and generate power supply system impedance characteristic data; Step S400: Based on the thermal buffer capacity data and the power supply system impedance characteristic data, a first power adjustment command data is generated through a power smoothing optimization algorithm; the first power adjustment command data is used to control the power adjustment process to avoid the resonant frequency region of the power supply system. Step S500: Based on the first power adjustment command data and the heat dissipation device operation data, a second power adjustment command sequence data for multiple heat dissipation devices is generated through a phase-shifted cooperative scheduling algorithm; the second power adjustment command sequence data is used to control the power adjustment time points of multiple heat dissipation devices to be staggered. In step S600, the second power adjustment command sequence data is sent to the corresponding heat dissipation device for execution to complete the heat dissipation regulation.

[0025] In this embodiment, operational status data refers to real-time or historical data generated during the operation of the server cluster, including server temperature data, computing load data, and cooling equipment operation data. Server temperature data reflects the real-time thermal state of various components within the server; computing load data indicates the current processing workload and resource usage; and cooling equipment operation data records operating parameters of cooling devices such as fan speed and liquid cooling pump power. This data forms the basis for assessing the server cluster's heat load and cooling requirements. Power supply system status data refers to real-time or historical data generated during the operation of the data center power supply system, including electrical parameters such as voltage, current, and power factor. This data is used to analyze the health and stability of the power supply system, particularly its response characteristics to power changes. The thermal inertia analysis model is a pre-trained mathematical model that can predict the temperature change trend of the server under different heat loads and quantify its thermal buffering capacity based on the server's operational status data. This model considers factors such as the heat capacity and thermal conductivity of server materials and the efficiency of cooling equipment.

[0026] In this embodiment, the thermal buffer capacity data refers to the total amount of additional heat that the server can absorb before its internal temperature rises to a preset safety threshold at the current temperature. This data quantifies the server's ability to withstand thermal load fluctuations in a short period, providing time margin and power adjustment space for heat dissipation control. The power supply system impedance frequency characteristics refer to the impedance response of the power supply system to changes in current or power at different frequencies. At certain specific frequencies, the power supply system may exhibit lower impedance, making it prone to resonance, thereby amplifying power fluctuations and affecting power supply stability. The power supply system impedance characteristic data refers to a dataset describing the impedance behavior of the power supply system at different frequencies, including curves of impedance amplitude and phase angle as a function of frequency, as well as identified strong resonant frequency points and highly sensitive frequency bands. This data is a key basis for evaluating the power supply system's sensitivity to power adjustment.

[0027] In this embodiment, the power smoothing optimization algorithm refers to an optimization algorithm for generating power adjustment instructions. Its goal is to minimize the power change rate and fluctuation amplitude during the power adjustment process while meeting heat dissipation requirements, and to ensure that the frequency components of the power adjustment avoid the resonant frequency region of the power supply system. The first power adjustment instruction data refers to a series of instructions generated by the power smoothing optimization algorithm, which guides the heat dissipation equipment in power adjustment. This instruction data not only specifies the magnitude of the power adjustment but also specifically considers the timing and frequency components of the adjustment to avoid impacting the power supply system. The phase-shifting cooperative scheduling algorithm refers to a scheduling algorithm for coordinating the power adjustments of multiple heat dissipation devices. Its goal is to stagger the peak times of power adjustments for different heat dissipation devices, thereby avoiding the superposition of power changes from multiple devices and further smoothing the overall power fluctuation. The second power adjustment instruction sequence data refers to a set of specific power adjustment instructions for each heat dissipation device generated by the phase-shifting cooperative scheduling algorithm. This sequence data specifies in detail the power adjustment magnitude, start time, duration, and adjustment rate for each heat dissipation device to achieve coordination and smoothness in overall heat dissipation control.

[0028] In this embodiment, the server cluster's operational status data and power supply system status data can be obtained in various ways. For example, operational status data can be obtained by manually periodically inspecting and recording server thermometer readings, calculating load statistics through system logs, and manually observing the operation indicator lights of the cooling devices. Power supply system status data can be obtained by manually reading electricity meters or measuring voltage and current using a handheld oscilloscope. As another implementation, operational status data can be obtained by deploying basic temperature sensors, load counters, and cooling device status indicators. Power supply system status data can be collected in real time using basic voltage and current sensors.

[0029] In this embodiment, based on the acquired operational status data, the thermal buffer capacity of the server cluster can be calculated using a pre-trained thermal inertia analysis model. For example, a static thermal capacity lookup table can be pre-established, which directly estimates the thermal buffer capacity based on the server model and the current ambient temperature. Alternatively, offline testing of the servers can be performed, recording their temperature rise curves under different loads, and then an empirical formula can be obtained through simple curve fitting for online calculation of the thermal buffer capacity. This thermal buffer capacity data represents the additional thermal load that the server can withstand before the temperature reaches a preset safety threshold.

[0030] In this embodiment, based on the acquired power supply system status data, the impedance frequency characteristics of the power supply system can be analyzed, and power supply system impedance characteristic data can be generated. For example, the fundamental impedance spectrum can be directly obtained by performing a Fourier transform on the real-time voltage and current data of the power supply system, which can then be used as impedance characteristic data. Alternatively, simple impedance measurements can be performed at several preset fixed frequency points, and then the impedance frequency characteristics of the power supply system at other frequency points can be approximated through linear interpolation or polynomial fitting.

[0031] In this embodiment, based on the thermal buffer capacity data and the power supply system impedance characteristic data, the first power adjustment command data can be generated using a power smoothing optimization algorithm. This first power adjustment command data is used to control the power adjustment process to avoid the resonant frequency region of the power supply system. For example, a rule-based decision system can be preset. When the thermal buffer capacity is lower than a certain threshold, the system will directly restrict the power adjustment command to the known non-resonant frequency region according to preset rules. Alternatively, a multi-dimensional lookup table can be constructed. Inputting the current thermal buffer capacity and the power supply system impedance characteristics, the corresponding power adjustment command can be directly retrieved from the table and output.

[0032] In this embodiment, based on the first power adjustment command data and the heat dissipation device operation data, the second power adjustment command sequence data for multiple heat dissipation devices can be generated through a phase-shifted cooperative scheduling algorithm. This second power adjustment command sequence data is used to control the power adjustment times of multiple heat dissipation devices to be staggered. For example, a simple sequential scheduling strategy can be adopted, where the power adjustment of one heat dissipation device is completed before the power adjustment of the next heat dissipation device is started. Alternatively, a fixed start-up delay time can be preset for each heat dissipation device, ensuring that their peak power adjustment times are staggered. Finally, the generated second power adjustment command sequence data is sent to the corresponding heat dissipation device for execution, thereby completing the heat dissipation control. For example, the command data can be manually input into the local controller of each heat dissipation device. Alternatively, the command data can be sent to each heat dissipation device via a wired communication interface, such as RS485 or an Ethernet interface.

[0033] In this embodiment, the intelligent dynamic heat dissipation control method for server clusters effectively avoids the impact of heat dissipation power adjustments on the power supply system by comprehensively considering the thermal inertia of the server and the impedance frequency characteristics of the power supply system. This method utilizes the server's thermal buffering capability to provide time margin for heat dissipation power adjustments and, combined with the impedance characteristics of the power supply system, plans the power adjustment path to avoid resonant regions. Furthermore, through phase-shifting collaborative scheduling, it further smooths the total power fluctuations caused by the superposition of power adjustments from multiple heat dissipation devices. Thus, while ensuring efficient heat dissipation of the server cluster, it improves the stability and operational reliability of the data center power supply system and reduces the risks caused by power fluctuations.

[0034] In one feasible implementation, the step of calculating the thermal buffer capacity data of a server cluster using a pre-trained thermal inertia analysis model includes: obtaining the current thermal capacity status data of each server component from the server temperature data through thermal capacity characteristic analysis; predicting the load change trend data within a preset first time period from the calculated load data, wherein the load change trend data includes the load growth rate and load fluctuation characteristics; calculating the remaining time data for each server component to reach a preset safe temperature threshold using a thermal buffer time window algorithm based on the current thermal capacity status data and the load change trend data; obtaining the thermal buffer capacity quantized data of each server component using a thermal buffer quantization algorithm based on the remaining time data and the preset safe temperature threshold data, wherein the thermal buffer capacity quantized data represents the tolerable thermal load increment and the corresponding buffer time window; and integrating the thermal buffer capacity quantized data of multiple server components to obtain the overall thermal buffer capacity data of the server cluster.

[0035] In this embodiment, the current thermal capacity status data of each server component is obtained from server temperature data through thermal capacity characteristic analysis. The aim is to acquire real-time thermal status information of key components within the server (such as CPU, GPU, memory modules, and storage devices). Thermal capacity characteristic analysis can be based on a pre-established component thermal model, combined with temperature data collected by sensors, to calculate the component's current stored heat energy, heat capacity, or its response to temperature changes. For example, by monitoring the component's surface and core temperatures, and combining this with its material properties (such as specific heat capacity and density) and geometric dimensions, the component's heat energy reserve at the current temperature can be estimated, thus forming the current thermal capacity status data.

[0036] In this embodiment, load change trend data within a preset first time period is predicted from computational load data. This load change trend data includes load growth rate and load fluctuation characteristics, aiming to understand the changes in the computing demand of the server cluster in advance, thereby predicting its future hot load generation. Computational load data may include indicators such as CPU utilization, memory usage, I / O throughput, and network traffic. The prediction method can employ machine learning models (such as Recurrent Neural Networks (RNNs) or Long Short-Term Memory Networks (LSTMs)) or statistical models (such as ARIMA) to predict the load growth rate (i.e., the expected rate of increase or decrease in load) and load fluctuation characteristics (i.e., the degree of periodic change or random disturbance in load) within a future period (e.g., 5 minutes, 15 minutes, or 30 minutes) based on historical load data and current load trends.

[0037] Based on this, according to the current thermal capacity status data and the load change trend data, the remaining time data for each server component to reach the preset safe temperature threshold is calculated using a thermal buffer time window algorithm. The core of this algorithm lies in dynamically evaluating the continuous operating time of a component without triggering the overheat protection mechanism by combining the component's current thermal state and expected thermal load changes. For example, the algorithm can establish a transient thermal model of the component, using the current temperature as the initial condition, converting the predicted load change trend into a heat generation rate, and then simulating the component temperature change over time until the preset safe temperature threshold is reached, thereby determining the time difference from the current moment to that threshold point, i.e., the remaining time data.

[0038] Furthermore, based on the remaining time data and the preset safe temperature threshold data, a thermal buffer capacity quantification data for each server component is obtained through a thermal buffer quantification algorithm. This thermal buffer capacity quantification data represents the tolerable thermal load increment and the corresponding buffer time window. The algorithm transforms the remaining time into an operable heat dissipation control index. Specifically, it calculates the maximum additional thermal load increment that a component can withstand within the remaining time, without exceeding the safe temperature threshold, and the time window within which that increment can be maintained. For example, if a component has 10 minutes of remaining time and its thermal capacity allows, the algorithm might quantify that it can withstand an additional 50W of power within 10 minutes, or an additional 100W of power within 5 minutes, thus forming thermal buffer capacity quantification data that includes the tolerable thermal load increment and the corresponding buffer time window.

[0039] In this embodiment, the final step is to integrate the quantified heat buffer capacity data of multiple server components to obtain the overall heat buffer capacity data of the server cluster. This integration process aims to provide a macroscopic, cluster-level view of heat buffer capacity. Integration methods may include weighted averaging of the quantified heat buffer capacity data of all components, taking the minimum value (based on the most vulnerable component), or comprehensive evaluation through a more complex cluster thermal model. For example, the component with the weakest heat buffer capacity in the cluster can be identified and used as a bottleneck to determine the short-term tolerable heat load increment and buffer time window for the entire cluster; or, by considering the thermal coupling effect between components, the additional heat that the cluster as a whole can absorb within a specific time period can be calculated.

[0040] In this embodiment, through the above technical solution, this application can analyze the current thermal capacity status and future load change trends of each server component in detail, starting from server temperature data and computing load data. This detailed analysis enables the thermal buffer time window algorithm to accurately calculate the remaining time for each component to reach the preset safe temperature threshold, and obtains the tolerable thermal load increment and corresponding buffer time window through the thermal buffer quantization algorithm. Finally, by integrating these component-level data, more accurate and dynamic overall thermal buffer capacity data of the server cluster is obtained. Compared with relying solely on a general pre-trained model, this solution improves the accuracy and real-time performance of thermal buffer capacity data calculation, providing reliable input for subsequent power smoothing optimization algorithms, thereby making heat dissipation control more precise and efficient. This effectively avoids the risk of server overheating, reduces unnecessary energy consumption, and improves the operational stability and energy efficiency ratio of the server cluster.

[0041] In one feasible implementation, the step of calculating the remaining time data for each server component to reach a preset safe temperature threshold using a thermal buffer time window algorithm based on the current thermal capacity status data and the load change trend data includes: calculating the heat capacity data of the server component absorbing heat at the current temperature based on the current thermal capacity status data; predicting the heat load increment data at different future time points based on the load change trend data; establishing a heat balance equation based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server component; predicting the time point data when the temperature of the server component reaches the preset safe temperature threshold based on the temperature rise rate data and the current temperature data of the server component; and calculating the time difference between the current time point and the time point data to obtain the remaining time data.

[0042] In this embodiment, when calculating the heat capacity data of a server component at the current temperature based on the current heat capacity state data, the heat capacity data is a key parameter for measuring the server component's ability to absorb or release heat at a specific temperature. This data can be obtained in various ways, for example, by calculating based on the material type, mass, and specific heat capacity parameter of the server component. Specifically, a material database for different server components (such as CPU, GPU, memory module, hard drive, etc.) can be pre-established, containing the specific heat capacity of each material. By combining the real-time monitored component mass (if variable) and the current temperature data, its heat capacity in the current state can be calculated. Alternatively, experimental testing or numerical simulation methods can be used to perform heat injection and temperature response measurements on the component at different temperature points, thereby fitting and obtaining its effective heat capacity data at the current temperature.

[0043] In this embodiment, when predicting the incremental heat load data at different future time points based on the load change trend data, the incremental heat load data reflects the additional heat generated by the server components due to increased computing tasks over a future period. Predicting this data is crucial for accurately assessing temperature changes. This prediction can be based on historical computing load data, using time series analysis models (such as ARIMA, exponential smoothing, etc.) or machine learning models (such as recurrent neural networks (RNN), long short-term memory networks (LSTM, etc.). These models can learn the load growth rate and load fluctuation characteristics, thereby predicting the changes in computing load at different time points (e.g., every minute) within a preset first time period (e.g., the next 5 minutes, 10 minutes, or longer), and then converting this into the corresponding incremental heat load. For example, the predicted load data can be converted into incremental heat load data based on the mapping relationship between indicators such as CPU utilization, memory utilization, and heat generation.

[0044] In this embodiment, when establishing a heat balance equation based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server component, the heat balance equation is a mathematical model describing the relationship between the heat input and output inside the server component and its own temperature change. This equation comprehensively considers the heat generated by the internal heat source of the component, the heat lost through conduction, convection, radiation, etc., and the heat stored within the component itself. For example, the lumped parameter method can be used to treat the server component as a uniformly heated whole, and its heat balance equation can be expressed as: “m × c × (dT / dt) =Q_gen - Q_diss”, where “m” is the component mass, “c” is the specific heat capacity, “dT / dt” is the temperature rise rate, “Q_gen” is the internal heat generation rate (related to the heat load increment), and “Q_diss” is the heat dissipation rate (related to ambient temperature, heat dissipation coefficient, etc.). By solving this equation, the temperature rise rate data under the current heat capacity data and heat load increment data can be obtained.

[0045] In this embodiment, when predicting the time point when the server component's temperature will reach the preset safe temperature threshold based on the temperature rise rate data and the server component's current temperature data, after obtaining the server component's current temperature data and temperature rise rate data, it is possible to predict when its temperature will reach the preset safe temperature threshold. If the temperature rise rate can be approximated as a constant within the prediction time period, a simple linear extrapolation method can be used for prediction: "Time point = Current time + (Preset safe temperature threshold - Current temperature) / Temperature rise rate". However, considering that the temperature rise rate may change over time, a more accurate method is to use numerical integration or iterative calculation. For example, the temperature change can be calculated according to the heat balance equation in each small time step and accumulated until the temperature reaches or exceeds the preset safe temperature threshold, thereby determining the corresponding time point.

[0046] In this embodiment, when calculating the time difference between the current time point and the predicted time point to obtain the remaining time data, once the time point when the server component temperature reaches the preset safe temperature threshold is determined, the time difference between the current time point and the predicted time point can be calculated through a simple subtraction operation. This time difference is the remaining time data, which intuitively represents how much time it will take for the server component to reach the upper limit of the safe temperature under the current operating state and predicted load trend.

[0047] In this embodiment, through the above technical solution, this application can accurately establish a thermal balance equation based on the current thermal capacity status data and future load change trend data of server components, and dynamically calculate the remaining time data for the server components to reach the preset safe temperature threshold. This refined calculation method overcomes the real-time insufficiency and accuracy deviation problems that may exist when relying solely on pre-trained models, making the acquisition of thermal buffer capacity data more accurate and timely. Specifically, by considering the actual thermal capacity of the components and the dynamic thermal load increment, the temperature change trend of the server components can be more realistically reflected, thereby providing reliable input for subsequent thermal buffer quantification algorithms. This not only improves the accuracy of the thermal buffer capacity assessment of the server cluster, but also provides a more solid foundation for the power smoothing optimization algorithm, ensuring that the server's thermal buffer capacity can be utilized more effectively when performing heat dissipation regulation, avoiding premature or late power adjustments due to misjudgment, thereby improving the response speed and safety of the entire heat dissipation regulation system.

[0048] In one feasible implementation, the step of establishing a heat balance equation based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server component includes: extracting material heat capacity parameter data, geometric dimension parameter data, and current temperature distribution data of the server component from the current heat capacity state data; calculating the heat generation rate data generated inside the server component based on the heat load increment data; calculating the heat capacity coefficient of the server component based on the material heat capacity parameter data and the geometric dimension parameter data; inputting the heat generation rate data, the heat capacity coefficient, and the current temperature distribution data into the heat conduction equation to solve for the temperature change response data of the server component; extracting the temperature change curve data over time from the temperature change response data, and calculating the temperature rise rate data.

[0049] In this embodiment, material thermal capacity parameters, geometric dimension parameters, and current temperature distribution data of the server component are extracted from the current thermal capacity state data. Material thermal capacity parameters refer to the inherent thermal properties of the materials used in the server component, such as specific heat capacity and thermal conductivity. These parameters are fundamental to describing the temperature change characteristics of the material during heat absorption or release and can be obtained by consulting material handbooks, experimental measurements, or data provided by the supplier. Geometric dimension parameters refer to the physical dimensions and shape information of the server component, such as volume, surface area, thickness, and internal structure. These parameters are crucial for determining heat transfer paths, heat capacity, and heat generation areas, and are typically obtained through CAD models or actual measurements. Current temperature distribution data refers to the spatial distribution of temperature values ​​at various points within the server component at a given moment. This reflects the component's current actual thermal state and serves as the initial condition for subsequent thermal balance calculations. It can be acquired in real-time or near real-time using embedded temperature sensor arrays, infrared thermal imagers, or numerical simulations.

[0050] In this embodiment, the heat generation rate data generated inside the server component is then calculated based on the heat load increment data. The heat load increment data represents the change in heat input to the server component over a future period due to factors such as increased computing tasks and changes in power consumption. The heat generation rate data refers to the heat generated inside the server component per unit time due to power consumption conversion. This data is usually directly related to the instantaneous power consumption of the component. For example, the power consumption of the CPU or GPU can be directly converted into its internal heat generation rate, which can be estimated in real time by monitoring the component's voltage and current or using a power consumption model.

[0051] Based on this, the thermal capacity coefficient of the server components is calculated according to the material heat capacity parameter data and geometric dimension parameter data. The thermal capacity coefficient is a comprehensive parameter that measures the ability of a server component to absorb or release heat as a whole or locally. It can be the total heat capacity of the component (mass multiplied by specific heat capacity), or it can be the equivalent heat capacity that takes into account the component's geometry and material distribution. This coefficient reflects the amount of heat that the component needs to absorb per unit degree of temperature increase and is one of the key parameters for establishing the heat balance equation.

[0052] Furthermore, the heat generation rate data, thermal capacity coefficient, and current temperature distribution data are input into the heat conduction equation to obtain the temperature change response data of the server components. The heat conduction equation is a partial differential equation describing the heat conduction process within an object; for example, the three-dimensional transient heat conduction equation considers heat generation, storage, and conduction, accurately simulating the temperature field changes within the component over time. Solving typically employs numerical methods, such as the finite element method (FEM), finite difference method (FDM), or finite volume method (FVM), discretizing the complex physical model into a system of algebraic equations for solution. Through these methods, the transient temperature field changes of the server components under given heat generation rate, thermal capacity coefficient, and initial temperature distribution can be obtained, i.e., the temperature change response data.

[0053] In this embodiment, finally, the temperature change curve data over time is extracted from the temperature change response data, and the temperature rise rate data is calculated. The temperature change response data is a set of temperature values ​​of server components at different times and locations obtained after solving the heat conduction equation. The temperature change curve data over time can be the trajectory of key points (such as hot spots) or average temperatures of components selected from the temperature change response data over time. The temperature rise rate data is obtained by mathematically processing the temperature change curve data over time, such as by calculating the time derivative (dT / dt), to obtain the temperature rise rate of the component at a specific moment or within a specific time period, which provides instantaneous trend information of component temperature changes.

[0054] In this embodiment, through the above technical solution, this application can accurately establish a heat conduction equation based on the heat capacity parameters, geometric dimension parameters, and current temperature distribution data of server components, combined with the heat generation rate data calculated from the heat load increment data. By solving this equation, the temperature change response data of the server components under dynamic heat load can be obtained, and then the temperature change curve data over time can be extracted to calculate the accurate temperature rise rate data. This method of establishing and solving the heat balance equation based on a physical model overcomes the problem of insufficient accuracy of traditional empirical models or simplified models in complex dynamic scenarios, and improves the accuracy and reliability of predicting the temperature rise rate of server components. This enables the subsequent heat buffer time window algorithm to more accurately calculate the remaining time for the server components to reach the preset safe temperature threshold, thereby providing a solid foundation for the assessment of the heat buffer capacity of the server cluster, ensuring the effectiveness and safety of the heat dissipation control strategy, and avoiding thermal management failure or overly conservative approaches due to inaccurate predictions.

[0055] In one feasible implementation, the step of analyzing the impedance frequency characteristics of the power supply system and generating power supply system impedance characteristic data based on the power supply system status data includes: calculating the basic impedance spectrum data of the power supply system based on the voltage and current data in the power supply system status data; injecting multiple test power disturbance signals of different frequencies into the power supply system, while simultaneously monitoring and recording the voltage response signal data of the power supply system; obtaining the actual impedance values ​​of the power supply system at each frequency point using an impedance frequency scanning algorithm based on the test power disturbance signals and the voltage response signal data; analyzing the actual impedance values ​​to identify frequency points where the impedance values ​​are lower than a preset first threshold, thus obtaining strong resonant frequency point data; analyzing the actual impedance values ​​to identify frequency bands where the impedance values ​​change by more than a preset second threshold within a preset frequency range, thus obtaining high-sensitivity frequency band data; and integrating the strong resonant frequency point data, the high-sensitivity frequency band data, and the basic impedance spectrum data to obtain the power supply system impedance characteristic data.

[0056] In this embodiment, the fundamental impedance spectrum data of the power supply system is calculated based on the voltage and current data in the power supply system status data. This step aims to preliminarily establish an impedance frequency characteristic model of the power supply system by analyzing the voltage and current data under normal operating conditions. Specifically, the voltage and current data in the time domain can be converted to the frequency domain using signal processing methods such as Fourier transform. Then, according to Ohm's law, the impedance values ​​at different frequencies are calculated, thereby obtaining the fundamental impedance spectrum data of the power supply system. This fundamental impedance spectrum data reflects the inherent frequency response characteristics of the power supply system under normal operating conditions, providing a benchmark for subsequent more refined impedance analysis.

[0057] In this embodiment, multiple test power disturbance signals of different frequencies are injected into the power supply system, while the voltage response signal data of the power supply system is monitored and recorded. This step is to actively detect the true impedance characteristics of the power supply system at different frequencies, especially in areas where resonance or high sensitivity may exist. The test power disturbance signal can be a small-amplitude, frequency-sweeping sine wave signal, applied to the power supply system through a specialized injection device (such as a programmable power supply or a signal generator combined with a power amplifier). Simultaneously with the injection of the disturbance signal, high-precision data acquisition equipment (such as an oscilloscope or power quality analyzer) is needed to monitor and record the voltage response signal data of the power supply system in real time at each injection frequency. These response signals will contain the true feedback of the power supply system to external disturbances and are crucial for calculating the actual impedance.

[0058] Based on this, the actual impedance values ​​of the power supply system at each frequency point are obtained using the test power disturbance signal and the voltage response signal data through an impedance frequency scanning algorithm. This step utilizes the injected test signal and the monitored voltage response signal to accurately calculate the actual impedance value of the power supply system at each test frequency point using a specialized impedance frequency scanning algorithm. The impedance frequency scanning algorithm typically involves performing frequency domain analysis (such as Fast Fourier Transform) on the test power disturbance signal and the voltage response signal, then calculating their ratio and considering the phase relationship to obtain the complex impedance value. These actual impedance values ​​can more accurately reflect the dynamic response of the power supply system at a specific frequency, compensating for potential deficiencies in calculations based solely on basic operating data.

[0059] Further, the actual impedance data is analyzed to identify frequency points where the impedance value is lower than a preset first threshold, thus obtaining strong resonant frequency point data. This step aims to identify strong resonant frequency points in the power supply system from the actual impedance data. Near the resonant frequency point, the impedance amplitude of the power supply system typically decreases. The preset first threshold is a critical value set based on the power supply system's design parameters, safe operation requirements, and historical experience. When the actual impedance value at a certain frequency point is lower than this threshold, it indicates that the power supply system has a relatively small resistance to current changes at that frequency, making it prone to large current fluctuations. Therefore, it is identified as a strong resonant frequency point and requires special attention and avoidance.

[0060] In this embodiment, the actual impedance data is analyzed to identify frequency ranges where the impedance value changes by more than a preset second threshold within a preset frequency range, thus obtaining high-sensitivity frequency range data. This step is used to identify frequency ranges in the power supply system that are highly sensitive to frequency changes. A high-sensitivity frequency range refers to a frequency range within which even a small change in the impedance value of the power supply system can lead to drastic fluctuations in the system response. By performing sliding window analysis or gradient analysis on the actual impedance data, it is possible to detect whether the change in the impedance value within a specific frequency range (e.g., a relatively narrow frequency interval) exceeds a preset second threshold. The preset second threshold is also determined based on system characteristics and safety margins. Identifying high-sensitivity frequency range data helps avoid large or rapid frequency component adjustments within this range during power regulation, thereby preventing system instability.

[0061] In this embodiment, the strong resonant frequency data, the high-sensitivity frequency band data, and the basic impedance spectrum data are integrated to obtain the power supply system impedance characteristic data. This step integrates the various impedance-related data obtained from the previous analysis to form a comprehensive and complete power supply system impedance characteristic data. The integration process may include annotating the strong resonant frequency data and the high-sensitivity frequency band data onto the basic impedance spectrum data, or generating a structured data package containing all this key information. This integrated power supply system impedance characteristic data not only includes the system's basic impedance response over a wide frequency range, but also highlights the strong resonant points and high-sensitivity regions that need to be avoided, providing accurate and reliable input for subsequent power smoothing optimization algorithms and ensuring that power adjustment commands can effectively avoid potential risk areas.

[0062] In this embodiment, by calculating the basic impedance spectrum data, the overall frequency response of the power supply system can be preliminarily understood. By injecting a test power disturbance signal and performing impedance frequency scanning, the actual impedance values ​​of the power supply system at various frequency points can be obtained, thus compensating for potential blind spots that may exist when relying solely on operational data. Based on this, strong resonant frequency points with impedance values ​​below a preset first threshold and highly sensitive frequency bands where impedance values ​​change by more than a preset second threshold within a preset frequency range are further identified. This refined impedance characteristic data, especially the identification of strong resonant points and highly sensitive frequency bands, enables the subsequent power smoothing optimization algorithm to more accurately plan the power adjustment process, effectively avoiding the resonant frequency region of the power supply system, thereby preventing drastic voltage or current fluctuations caused by power adjustments and improving the stability and reliability of the power supply system. This not only ensures the stable operation of the server cluster but also extends the service life of the power supply equipment, providing a solid power guarantee for intelligent dynamic heat dissipation control.

[0063] In one feasible implementation, the step of analyzing the actual impedance numerical data and identifying frequency points where the impedance value is lower than a preset first threshold to obtain strong resonant frequency point data includes: performing frequency domain analysis on the actual impedance numerical data to obtain impedance-frequency characteristic curve data; identifying frequency points where the impedance amplitude is lower than a preset amplitude threshold on the impedance-frequency characteristic curve data to obtain frequency point set data; selecting frequency points whose phase angle is close to a preset resonant phase angle from the frequency point set data to obtain preliminary resonant frequency point data; verifying the stability of the preliminary resonant frequency point data, the verification method including re-injecting a test signal of the corresponding frequency and monitoring the voltage amplification factor; and determining the frequency points that pass the stability verification as the final strong resonant frequency point data based on the stability verification results.

[0064] In this embodiment, frequency domain analysis is performed on the actual impedance numerical data. For example, signal processing techniques such as Fast Fourier Transform (FFT), wavelet transform, or sliding Fourier transform are used to convert discrete impedance measurement points into continuous impedance-frequency response curve data. This curve intuitively shows the law of impedance change with frequency in the power supply system, providing a comprehensive view for subsequent resonant point identification and helping to capture more subtle impedance change characteristics. On the obtained impedance-frequency response curve data, the system automatically scans and identifies all frequency points where the impedance amplitude is lower than a preset amplitude threshold. This preset amplitude threshold can be dynamically adjusted according to the power supply system design specifications, historical operating data, empirical values, or through machine learning models, serving as the basis for preliminary screening of potential resonant points. These identified frequency points constitute a frequency point set data, serving as the basis for further refined analysis.

[0065] In this embodiment, the resonance phenomenon is not only manifested as a change in impedance amplitude, but also accompanied by phase angle characteristics. For example, in series resonance, the impedance phase angle approaches zero; in parallel resonance, the impedance phase angle approaches ±90 degrees. Therefore, by further filtering the frequency point set data to identify those frequency points with phase angles close to the preset resonance phase angle (e.g., close to 0 degrees or ±90 degrees, allowing for a certain deviation range), the frequency points with resonance characteristics can be located more accurately, thereby obtaining preliminary resonance frequency point data and effectively eliminating impedance amplitude drops caused by other non-resonant factors, such as impedance valleys caused by certain filter or load characteristics.

[0066] In this embodiment, to ensure that the identified initial resonant frequency point is a true strong resonant point and poses a potential threat to system stability, stability verification is required. Specifically, the system re-injects a test signal corresponding to the initial resonant frequency point into the power supply system. The amplitude of this test signal should be small enough to avoid interfering with normal operation, but sufficient to excite a potential resonant response. Simultaneously, the voltage response of the power supply system at this frequency is monitored in real time, and the voltage amplification factor is calculated. If the voltage amplification factor is higher than normal (e.g., exceeding a preset amplification factor threshold), it indicates that the frequency point does indeed have a strong resonant effect, potentially causing drastic voltage or current fluctuations, thus affecting the stable operation of the server cluster. Only frequency points that pass the above stability verification are ultimately confirmed as strong resonant frequency points. This means that these frequency points not only exhibit resonant characteristics in impedance amplitude and phase angle, but also, through actual testing, are confirmed to cause amplification of system voltage or current, thus posing a real resonant risk to the power supply system.

[0067] In this embodiment, through the above technical solution, this application no longer relies solely on a single impedance amplitude threshold for identifying strong resonant frequency points in a power supply system. Instead, it introduces a multi-dimensional comprehensive analysis and verification mechanism. First, complete impedance-frequency characteristic curve data is obtained through frequency domain analysis, laying the foundation for a comprehensive understanding of the system's impedance behavior. Second, after initially screening out frequency points with low impedance amplitudes, further refined screening is performed by combining phase angle characteristics. This effectively distinguishes true resonant points from non-resonant impedance drops, improving the accuracy of initial identification. More importantly, by re-injecting test signals and monitoring voltage amplification for stability verification, this application can confirm the strength and potential hazards of resonant points from the perspective of actual operational effects, avoiding potential deviations in theoretical analysis. This multi-stage, multi-dimensional identification and verification process ensures that the finally determined strong resonant frequency point data is highly accurate and reliable. Therefore, based on these precisely identified strong resonant frequency points, the subsequent power smoothing optimization algorithm can more accurately plan the power adjustment process, effectively avoid truly risky resonant regions, thereby improving the operational stability and reliability of the server cluster power supply system, avoiding equipment damage or service interruption caused by resonance, while optimizing the efficiency of heat dissipation control, and ensuring the continuous and efficient operation of the server cluster.

[0068] In one feasible implementation, the step of generating first power adjustment command data based on the thermal buffer capacity data and the power supply system impedance characteristic data using a power smoothing optimization algorithm includes: extracting thermal buffer time window data and thermal buffer capacity limit data from the thermal buffer capacity data; extracting resonant frequency range data that must be avoided and allowable adjustment frequency range data from the power supply system impedance characteristic data; determining time constraint data for power adjustment based on the thermal buffer time window data and the thermal buffer capacity limit data; determining frequency constraint data for power adjustment based on the resonant frequency range data and the allowable adjustment frequency range data; inputting the time constraint data and the frequency constraint data into a dual-constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy both thermal buffer constraints and impedance constraints; and selecting the scheme with the optimal smoothness from the set of candidate power adjustment schemes to obtain the first power adjustment command data.

[0069] In this embodiment, heat buffer time window data and heat buffer capacity limit data are extracted from the heat buffer capacity data. The heat buffer time window data refers to the length of time the server cluster can continuously withstand additional heat load without exceeding a preset safe temperature threshold. The heat buffer capacity limit data represents the maximum amount of additional heat load the server cluster can absorb within a given time window. These data are crucial for ensuring server temperature safety during power adjustments. They can be parsed from the heat buffer capacity data, for example, by analyzing the tolerable heat load increment and corresponding buffer time window contained in the quantified heat buffer capacity data.

[0070] In this embodiment, the resonant frequency range that must be avoided and the allowable adjustment frequency range are extracted from the power supply system impedance characteristic data. The power supply system impedance characteristic data details the impedance response of the power supply system at different frequencies. The resonant frequency range that must be avoided refers to the frequency range where the power supply system impedance is abnormally low or high, easily triggering resonance and causing large fluctuations in voltage or current. The allowable adjustment frequency range refers to the frequency range where the power supply system impedance characteristics are relatively stable, and power adjustment operations within this range have a minimal impact on the stability of the power supply system. These data are crucial for ensuring that the power adjustment process does not trigger instability in the power supply system.

[0071] In this embodiment, the time constraint data for power adjustment is determined based on the heat buffer time window data and the heat buffer capacity limit data. The time constraint data for power adjustment is set based on the heat buffer capacity of the server cluster. It stipulates that the power adjustment operation must be completed within the heat buffer time window of the server cluster, and the increase in heat load generated during the adjustment process cannot exceed the heat buffer capacity limit. For example, if the server cluster can withstand an additional 10kW of heat load in the next 5 minutes, then the duration of the power adjustment cannot exceed 5 minutes, and the adjustment magnitude cannot cause the instantaneous heat load increase to exceed 10kW.

[0072] In this embodiment, the frequency constraint data for power adjustment is determined based on the resonant frequency range data and the allowable adjustment frequency range data. The frequency constraint data for power adjustment is set based on the impedance characteristics of the power supply system. It requires that the frequency components of the power adjustment command must avoid the resonant frequency range of the power supply system and be distributed as close as possible within the allowable adjustment frequency range. This can be achieved by performing a Fourier transform on the power adjustment command, analyzing its frequency spectrum, ensuring that its main energy is concentrated in the safe frequency range, and avoiding large disturbances at the resonant frequency point.

[0073] Furthermore, the time constraint data and the frequency constraint data are input into a dual-constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy both thermal buffer constraints and impedance constraints. The dual-constraint optimization algorithm is a mathematical optimization method capable of handling multiple interrelated constraints simultaneously. It uses the aforementioned time constraint data and frequency constraint data as input to the algorithm, searching for a series of feasible power adjustment schemes while satisfying these constraints. These schemes constitute the set of candidate power adjustment schemes, each representing a power adjustment strategy that achieves a balance between thermal safety and electrical stability. This algorithm can be implemented using techniques such as multi-objective optimization and constraint satisfaction problem solving.

[0074] In this embodiment, the scheme with the optimal smoothness is selected from the candidate power adjustment scheme set data to obtain the first power adjustment command data. The scheme with the optimal smoothness refers to the scheme with the smoothest power adjustment process and the smallest rate of change, under the premise of satisfying all thermal buffering and impedance constraints. Selecting the scheme with the optimal smoothness helps to reduce the impact on the server cluster and power supply system, reduce the risk of transient response, and improve the overall stability and reliability of the system. Smoothness can be quantified by evaluating the derivative, rate of change, or spectral characteristics of the power adjustment command, for example, selecting the scheme with the smallest power change slope or the fewest high-frequency components.

[0075] In this embodiment, through the above technical solution, this application can deeply integrate the thermal buffering capability of the server cluster with the electrical stability requirements of the power supply system. First, by accurately extracting thermal buffering time window data and thermal buffering capacity limit data from the thermal buffering capacity data, and identifying the resonant frequency range to be avoided and the allowable adjustment frequency range data from the power supply system impedance characteristic data, clear and quantifiable time and frequency constraints are set for the power adjustment operation. Subsequently, these dual constraints are input into a dual-constraint optimization algorithm, which can systematically generate a series of candidate power adjustment schemes that simultaneously meet the requirements of thermal safety and electrical stability. Finally, the scheme with the optimal smoothness is selected from these candidate schemes as the first power adjustment command data, ensuring that the power adjustment process can efficiently utilize the server's thermal buffering capability for heat dissipation while effectively avoiding the resonance risk of the power supply system. This avoids system instability or low heat dissipation efficiency caused by improper power adjustment, improving the intelligence, safety, and reliability of server cluster heat dissipation control.

[0076] In one feasible implementation, the step of inputting the time constraint data and the frequency constraint data into a dual-constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy the thermal buffer constraint and the impedance constraint includes: constructing a thermal buffer constraint function based on the time constraint data, wherein the thermal buffer constraint function is used to constrain the power adjustment process to be completed within the thermal buffer time window and not exceed the thermal buffer capacity limit; constructing an impedance constraint function based on the frequency constraint data, wherein the impedance constraint function is used to constrain the frequency components of the power adjustment to avoid the resonant frequency range and be mainly distributed within the allowable adjustment frequency range; establishing a mathematical model of the optimization problem based on the thermal buffer constraint function and the impedance constraint function, wherein the model takes the power adjustment smoothness as the optimization objective and the thermal buffer constraint function and the impedance constraint function as constraints; solving the mathematical model of the optimization problem using a sequential quadratic programming algorithm to obtain preliminary optimization solution data that satisfies the dual constraints; and performing feasibility verification and boundary condition re-satisfaction processing on the preliminary optimization solution data to obtain the set of candidate power adjustment schemes.

[0077] In this embodiment, a heat buffer constraint function is constructed based on time constraint data. This function aims to ensure that the power adjustment process is completed within the heat buffer time window of the server cluster, and that the adjustment magnitude does not exceed the heat buffer capacity limit. The heat buffer constraint function can be expressed as a series of inequalities or equations. For example, by taking into account parameters such as the server's instantaneous temperature change rate, current temperature, preset safe temperature threshold, and heat buffer capacity limit, it ensures that at any point in time, the server temperature will not exceed the safe range, and the additional heat load generated by the power adjustment is always within the tolerable heat buffer capacity. This function can be based on a thermodynamic model, such as a lumped parameter model or a distributed parameter model, to describe the temperature response of the server components, thereby relating the power adjustment amount to temperature changes.

[0078] In this embodiment, an impedance constraint function is constructed based on frequency constraint data. This function constrains the frequency components of the power adjustment to avoid the resonant frequency range of the power supply system and to be mainly distributed within the allowable adjustment frequency range. The impedance constraint function can be constructed using frequency domain analysis methods, for example, by comparing the Fourier transform result of the power adjustment command with the impedance characteristic data of the power supply system. The function can be set such that at strong resonant frequency points or highly sensitive frequency bands of the power supply system, the spectral components of the power adjustment command must be below a certain preset extremely low threshold to avoid triggering resonance; while within the allowable adjustment frequency range, the spectral components of the power adjustment command can be distributed relatively freely, but still need to meet certain smoothness requirements.

[0079] Based on this, and using the aforementioned heat buffer constraint function and impedance constraint function, a mathematical model for the optimization problem is established. This model takes power adjustment smoothness as the optimization objective and the heat buffer constraint function and impedance constraint function as constraints. Power adjustment smoothness is typically quantified by minimizing the second derivative (i.e., acceleration) of the power adjustment command or the sum of the squares of its rates of change, to avoid drastic power fluctuations. This mathematical model can be a nonlinear programming problem, where the decision variables are the sequence of power adjustment commands changing over time, the objective function is an indicator of smoothness, and the heat buffer constraint function and impedance constraint function serve as inequality or equality constraints to ensure that the solution is physically and electrically feasible.

[0080] In this embodiment, to efficiently solve the mathematical model data of the aforementioned optimization problem, this application employs a sequential quadratic programming algorithm (SQP) to obtain preliminary optimization solution data that satisfies the dual constraints. SQP is an iterative optimization method, particularly suitable for solving optimization problems with nonlinear objective functions and nonlinear constraints. It solves the problem by approximating it as a quadratic programming subproblem in each iteration, and then uses the solution of the subproblem to update the search direction and step size. This method can effectively handle complex constraints and has a fast convergence speed and good global convergence, thus efficiently finding a power adjustment scheme that satisfies both heat buffer and impedance constraints.

[0081] In this embodiment, the preliminary optimized solution data is finally subjected to feasibility verification and boundary condition re-satisfaction processing to obtain the final set of candidate power adjustment schemes. Feasibility verification aims to check whether the preliminary optimized solution strictly satisfies all thermal buffer constraints and impedance constraints; for example, it checks for any numerical errors causing minor boundary overflows. Boundary condition re-satisfaction processing involves fine-tuning the preliminary solution when minor boundary overflows or non-compliance with strict boundary conditions are found, ensuring it fully conforms to all preset constraints. For example, the solution is adjusted back to the feasible region using projection methods or penalty function methods, ensuring that the final power adjustment scheme is safe, reliable, and fully compliant in practical applications.

[0082] In this embodiment, through the above technical solution, this application provides a structured and robust optimization method for generating power adjustment commands that simultaneously satisfy both heat buffer and power supply system impedance constraints. Specifically, by constructing precise heat buffer constraint functions and impedance constraint functions based on time constraint data and frequency constraint data respectively, complex physical and electrical constraints can be transformed into explicit mathematical expressions, laying a solid foundation for subsequent optimization. Based on this, a mathematical model with power adjustment smoothness as the optimization objective is established, and a sequential quadratic programming algorithm is used to solve it. This not only efficiently finds preliminary optimization solutions that satisfy the dual constraints but also ensures that the generated power adjustment commands have good smoothness, avoiding drastic power fluctuations, thereby effectively reducing the impact on server hardware and potential threats to power supply system stability. Furthermore, feasibility verification and boundary condition re-satisfaction processing of the preliminary optimization solution data further ensure the safety and reliability of the final candidate power adjustment scheme set data in practical applications, ensuring that the power adjustment commands will not cause server overheating or trigger power supply system resonance under any circumstances, thereby improving the operational stability and energy efficiency of the server cluster.

[0083] In one feasible implementation, the step of generating a second power adjustment command sequence data for multiple heat dissipation devices based on the first power adjustment command data and the heat dissipation device operation data using a phase-shifted cooperative scheduling algorithm includes: extracting dynamic response characteristic data of multiple heat dissipation devices from the heat dissipation device operation data, the dynamic response characteristic data including device start-up delay time, power rise time, steady-state holding time, and shutdown decay time; decomposing the first power adjustment command data to obtain total power adjustment task data, the total power adjustment task data including overall power adjustment amplitude, adjustment start time, and adjustment duration; obtaining sub-task data suitable for the characteristics of each heat dissipation device using a response characteristic matching algorithm based on the dynamic response characteristic data and the total power adjustment task data; calculating the optimal execution time point data for each sub-task using a timing misalignment optimization algorithm based on the time parameters in the dynamic response characteristic data and the sub-task data, the optimal execution time point data being used to control the peak power adjustment times of different heat dissipation devices to be staggered; and combining the sub-task data and the optimal execution time point data to obtain the second power adjustment command sequence data.

[0084] In this embodiment, dynamic response characteristic data of multiple heat dissipation devices are extracted from the operating data of the heat dissipation devices. This dynamic response characteristic data includes device start-up delay time, power rise time, steady-state hold time, and shutdown decay time. Dynamic response characteristic data refers to the inherent properties of a heat dissipation device that change its power output or heat dissipation capacity over time after receiving a power adjustment command. These characteristics are crucial for precise control and coordinated scheduling. Device start-up delay time is the time interval between receiving a start command and the device beginning its actual response. This is typically determined by the device's mechanical inertia, electrical start-up characteristics, or the response speed of the control system. Power rise time is the time required for the device to reach its target power output from a start-up state or a lower power state; for example, the time required for a fan to accelerate from low speed to high speed, or for a compressor to operate from low load to high load. Steady-state hold time is the time it takes for the device to stably maintain its target power output after reaching it. This is typically related to the device's operational stability, control accuracy, and external environmental conditions. Shutdown decay time is the time required for the device to go from operating to complete shutdown or for its power output to drop to its minimum. These data can be obtained through pre-performance testing of the heat dissipation devices, analysis of historical operating data, or specifications provided by the manufacturer. For example, by sending a step power command to the heat dissipation device in a controlled environment and monitoring the changes in parameters such as power output and temperature over time, the aforementioned time parameters can be extracted.

[0085] In this embodiment, total power adjustment task data is obtained by decomposing the first power adjustment instruction data. This total power adjustment task data includes the overall power adjustment magnitude, adjustment start time, and adjustment duration. The total power adjustment task data is a high-level abstraction and decomposition of the first power adjustment instruction data generated by the power smoothing optimization algorithm, transforming it into an executable data package describing the overall power adjustment requirements of the cooling system. The overall power adjustment magnitude refers to the net power adjustment amount required for the entire server cluster cooling system, which can be an increase or decrease. The adjustment start time refers to the predetermined time point at which the entire cooling system begins to execute power adjustments according to the first power adjustment instruction. The adjustment duration refers to the time length required to complete the overall power adjustment magnitude. This decomposition process can be determined by parsing the timestamp, target power value, and current power value in the first power adjustment instruction data. For example, if the first power adjustment instruction is a power curve that changes over time, the integral or difference of the curve over a specific time period can be calculated to obtain the overall power adjustment magnitude and determine its start and end times.

[0086] In this embodiment, based on the dynamic response characteristic data and the total power adjustment task data, a response characteristic matching algorithm is used to obtain sub-task data suitable for the characteristics of each heat dissipation device. The response characteristic matching algorithm is an intelligent allocation strategy designed to rationally decompose the total power adjustment task into multiple independent sub-tasks based on the unique dynamic response characteristics of each heat dissipation device. Sub-task data consists of power adjustment instructions assigned to a single heat dissipation device, the content and timing of which match the device's dynamic response characteristics. This algorithm can consider factors such as the device's power capacity, efficiency curve, and response speed. For example, for devices with fast response speeds and large power capacities, a larger power adjustment amplitude or a faster adjustment rate can be assigned; for devices with slow response speeds or small power capacities, a smaller adjustment amplitude or a smoother adjustment curve is assigned. This can be achieved through an optimization model, for example, by decomposing the total power adjustment task into sub-tasks that satisfy the constraints of each device, with the goal of minimizing the total adjustment time or maximizing system efficiency.

[0087] In this embodiment, based on the time parameters in the dynamic response characteristic data and the sub-task data, the optimal execution time point data for each sub-task is calculated using a timing misalignment optimization algorithm. This optimal execution time point data is used to control the peak power adjustment times of different heat dissipation devices to be staggered. The purpose of the timing misalignment optimization algorithm is to avoid multiple heat dissipation devices reaching their power adjustment peaks simultaneously through fine-grained scheduling, thereby smoothing the instantaneous power demand of the entire heat dissipation system. The optimal execution time point data refers to the optimal time point at which each heat dissipation device begins executing its assigned sub-task, minimizing the impact of the overall power adjustment process on the power supply system while ensuring heat dissipation effectiveness. This algorithm can construct a multi-objective optimization model, with objectives including minimizing instantaneous power fluctuations, minimizing total adjustment time, and maximizing heat dissipation efficiency. Constraints include the dynamic response characteristics of each heat dissipation device (such as startup delay and power rise time), the requirements of the total power adjustment task, and the instantaneous power handling capacity of the power supply system. For example, genetic algorithms, particle swarm optimization algorithms, or rule-based heuristic algorithms can be used to search for the optimal combination of time points. By predicting the power response curve of each device and superimposing these curves, the algorithm can identify potential instantaneous power peaks and achieve a smooth power transition by adjusting the startup time or adjustment rate of each device to stagger these peaks.

[0088] In this embodiment, the subtask data and the optimal execution time point data are combined to obtain the second power adjustment instruction sequence data. The second power adjustment instruction sequence data is the final generated detailed instruction set that can be directly sent to each heat dissipation device for execution. This typically involves associating each device's subtasks (including target power, adjustment range, adjustment rate, etc.) with the calculated optimal execution time point to form a time-series instruction. For example, it can be a list or data structure containing device ID, instruction type (start, accelerate, decelerate, stop), target power value, duration, and precise execution timestamp.

[0089] In this embodiment, the above technical solution effectively addresses the problem of instantaneous power fluctuations that may result from simultaneous power adjustments by multiple heat dissipation devices. Specifically, by extracting the dynamic response characteristic data of the heat dissipation devices and combining it with the total power adjustment task, a response characteristic matching algorithm decomposes the total task into sub-tasks suitable for each device. Based on this, a timing-shifting optimization algorithm calculates the optimal execution time based on the dynamic response time parameters of each device, ensuring that the peak power adjustment times of different heat dissipation devices are staggered. This staggered collaborative scheduling mechanism avoids concentrated bursts of instantaneous power demand, thereby reducing the impact on the power supply system and improving its stability. Simultaneously, since each heat dissipation device executes its task according to its own characteristics and the optimized time point, not only is heat dissipation efficiency guaranteed, but the lifespan of the devices is also extended, achieving refined and intelligent heat dissipation control of the server cluster.

[0090] In the embodiments of this application, the intelligent dynamic heat dissipation control method for server clusters obtains server cluster operating status and power supply system status data, calculates thermal buffer capacity and impedance characteristics, generates optimized power adjustment instructions and staggers the adjustment time of heat dissipation equipment, thereby achieving heat dissipation control while reducing power surges, effectively suppressing the impact of power surges on the power supply system, improving the power stability of the data center, and reducing resonance risk.

[0091] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the intelligent dynamic heat dissipation control method for server clusters in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0092] This application also provides an intelligent dynamic heat dissipation control system for server clusters, see reference. Figure 2 The server cluster intelligent dynamic heat dissipation control system includes: a memory 10, a processor 20, and a server cluster intelligent dynamic heat dissipation control program stored on the memory 10 and capable of running on the processor 20. The server cluster intelligent dynamic heat dissipation control program is configured to implement the steps of the server cluster intelligent dynamic heat dissipation control method.

[0093] The intelligent dynamic heat dissipation control system for server clusters provided in this application, employing the intelligent dynamic heat dissipation control method for server clusters in the above embodiments, can improve the power stability of data centers and reduce the risk of resonance. Compared with the prior art, the beneficial effects of the intelligent dynamic heat dissipation control system for server clusters provided in this application are the same as those of the intelligent dynamic heat dissipation control method for server clusters provided in the above embodiments, and other technical features of the intelligent dynamic heat dissipation control system for server clusters are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0094] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. A method for intelligent dynamic heat dissipation control of a server cluster, characterized in that, The method includes: Obtain the server cluster's operating status data and power supply system status data, including server temperature data, computing load data, and heat dissipation equipment operating data; Based on the operational status data, the thermal buffer capacity data of the server cluster is calculated using a pre-trained thermal inertia analysis model; the thermal buffer capacity data is used to represent the additional thermal load that the server can withstand before the temperature reaches a preset safety threshold. Based on the power supply system status data, the impedance frequency characteristics of the power supply system are analyzed to generate power supply system impedance characteristic data. Based on the thermal buffer capacity data and the power supply system impedance characteristic data, a first power adjustment command data is generated through a power smoothing optimization algorithm; the first power adjustment command data is used to control the power adjustment process to avoid the resonant frequency region of the power supply system. Based on the first power adjustment command data and the heat dissipation device operation data, a second power adjustment command sequence data for multiple heat dissipation devices is generated through a phase-shifted cooperative scheduling algorithm; the second power adjustment command sequence data is used to control the power adjustment time points of multiple heat dissipation devices to be staggered. The second power adjustment command sequence data is sent to the corresponding heat dissipation device for execution, thereby completing the heat dissipation regulation.

2. The intelligent dynamic heat dissipation control method for server clusters as described in claim 1, characterized in that, The steps for calculating the hot buffer capacity data of the server cluster based on the aforementioned operational status data using a pre-trained thermal inertia analysis model include: The current thermal capacity status data of each server component is obtained from the server temperature data through thermal capacity characteristic analysis; The load change trend data within a preset first time period is predicted from the calculated load data, and the load change trend data includes the load growth rate and load fluctuation characteristics. Based on the current thermal capacity status data and the load change trend data, the remaining time for each server component to reach the preset safe temperature threshold is calculated using a thermal buffer time window algorithm. Based on the remaining time data and the preset safe temperature threshold data, the thermal buffer capacity quantization data of each server component is obtained through the thermal buffer quantization algorithm. The thermal buffer capacity quantization data is used to represent the tolerable thermal load increment and the corresponding buffer time window. By integrating the quantified hot buffer capacity data of multiple server components, the overall hot buffer capacity data of the server cluster is obtained.

3. The intelligent dynamic heat dissipation control method for server clusters as described in claim 2, characterized in that, The steps for calculating the remaining time for each server component to reach a preset safe temperature threshold based on the current thermal capacity status data and the load change trend data using a thermal buffer time window algorithm include: Calculate the heat capacity data of the server component at the current temperature based on the current heat capacity status data; Predict the heat load increment data at different future time points based on the load change trend data; A heat balance equation is established based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server components. Based on the temperature rise rate data and the current temperature data of the server components, predict the time point when the temperature of the server components will reach the preset safe temperature threshold. Calculate the time difference between the current time point and the data at the specified time point to obtain the remaining time data.

4. The intelligent dynamic heat dissipation control method for server clusters as described in claim 3, characterized in that, The steps for establishing a heat balance equation based on the heat capacity data and the heat load increment data to obtain the temperature rise rate data of the server components include: Extract the material thermal capacity parameters, geometric dimension parameters, and current temperature distribution data of the server components from the current thermal capacity status data; Calculate the heat generation rate data inside the server components based on the heat load increment data; Calculate the thermal capacity coefficient of the server component based on the material heat capacity parameter data and the geometric dimension parameter data; The heat generation rate data, the heat capacity coefficient, and the current temperature distribution data are input into the heat conduction equation to solve for the temperature change response data of the server components. Extract the temperature change curve data over time from the temperature change response data, and calculate the temperature rise rate data.

5. The intelligent dynamic heat dissipation control method for server clusters as described in claim 1, characterized in that, The steps for analyzing the impedance frequency characteristics of the power supply system and generating impedance characteristic data based on the power supply system status data include: Calculate the basic impedance spectrum data of the power supply system based on the voltage and current data in the power supply system status data; Multiple test power disturbance signals of different frequencies are injected into the power supply system, while the voltage response signal data of the power supply system is monitored and recorded. Based on the test power disturbance signal and the voltage response signal data, the actual impedance values ​​of the power supply system at each frequency point are obtained through the impedance frequency scanning algorithm. Analyze the actual impedance values ​​to identify frequency points where the impedance values ​​are lower than a preset first threshold, and obtain strong resonant frequency point data; Analyze the actual impedance data to identify frequency ranges where the impedance value changes more than a preset second threshold within a preset frequency range, and obtain high-sensitivity frequency range data; By integrating the strong resonant frequency data, the high-sensitivity frequency band data, and the basic impedance spectrum data, the impedance characteristic data of the power supply system is obtained.

6. The intelligent dynamic heat dissipation control method for server clusters as described in claim 5, characterized in that, The steps of analyzing the actual impedance data, identifying frequency points where the impedance value is lower than a preset first threshold, and obtaining strong resonant frequency point data include: Frequency domain analysis was performed on the actual impedance numerical data to obtain impedance-frequency characteristic curve data; On the impedance-frequency characteristic curve data, frequency points with impedance amplitudes lower than a preset amplitude threshold are identified to obtain frequency point set data; From the set of frequency points, frequency points whose phase angles are close to the preset resonant phase angle are selected to obtain preliminary resonant frequency point data; The stability of the preliminary resonant frequency data is verified by means of re-injecting the test signal at the corresponding frequency and monitoring the voltage amplification factor. Based on the stability verification results, the frequency points that passed the verification were determined as the final strong resonant frequency point data.

7. The intelligent dynamic heat dissipation control method for server clusters as described in claim 1, characterized in that, The steps for generating first power adjustment command data based on the hot buffer capacity data and the power supply system impedance characteristic data using a power smoothing optimization algorithm include: Extract hot buffer time window data and hot buffer capacity limit data from the hot buffer capacity data; Extract the resonant frequency range that must be avoided and the allowable adjustment frequency range from the impedance characteristic data of the power supply system; The time constraint data for power adjustment is determined based on the hot buffer time window data and the hot buffer capacity limit data. The frequency constraint data for power adjustment is determined based on the resonant frequency range data and the allowable adjustment frequency range data. The time constraint data and the frequency constraint data are input into the dual constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy the thermal buffer constraint and the impedance constraint. The scheme with the best smoothness is selected from the set of candidate power adjustment schemes to obtain the first power adjustment command data.

8. The intelligent dynamic heat dissipation control method for server clusters as described in claim 7, characterized in that, The steps of inputting the time constraint data and the frequency constraint data into a dual-constraint optimization algorithm to obtain a set of candidate power adjustment schemes that simultaneously satisfy the thermal buffer constraint and impedance constraint include: A hot buffer constraint function is constructed based on the time constraint data. The hot buffer constraint function is used to constrain the power adjustment process to be completed within the hot buffer time window and not exceed the hot buffer capacity limit. An impedance constraint function is constructed based on the frequency constraint data. The impedance constraint function is used to constrain the frequency components of power adjustment to avoid the resonant frequency range and to be mainly distributed within the allowable adjustment frequency range. A mathematical model for the optimization problem is established based on the heat buffer constraint function and the impedance constraint function. The model takes the power adjustment smoothness as the optimization objective and the heat buffer constraint function and the impedance constraint function as the constraint conditions. The mathematical model data of the optimization problem is solved using a sequential quadratic programming algorithm to obtain preliminary optimization solution data that satisfies the dual constraints; The feasibility of the preliminary optimized solution data is verified and the boundary conditions are satisfied again to obtain the set of candidate power adjustment schemes.

9. The intelligent dynamic heat dissipation control method for server clusters as described in claim 1, characterized in that, The step of generating a second power adjustment instruction sequence data for multiple heat dissipation devices based on the first power adjustment instruction data and the heat dissipation device operation data using a phase-shifted cooperative scheduling algorithm includes: Extract dynamic response characteristic data of multiple heat dissipation devices from the heat dissipation device operation data. The dynamic response characteristic data includes device start-up delay time, power rise time, steady state hold time, and shutdown decay time. The total power adjustment task data is obtained by decomposing the first power adjustment instruction data. The total power adjustment task data includes the overall power adjustment range, adjustment start time, and adjustment duration. Based on the dynamic response characteristic data and the total power adjustment task data, sub-task data suitable for the characteristics of each heat dissipation device is obtained through a response characteristic matching algorithm; Based on the time parameters in the dynamic response characteristic data and the sub-task data, the optimal execution time point data of each sub-task is calculated by the timing misalignment optimization algorithm. The optimal execution time point data is used to control the peak power adjustment times of different heat dissipation devices to be staggered. The subtask data and the optimal execution time point data are combined to obtain the second power adjustment instruction sequence data.

10. A server cluster intelligent dynamic heat dissipation control system, characterized in that, The server cluster intelligent dynamic heat dissipation control system includes: a memory, a processor, and a server cluster intelligent dynamic heat dissipation control program stored in the memory and executable on the processor. The server cluster intelligent dynamic heat dissipation control program is configured to implement the steps of the server cluster intelligent dynamic heat dissipation control method as described in any one of claims 1 to 9.

Images