Environment control system and control method for large-scale server cluster

Abstract

The invention relates to the field of computer system environment management, and discloses an environment control system and method for a large-scale server cluster, and the system comprises a data collection and preprocessing module which is used for collecting temperature observation data from a plurality of sensors disposed in the server cluster, and carrying out the preprocessing of the data; the mathematical modeling and discretization module is used for constructing a convection-diffusion type heat conduction model based on the internal structure and physical parameters of the server cluster and discretizing the model; and a data assimilation and state estimation module. By integrating an advanced data assimilation technology and cooperating with real-time temperature observation data and a mathematical model, accurate estimation of a temperature field and heat source distribution in a server cluster is realized, so that the accuracy of state estimation is improved, the prediction capability of a system on an unmeasured state is optimized, a control decision is ensured to be carried out based on the most accurate information, and the reliability of state estimation is improved. And the real-time performance and the reliability of system response are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer system environment management, and in particular to an environment control system and a control method for a large-scale server cluster. Background Art

[0002] In today's information age, large-scale server clusters serve as the core support of data centers, carrying the critical task of processing and storing massive amounts of data. As the scale of data centers continues to expand, how to effectively manage the operating environment of server clusters, especially temperature control, has become an important issue in improving system reliability and efficiency. Traditional environmental control systems usually use centralized air conditioning cooling combined with local fan heat dissipation to maintain the operation of equipment within a safe temperature range by setting fixed temperature thresholds. These systems may include basic temperature monitoring and control hardware that can respond to environmental changes to a certain extent and adjust cooling strategies to prevent equipment overheating.

[0003] However, with the increase in server density and computing demands, traditional control methods have gradually revealed their limitations. First, these systems often use oversimplified data processing methods and fail to fully utilize real-time data to optimize control strategies, resulting in response delays and uneven cooling. Second, static temperature settings are difficult to adapt to rapidly changing workloads and external environments, which may cause energy waste or fail to respond to local hot spots in a timely manner. In addition, the lack of customized control solutions for specific servers or server areas has prevented the system's overall energy efficiency and performance optimization potential from being fully utilized. Therefore, there is an urgent need to develop a smarter, more flexible and more efficient server cluster environment control method that can achieve dynamic temperature management based on precise data analysis to adapt to the complex and changing needs of large-scale data centers. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an environmental control system and control method for a large-scale server cluster, which solves the problems of the existing server cluster environmental control system in terms of insufficient data processing, slow response speed and lack of targeted control strategies.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an environmental control system for a large-scale server cluster, comprising: A data collection and preprocessing module, which is used to collect temperature observation data from multiple sensors set inside the server cluster and preprocess the data; Mathematical modeling and discretization module, which is used to build a convection-diffusion heat conduction model based on the internal structure and physical parameters of the server cluster, and discretize the model; The data assimilation and state estimation module is used to fuse the data processed by the data acquisition and preprocessing module with the discretization model of the mathematical modeling and discretization module to estimate the temperature field and heat source term distribution inside the server cluster; An optimal control solution module, which is used to calculate the control input of the cooling system based on the output of the data assimilation and state estimation module to optimize the temperature distribution of the server cluster; The execution and feedback module is used to implement the control strategy determined by the optimal control solution module and feed back the execution results to the data acquisition and preprocessing module.

[0006] Preferably, the data acquisition and preprocessing module includes: A synchronization unit, which is used to time-align temperature data measured by multiple sensors within the server cluster; The noise filtering unit is used to filter the time-aligned temperature data.

[0007] Preferably, the mathematical modeling and discretization module includes: A grid generation unit, which is used to perform spatial grid division on the internal area of ​​the server cluster; The time discrete unit is used to process the temperature field evolution process in time segments.

[0008] Preferably, the data assimilation and state estimation module includes: A target functional construction unit, which is used to construct a target functional based on temperature observation data and model prediction data; An adjoint solving unit is used to solve the adjoint equation according to the target functional and the discretization model.

[0009] Preferably, the data assimilation and state estimation module further includes a parameter updating unit, which is used to update the distribution parameters of the heat source term according to the output of the adjoint solution unit, specifically including iterative optimization of the objective functional: ; in, Indicates at location and the model-predicted temperature at time t, represents the corresponding observed temperature, β and is a regularization parameter used to control the smoothness of the solution and prevent overfitting, q is the heat source term, u is the control input, M is the number of sensors, and H is the length of the observation time window.

[0010] Preferably, the optimal control solution module includes: A gradient calculation unit, which is used to calculate the gradient of the control input according to the state estimation result output by the data assimilation and state estimation module and the heat source term distribution parameter; A control input updating unit is used to iteratively update the control input according to the gradient.

[0011] Preferably, the optimal control solution module further includes an optimal control strategy output unit, which is used to output a control strategy for controlling the cooling device.

[0012] Preferably, the execution and feedback module includes: A control signal output unit, which is used to convert the optimal control strategy into a control signal for the cooling execution device; A feedback unit is used to transmit the regulated temperature data to the data acquisition and preprocessing module.

[0013] Preferably, the feedback unit is configured to periodically sample and update the input data of the data acquisition and preprocessing module.

[0014] A method for controlling the environment of a large-scale server cluster comprises the following steps: S1. Temperature data collection: collect temperature observation data from sensors at multiple locations within the server cluster; S2, data assimilation process: the collected temperature observation data is input into the data assimilation and state estimation module for data assimilation to estimate the temperature field distribution and heat source items inside the server cluster; S3, control input calculation: according to the data assimilation results, the control input of the cooling system is calculated using the optimal control solution module; S4, execution and feedback: implement the calculated control input to adjust the cooling equipment, and feed back the adjustment results for data collection and preprocessing in the next cycle.

[0015] The present invention provides an environmental control system and control method for a large-scale server cluster, which has the following beneficial effects: 1. By integrating advanced data assimilation technology, the present invention can efficiently fuse real-time temperature observation data with mathematical models to achieve accurate estimation of the temperature field and heat source distribution inside the server cluster. This method not only improves the accuracy of state estimation, but also optimizes the system's ability to predict unmeasured states, ensuring that control decisions are made based on the most accurate information, greatly improving the real-time and reliability of system response.

[0016] 2. The present invention uses partial differential equations to model the thermal environment of the server cluster and develops an optimal control strategy based on this model. The present invention can achieve refined management of the temperature of the server cluster. The control strategy calculated by mathematical optimization methods directly affects the operation of the cooling system, such as fan speed and air conditioning settings, thereby maximizing energy efficiency and extending hardware life while ensuring server performance.

[0017] 3. The present invention can monitor the effect of environmental control in real time through the design of execution and feedback modules, and dynamically adjust the control strategy according to the feedback data. This closed-loop control mechanism not only ensures that the system can quickly adapt to changes in server load and environmental conditions, but also further optimizes energy management and environmental comfort through continuous learning and adaptation. This is particularly helpful in dealing with frequently changing operating conditions and growing cooling needs in large-scale data center environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the system module of the present invention; Figure 2 It is a schematic diagram of the data acquisition and preprocessing module of the present invention; Figure 3 It is a schematic diagram of the mathematical modeling and discretization module of the present invention; Figure 4 It is a schematic diagram of the data assimilation and state estimation module of the present invention; Figure 5 It is a schematic diagram of the optimal control solution module of the present invention; Figure 6 It is a schematic diagram of the execution and feedback module of the present invention; Figure 7 The figure is a flow chart of the control method steps of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1:

[0020] Please see attached Figure 1 The embodiment of the present invention provides an environmental control system for a large-scale server cluster, including: A data collection and preprocessing module, which is used to collect temperature observation data from multiple sensors set inside the server cluster and preprocess the data; Mathematical modeling and discretization module, which is used to build a convection-diffusion heat conduction model based on the internal structure and physical parameters of the server cluster, and discretize the model; The data assimilation and state estimation module is used to fuse the data processed by the data acquisition and preprocessing module with the discretization model of the mathematical modeling and discretization module to estimate the temperature field and heat source term distribution inside the server cluster; An optimal control solution module, which is used to calculate the control input of the cooling system based on the output of the data assimilation and state estimation module to optimize the temperature distribution of the server cluster; The execution and feedback module is used to implement the control strategy determined by the optimal control solution module and feed back the execution results to the data acquisition and preprocessing module.

[0021] Specifically, the present invention provides an environmental control system and a control method for a large-scale server cluster, aiming to accurately manage and optimize the temperature distribution within the server cluster through advanced data assimilation technology and optimal control strategy, thereby improving energy efficiency and protecting hardware from overheating damage.

[0022] The data acquisition and preprocessing module includes multiple temperature sensors distributed in key hot spots of the server cluster. These sensors collect temperature data in real time and send it to the synchronization unit for time alignment to ensure that all data points are processed on the same time basis. The time-aligned data enters the noise filtering unit, which uses digital filtering technology to remove high-frequency noise in the data and retain useful temperature information.

[0023] The mathematical modeling and discretization module is responsible for establishing the physical model of convection-diffusion heat conduction. The grid generation unit divides the internal area of ​​the server cluster into multiple small areas to form a calculation grid, while the time discretization unit divides the continuous time into discrete time steps for subsequent calculation processing.

[0024] The data assimilation and state estimation module uses an improved four-dimensional variational assimilation (4D-Var) method. This module combines the preprocessed data from the data acquisition and preprocessing module with the model from the mathematical modeling and discretization module to optimize the model parameters and initial states. The target functional construction unit constructs the target functional based on minimizing the error between the observed data and the model prediction, and the adjoint solution unit calculates the gradient of the target functional and back-propagates to optimize the model parameters.

[0025] The optimal control solution module calculates the optimal value of the control input based on the data assimilation results using the gradient descent method or other optimization algorithms. The gradient calculation unit first calculates the impact of the control variable on the system performance index, and then the control input update unit iteratively updates the control input until the best solution is found.

[0026] The execution and feedback module controls the signal output unit to convert the optimal control strategy into specific execution instructions, such as adjusting the fan speed, air conditioning temperature setting, etc., to implement temperature control. The execution results are collected by the feedback unit and sent back to the data acquisition and preprocessing module for data update and processing in the next cycle.

[0027] Please see attached Figure 2, the data acquisition and preprocessing module includes: A synchronization unit, which is used to time-align temperature data measured by multiple sensors within the server cluster; The noise filtering unit is used to filter the time-aligned temperature data.

[0028] Specifically, in this embodiment, the data acquisition and preprocessing module has two key components: a synchronization unit and a noise filtering unit. The synchronization unit is responsible for time synchronization of temperature data from different sensors within the server cluster to ensure that all data points are analyzed and processed on the same time standard. This step is crucial because time synchronization errors may lead to deviations in data analysis results and affect the final environmental control decision.

[0029] Specifically, the synchronization unit marks the received data with a timestamp and corrects any time deviation through an internal clock or a network time protocol (NTP) to achieve high-precision time synchronization. For example, some embodiments may use more advanced synchronization technologies, such as the IEEE-1588 Precision Time Protocol, to further improve the accuracy of synchronization.

[0030] The noise filtering unit is responsible for processing the time-aligned data and removing random noise or environmental noise from the data through various digital filtering techniques. In one possible implementation, the unit can apply a low-pass filter, a Kalman filter, or other statistical methods to reduce the impact of measurement errors and external interference. As an option, for extreme noise environments, this module can also integrate machine learning algorithms to dynamically adjust the filtering parameters to adapt to changing environmental conditions.

[0031] It should be noted that the synchronized and filtered data will provide reliable input for the subsequent data assimilation and state estimation modules, which are the basis for further optimizing the temperature control strategy by combining mathematical models with measured data. It is understandable that the accuracy and consistency of the data directly affect the effectiveness and response speed of the entire control system.

[0032] Please see attached Figure 3 , the mathematical modeling and discretization module includes: A grid generation unit, which is used to perform spatial grid division on the internal area of ​​the server cluster; The time discrete unit is used to process the temperature field evolution process in time segments.

[0033] Specifically, in this embodiment, the mathematical modeling and discretization module first constructs a corresponding heat conduction model according to the specific layout and physical characteristics of the server cluster. Specifically, the module includes two main functional units: a grid generation unit and a time discretization unit. The grid generation unit is responsible for dividing the physical space of the server cluster into a finite computational grid so that the heat conduction equation can be solved at these discrete grid points. Exemplarily, a finite element method can be used to generate a grid, which can adapt to complex server cluster geometries and ensure the accuracy of the calculation.

[0034] The time discrete unit discretizes the continuous time variable so that the temperature field can be calculated at each discrete time step. This process usually uses an explicit or implicit time integration scheme, such as the Euler method or the Crank-Nicolson method. In one possible implementation, the time discrete unit can select a higher-order time discretization strategy to improve the stability and accuracy of the solution.

[0035] To implement this module, a mathematical model was established based on the following convection-diffusion equation: ; Where T represents the temperature field, u represents the air flow velocity vector caused by the server fan and air conditioner, α is the thermal diffusivity, and q is the heat source term generated by the server hardware. is the Laplace operator of the temperature field T, representing the spatial effect of heat diffusion, is the rate of change of the temperature field T over time. It can be understood that the parameters in the model, such as α and u, can be adjusted according to the specific server configuration and operating conditions.

[0036] It should be noted that, through the above mathematical modeling and discretization processing, the present invention can provide strong theoretical support and calculation tools for the server cluster environment control system, ensuring that the system can achieve accurate temperature control and optimization in actual operation. In addition, the continuous optimization and updating of the model can cope with future technological development and new operation and maintenance requirements, ensuring the long-term effectiveness and flexibility of the system.

[0037] Please see attached Figure 4 , the data assimilation and state estimation module includes: A target functional construction unit, which is used to construct a target functional based on temperature observation data and model prediction data; An adjoint solving unit, which is used to solve the adjoint equation according to the target functional and the discretization model; The data assimilation and state estimation module further includes a parameter updating unit, which is used to update the heat source term distribution parameters according to the output of the adjoint solution unit, specifically including iterative optimization of the target functional (the target functional can be expressed as the following form): ; in, Indicates at location and the model-predicted temperature at time t, represents the corresponding observed temperature, β and is a regularization parameter used to control the smoothness of the solution and prevent overfitting, q is the heat source term, u is the control input, M is the number of sensors, and H is the length of the observation time window.

[0038] Specifically, in this embodiment, the data assimilation and state estimation module includes a target functional construction unit and an adjoint solution unit. The target functional construction unit is responsible for constructing an optimization problem based on real-time and historical temperature data and the prediction model obtained from the mathematical modeling and discretization module. This optimization problem aims to minimize the difference between the observed data and the model prediction, so as to accurately estimate the current temperature field and related parameters.

[0039] The adjoint solver is responsible for calculating the gradient of this objective functional and using the adjoint method to back-propagate the error, thereby optimizing the state variable T and the control variable u, as well as the heat source term q. The core advantage of this method is that it can effectively handle large-scale optimization problems and is suitable for complex server cluster environments.

[0040] As an option, this module can also integrate advanced data assimilation techniques, such as Kalman filtering or particle filtering, to further improve the accuracy and robustness of state estimation. In one possible implementation, an ensemble Kalman filter can be used to handle nonlinear and non-Gaussian problems, which is particularly important for dynamically changing server cluster environments.

[0041] It should be noted that through the application of these technologies, the data assimilation and state estimation module can not only provide real-time and accurate temperature field estimation, but also provide reliable input for the optimal control solution module, thereby achieving more effective environmental control strategies. In addition, the design and implementation of this module considers easy integration and expansion, supports future technology upgrades and the addition of new functions to cope with changing technical and operational requirements.

[0042] Please see attached Figure 5 , the optimal control solution module includes: A gradient calculation unit, which is used to calculate the gradient of the control input according to the state estimation result output by the data assimilation and state estimation module and the heat source term distribution parameter; a control input updating unit, configured to iteratively update the control input according to the gradient; The optimal control solution module further includes an optimal control strategy output unit, which is used to output a control strategy for controlling the cooling device.

[0043] Specifically, in this embodiment, the optimal control solution module includes a gradient calculation unit and a control input update unit. The gradient calculation unit uses the state and parameter estimates obtained from the data assimilation and state estimation module, such as temperature field and heat source distribution, to calculate the gradient of the control variable (such as cooling fan speed, air conditioning setting, etc.) by solving the corresponding optimization problem. This calculation is based on the goal of maximizing the system performance index or minimizing the energy consumption. The performance index usually used is the total energy consumption of the system or the temperature deviation of a specific area.

[0044] Specifically, the objective function can be defined as: ; Where J is the objective function, which is used to optimize the control input, is the temperature distribution predicted by the model, is the target temperature distribution, is the control input, λ and is a regularization parameter used to balance the temperature error and the control input cost, and T is the length of the observation or control time period.

[0045] The control input update unit iteratively updates the control strategy based on the calculated gradient. This process usually uses gradient descent or other advanced optimization algorithms, such as conjugate gradient method or quasi-Newton method, to ensure that the optimal solution is found quickly and accurately.

[0046] In some embodiments, the optimal control solution module also includes a real-time adjustment unit that can adjust the control strategy according to real-time data of environmental changes or server load. This dynamic adjustment capability enables the system to flexibly respond to sudden temperature peaks or equipment failures, further improving the reliability and efficiency of the system.

[0047] For example, an adaptive control algorithm can be implemented to allow the system to automatically adjust the regularization parameters λ and , to optimize the performance and energy efficiency of the control strategy. This adaptive algorithm is based on online learning or artificial intelligence technology, which can continuously learn from system operation data and improve the control strategy.

[0048] It should be noted that, through this detailed technical description, the optimal control solution module of the present invention can not only provide technical personnel with specific implementation solutions, but also inspire the development of more innovative control strategies and algorithms. In addition, the design of the module allows for easy integration into existing server management systems, providing a cost-effective and easy-to-implement solution to optimize the energy management and operation and maintenance efficiency of large-scale data centers. It is understandable that the development and application of such a module will have a profound impact on the sustainable operation of data centers.

[0049] Please see attached Figure 6 , the execution and feedback module includes: A control signal output unit, which is used to convert the optimal control strategy into a control signal for the cooling execution device; A feedback unit, which is used to transmit the regulated temperature data to the data acquisition and preprocessing module; The feedback unit is configured to periodically sample and update the input data of the data acquisition and preprocessing module.

[0050] Specifically, in this embodiment, the execution and feedback module includes a control signal output unit and a feedback unit. The control signal output unit is responsible for receiving the control decisions from the optimal control solution module and converting these decisions into specific operation instructions, such as adjusting the fan speed of the cooling system, the air conditioning temperature setting, etc. These instructions directly affect the physical operating environment of the server cluster and help maintain or reach the target temperature state.

[0051] Specifically, the control signal output unit may include multiple actuator interfaces for communicating with different types of physical control devices, such as fan controllers, air conditioning control systems, etc. These interfaces may use standard industrial communication protocols, such as Modbus or BACnet, to ensure compatibility and reliability with existing facilities.

[0052] The feedback unit is responsible for collecting actual environmental data after execution, such as temperature, humidity, etc., and monitoring these parameters in real time through sensors. The collected data is sent back to the data acquisition and preprocessing module to verify the effect of the control strategy and provide new inputs for the data assimilation and state estimation module, thereby realizing dynamic adjustment of the control strategy.

[0053] In some embodiments, the feedback unit also includes an evaluation mechanism for analyzing the effect of the control strategy in real time, such as by comparing the deviation between the expected temperature and the actual temperature. This evaluation result can be used to refine or adjust the control algorithm to improve the responsiveness and efficiency of the system.

[0054] For example, machine learning algorithms can be integrated into the system to enable the execution and feedback modules to self-optimize based on historical data and real-time feedback. This intelligent feedback mechanism can further improve the system's adaptive capabilities, optimize energy use and reduce maintenance requirements.

[0055] It should be noted that the design of this module takes into account the real-time and reliability of the system, ensuring that the control strategy is implemented quickly and accurately and adjusted in time to cope with the rapid changes in the data center environment. In addition, the implementation of the execution and feedback module supports the scalability and flexibility of the system, allowing the integration of future technologies, such as adding more types of sensors or control devices, to support a wider range of application scenarios.

[0056] It can be understood that, through the above detailed description, the execution and feedback module of the present invention provides a feasible and efficient technical solution for achieving accurate and effective server cluster environment control. This not only enhances the operating efficiency of the system, but also provides a solid foundation for maintaining the optimal working state of the equipment.

[0057] Embodiment 2: Please see attached Figure 7 , an environmental control method for a large-scale server cluster, comprising the following steps: S1. Temperature data collection: collect temperature observation data from sensors at multiple locations within the server cluster; S2, data assimilation process: the collected temperature observation data is input into the data assimilation and state estimation module for data assimilation to estimate the temperature field distribution and heat source items inside the server cluster; S3, control input calculation: according to the data assimilation results, the control input of the cooling system is calculated using the optimal control solution module; S4, execution and feedback: implement the calculated control input to adjust the cooling equipment, and feed back the adjustment results for data collection and preprocessing in the next cycle.

[0058] Specifically, in this embodiment, step S1 includes setting a plurality of sensors in key hot spots of the server cluster, such as near the CPU, above the memory module, and near the power supply unit, etc. These sensors can monitor and record the temperature data of their respective locations in real time, and each sensor is equipped with a high-precision temperature measurement device to ensure the accuracy and reliability of the data.

[0059] Specifically, sensors are usually thermocouples or thermistors, which are sensitive to changes in ambient temperature and can provide accurate temperature measurements over a wide temperature range. Each sensor is connected to the data acquisition system via a wired or wireless network to transmit measurement data in real time. As an option, the sensor can be configured with its own cache unit to temporarily store data when the network connection is unstable.

[0060] In some embodiments, step S1 also includes using data preprocessing techniques, such as data filtering and outlier detection, to improve data quality. These techniques help remove erroneous data caused by environmental noise and equipment failures, ensuring that only valid temperature readings are passed to the next link in the system. Data preprocessing can be performed on the sensor side or on a centralized data processing server.

[0061] For example, the collection frequency of temperature data can be set to be dynamically adjusted according to the workload of the server. When the server load is high, the collection frequency is increased to more accurately monitor possible rapid temperature changes; when the load is low, the collection frequency is reduced to save resources.

[0062] It should be noted that the temperature data collected through this step provides real-time and accurate basic information for the environmental control of the server cluster. In addition, the implementation of this step provides scalability for possible future technology upgrades, such as the introduction of more advanced sensor technology or more efficient data transmission solutions. It is understandable that the accurate collection and processing of these temperature data is crucial to achieving efficient environmental control in subsequent steps.

[0063] In this embodiment, the data assimilation and state estimation module uses advanced data assimilation technology to combine real-time temperature observation data with a pre-established heat conduction mathematical model to optimize model parameters and estimate state variables that are not directly observed. This process includes the following key steps: First, the module receives the preprocessed temperature data from step S1, which includes the temperature values ​​measured by each sensor and their corresponding timestamps. Specifically, the target functional construction unit establishes an optimization problem to minimize the difference between the model prediction and the actual observation.

[0064] In some embodiments, solving the adjoint equation involves calculating the gradient of the target functional with respect to model parameters and state variables, and this gradient information is used to guide the update of the parameters to improve the accuracy of the model prediction.

[0065] As an option, this module can also use advanced data assimilation techniques such as Kalman filters or ensemble Kalman filters, which are particularly suitable for dealing with dynamically changing systems and nonlinear problems. This method can adjust model parameters in real time in the continuously updated data stream to better reflect the actual system status.

[0066] It should be noted that, through the implementation of step S2, the present invention can ensure that the input of the control system reflects the actual current operation of the server cluster, thereby greatly improving the responsiveness and effectiveness of the environmental control strategy. In addition, the technical solution of this step provides strong flexibility and scalability, supporting the integration of more types of data assimilation technologies and algorithms in the future to adapt to new application requirements and challenges.

[0067] In this embodiment, step S3 is dedicated to using the accurate temperature field and heat source distribution estimation results obtained from the data assimilation and state estimation module. The optimal control solution module uses advanced optimization algorithms to calculate the required control inputs, which directly affect the operation of the cooling system, such as the fan speed and the temperature setting of the air conditioner. The calculation of the control strategy is based on an optimization problem, whose goal is to minimize energy consumption while ensuring that the server does not exceed the temperature threshold. This can be achieved by solving the following optimization problem: ; in, The optimization goal is to find a control strategy , so that the following integral expression is minimized, Indicates the integration of the following expression over the time interval [0, T], where T is the total time considered, Indicates control input The energy cost is is a weight parameter that controls the proportion of energy cost in the total cost, Represents the spatial region (i.e., all regions of the server cluster) are integrated, where is a loss function that measures the model's predicted temperature at location x and time t. With target temperature The deviation between represents the integral of the spatial variable x, represents the integral of the time variable t.

[0068] Specifically, an optimal control algorithm such as gradient descent or conjugate gradient method is used to iteratively solve the above optimization problem. The gradient calculation unit calculates the gradient of the objective function with respect to the control variable in each iteration step, and then the control input update unit adjusts the control input according to these gradients to gradually approach the optimal solution.

[0069] In some embodiments, this step also includes real-time monitoring of the performance of the algorithm to adjust the algorithm parameters or switch algorithm strategies. For example, if the current algorithm fails to converge to a satisfactory solution within a predetermined time, the system can automatically switch to another algorithm or dynamically adjust the regularization parameters. to improve performance.

[0070] As an option, the optimal control solution module can integrate machine learning technology to learn and predict the most effective control strategy from historical operation data. This adaptive control strategy can be dynamically adjusted according to real-time data to improve the system's response speed and control accuracy to environmental changes.

[0071] It should be noted that, through such fine control input calculation and optimization, the present invention can significantly improve the energy efficiency and operation efficiency of the data center, reduce energy consumption, and ensure the stability of the server environment and the optimization of server performance. In addition, the flexible design of this step and the application of advanced algorithms provide strong support for the expansion and upgrading of future technologies.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An environmental control system for a large-scale server cluster, characterized in that: include: A data collection and preprocessing module, which is used to collect temperature observation data from multiple sensors set inside the server cluster and preprocess the data; Mathematical modeling and discretization module, which is used to build a convection-diffusion heat conduction model based on the internal structure and physical parameters of the server cluster, and discretize the model; The data assimilation and state estimation module is used to fuse the data processed by the data acquisition and preprocessing module with the discretization model of the mathematical modeling and discretization module to estimate the temperature field and heat source term distribution inside the server cluster; An optimal control solution module, which is used to calculate the control input of the cooling system based on the output of the data assimilation and state estimation module to optimize the temperature distribution of the server cluster; The execution and feedback module is used to implement the control strategy determined by the optimal control solution module and feed back the execution results to the data acquisition and preprocessing module.

2. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The data acquisition and preprocessing module includes: A synchronization unit, which is used to time-align temperature data measured by multiple sensors within the server cluster; The noise filtering unit is used to filter the time-aligned temperature data.

3. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The mathematical modeling and discretization module includes: A grid generation unit, which is used to perform spatial grid division on the internal area of ​​the server cluster; The time discrete unit is used to process the temperature field evolution process in time segments.

4. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The data assimilation and state estimation module includes: A target functional construction unit, which is used to construct a target functional based on temperature observation data and model prediction data; An adjoint solving unit is used to solve the adjoint equation according to the target functional and the discretization model.

5. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The data assimilation and state estimation module further includes a parameter updating unit, which is used to update the heat source term distribution parameters according to the output of the adjoint solution unit, specifically including iterative optimization of the objective functional: ； in, Indicates at location and the model-predicted temperature at time t, represents the corresponding observed temperature, β and is a regularization parameter used to control the smoothness of the solution and prevent overfitting, q is the heat source term, u is the control input, M is the number of sensors, and H is the length of the observation time window.

6. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The optimal control solution module includes: A gradient calculation unit, which is used to calculate the gradient of the control input according to the state estimation result output by the data assimilation and state estimation module and the heat source term distribution parameter; A control input updating unit is used to iteratively update the control input according to the gradient.

7. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The optimal control solution module further includes an optimal control strategy output unit, which is used to output a control strategy for controlling the cooling device.

8. The environmental control system of a large-scale server cluster according to claim 1, characterized in that: The execution and feedback module includes: A control signal output unit, which is used to convert the optimal control strategy into a control signal for the cooling execution device; A feedback unit is used to transmit the regulated temperature data to the data acquisition and preprocessing module.

9. The environmental control system of a large-scale server cluster according to claim 8, characterized in that: The feedback unit is configured to periodically sample and update the input data of the data acquisition and preprocessing module.

10. A method for controlling the environment of a large-scale server cluster, according to the environmental control system of a large-scale server cluster according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Temperature data collection: collect temperature observation data from sensors at multiple locations within the server cluster; S2, data assimilation process: the collected temperature observation data is input into the data assimilation and state estimation module for data assimilation to estimate the temperature field distribution and heat source items inside the server cluster; S3, control input calculation: according to the data assimilation results, the control input of the cooling system is calculated using the optimal control solution module; S4, execution and feedback: implement the calculated control input to adjust the cooling equipment, and feed back the adjustment results for data collection and preprocessing in the next cycle.

Images