Cabinet-level computing power server dynamic temperature regulation and load optimization method

Abstract

The invention discloses a cabinet-level computing power server dynamic temperature regulation and load optimization method, and relates to the technical field of data centers, and the method comprises the following steps: collecting data, respectively building a thermodynamic model and a price prediction model, predicting a thermodynamic state, an electric power spot market price, and a renewable energy source supply condition, outputting and executing an optimal refrigeration system control plan by adopting an MPC algorithm; in combination with electric power spot market price data, cabinet-level computing power server energy consumption data and renewable energy supply data, the environment temperature of the cabinet-level computing power server and refrigeration system output are adjusted in real time through an MPC algorithm, and the future state of the cabinet-level computing power server can be accurately predicted. The MPC algorithm predicts the thermodynamic state of the cabinet-level computing power server and dynamically optimizes the temperature regulation strategy, the real-time performance is good, the temperature and the load can be adjusted in time, the stability and the energy efficiency of the cabinet-level computing power server are ensured, and safe and stable operation of the data center and the minimization of the energy consumption cost are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of data centers, and in particular to a method for dynamic temperature control and load optimization of a cabinet-level computing server. Background Art

[0002] A data center is a physical room, building or facility that hosts IT infrastructure. It is used to build, run and deliver applications and services, as well as store and manage data related to these applications and services. It is used to achieve centralized processing, storage, transmission, exchange and management of data information. It plays a key role in modern information technology infrastructure, providing reliable data storage and processing services for various organizations and enterprises. Data centers usually include large servers, network equipment, storage systems, cooling equipment and other infrastructure, designed to ensure that servers continue to operate efficiently while maintaining data security and reliability. These facilities use highly advanced technologies such as virtualization, cloud computing and automated management systems to improve the efficiency and scalability of data centers.

[0003] The main functions of a data center include data storage, processing, backup, and recovery, while providing high-speed network connections so that users can quickly access their information stored in the data center. In the digital age, data centers are critical to supporting enterprise operations, cloud services, big data analysis, and other aspects. Cabinet-level computing servers are key components within data centers. They provide the necessary computing power to process and store data. In order to ensure that cabinet-level computing servers operate stably, efficiently, and safely, while improving energy efficiency and user experience, it is usually necessary to perform temperature control and load optimization on cabinet-level computing servers.

[0004] Most of the existing temperature control and load optimization methods for cabinet-level computing servers have a single process and cannot accurately predict the future state of the cabinet-level computing servers, resulting in poor control and optimization effects, and poor real-time performance. It is impossible to achieve a sufficiently fast response speed, resulting in the inability to adjust the temperature and load in time, thereby affecting the stability and energy efficiency of the cabinet-level computing servers, and thus unable to ensure the safe and stable operation of the data center and the minimization of energy consumption costs. Therefore, the present invention proposes a dynamic temperature control and load optimization method for cabinet-level computing servers to solve the problems existing in the prior art. Summary of the invention

[0005] In response to the above problems, the purpose of the present invention is to propose a dynamic temperature control and load optimization method for a cabinet-level computing power server, so as to solve the problem that the existing temperature control and load optimization methods for cabinet-level computing power servers cannot accurately predict the future state of the cabinet-level computing power server, resulting in poor control and optimization effects, and cannot achieve a sufficiently fast response speed, resulting in the inability to adjust the temperature and load in time.

[0006] In order to achieve the purpose of the present invention, the present invention is implemented by the following technical scheme: a cabinet-level computing server dynamic temperature control and load optimization method, comprising the following steps:

[0007] Step 1: Collect electricity spot market price data, cabinet-level computing server energy consumption data, and renewable energy supply data, and perform data cleaning and preprocessing on the collected data;

[0008] Step 2: Based on the data collected in step 1, a thermodynamic model of the cabinet-level computing server and a price prediction model of the electricity spot market are established respectively;

[0009] Step 3: Predict the thermodynamic state of the cabinet-level computing server through the thermodynamic model, and predict the spot electricity market price and renewable energy supply through the price prediction model;

[0010] Step 4: Based on the current and predicted electricity spot market prices, renewable energy supply, and the thermodynamic state of the cabinet-level computing server, a multi-objective optimization algorithm is used to output the optimal cooling system control plan. The MPC algorithm is then used to modify the cooling system control plan in combination with the actual operating status of the system.

[0011] Step 5: According to the cooling system control plan output by the MPC algorithm in step 4, the ambient temperature of the cabinet-level computing server and the output of the cooling system are adjusted dynamically in real time.

[0012] Further improvements are as follows: in step one, the electricity spot market price data includes electricity spot market price fluctuation data and electricity spot market price peak-valley difference data, the cabinet-level computing server energy consumption data includes ambient temperature data, equipment heat generation data and refrigeration system energy consumption data, and the renewable energy supply data includes wind energy supply data and solar energy supply data.

[0013] A further improvement is that in step one, the specific steps of data cleaning and preprocessing of the collected data are: first, using a missing value processing algorithm to process missing values ​​in the data, then using an outlier detection algorithm to process outliers in the data, and finally standardizing the data.

[0014] Further improvements are as follows: in step 2, the specific steps for establishing the thermodynamic model of the cabinet-level computing power server are: first determine the scope that the model needs to cover and clarify the model objectives, then establish a physical model based on the collected energy consumption data of the cabinet-level computing power server, then use thermodynamic principles to analyze the thermal behavior of the server, and then convert the thermodynamic analysis into a mathematical model, that is, the thermodynamic model of the cabinet-level computing power server.

[0015] Further improvement lies in: in the step 2, the specific steps of establishing the price prediction model of the electricity spot market are: first extracting useful feature data from the collected electricity spot market price data and renewable energy supply data, then selecting the type of prediction model, and then training the selected prediction model with the extracted useful feature data, adjusting the model parameters to obtain the best performance, and obtaining the price prediction model of the electricity spot market.

[0016] A further improvement is that in step five, when the electricity price is at a peak or the renewable energy supply is lower than a preset value, the temperature setting of the cabinet-level computing server in the data center is increased, and the heat capacity of the data center building and the server cluster is used to absorb heat to reduce the energy consumption of the refrigeration system.

[0017] A further improvement is that in step five, when electricity prices are low or the supply of renewable energy is higher than a preset value, the output of the refrigeration system is increased, the ambient temperature of the cabinet-level computing server is reduced, and cold energy is reserved to cope with high-load periods.

[0018] A further improvement is that in step five, the deviation between the actual operation of the cabinet-level computing server and the control plan of the refrigeration system is continuously monitored and corrected in the next control cycle.

[0019] The beneficial effects of the present invention are as follows: the present invention combines the electricity spot market price data, the energy consumption data of the cabinet-level computing power server and the renewable energy supply data, and adjusts the ambient temperature and the output of the refrigeration system of the cabinet-level computing power server in real time through the MPC algorithm, and can accurately predict the future state of the cabinet-level computing power server, and improve the regulation and optimization effects. The MPC algorithm predicts the thermodynamic state of the cabinet-level computing power server and dynamically optimizes the temperature control strategy. It has good real-time performance and can adjust the temperature and load in time to ensure the stability and energy efficiency of the cabinet-level computing power server, ensure the safe and stable operation of the data center and minimize the energy consumption cost, and at the same time improve the absorption of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a flow chart of the dynamic temperature control and load optimization method of the cabinet-level computing server of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments 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.

[0023] See also Figure 1 This embodiment provides a method for dynamic temperature control and load optimization of a cabinet-level computing server, including the following steps:

[0024] Step 1: Data Collection

[0025] Collect electricity spot market price fluctuation data and electricity spot market price peak-valley price difference data as backup electricity spot market price data, collect cabinet-level computing server ambient temperature data, equipment heat data and cooling system energy consumption data as backup cabinet-level computing server energy consumption data, collect wind power supply data and solar energy supply data as backup renewable energy supply data, and provide a basis for subsequent model establishment by collecting electricity spot market price data, cabinet-level computing server energy consumption data and renewable energy supply data;

[0026] After the above data are collected, the missing values ​​in the data are first processed using the missing value processing algorithm, and then the outlier detection algorithm is used to process the outliers in the data. Finally, the data is standardized to achieve the cleaning and preprocessing of each data to ensure the quality and consistency of the data, thereby improving the accuracy and reliability of subsequent modeling;

[0027] Step 2: Model building

[0028] First, determine the scope that the thermodynamic model needs to cover. The coverage of the thermodynamic model in this embodiment includes the thermal environment of the entire cabinet-level computing server. At the same time, clarify the goal of the thermodynamic model, such as predicting the maximum temperature of the server and evaluating the efficiency of the heat dissipation system. Then, establish a physical model based on the energy consumption data of the cabinet-level computing server collected in step 1, including the layout of the internal components of the server, the air flow path and the cultivation of the refrigeration system. The physical model construction process is as follows: Analyze the collected cabinet-level computing server energy consumption data to find out the relationship between energy consumption and server operating status. According to the results of data analysis, put forward reasonable assumptions. This embodiment assumes that the power consumption of the server is proportional to the CPU utilization. Based on the proposed assumptions, establish a physical model of the energy consumption of the cabinet-level computing server, and its expression is as follows:

[0029] Pt=P s +P d +P c

[0030] Among them, Pt is the total power consumption of the cabinet-level computing server, P s is the static power consumption, that is, the power consumption of the cabinet-level computing server in idle state, P d is the dynamic power consumption, that is, the additional power consumption of the cabinet-level computing server when processing tasks, P c Cooling power consumption, that is, the power consumption of the cooling system required to maintain the normal operation of the cabinet-level computing server;

[0031] Then, the thermal behavior of the server is analyzed using the principles of thermodynamics, and then the thermodynamic analysis is converted into a mathematical model, that is, the thermodynamic model of the cabinet-level computing server is obtained. The conversion process of the mathematical model is as follows:

[0032] According to the thermodynamic analysis results, relevant data of the thermodynamic analysis server is obtained, including but not limited to the power consumption, size, material properties, ambient temperature and working parameters of the cooling system of the thermodynamic analysis server, and then the first law of thermodynamics (energy conservation) is applied to analyze the thermal behavior of the thermodynamic analysis server, calculate the heat generated by the thermodynamic analysis server in the working state, and analyze how the heat is transferred inside the thermodynamic analysis server and the surrounding environment, and then according to the thermodynamic principle, establish an energy balance equation as a data model, the model formula is as follows:

[0033] Qi=Q o +Q l

[0034] Among them, Qi is the heat generated inside the server according to the thermodynamic principle, and Q o is the heat dissipated by the cooling system, Q l is other heat loss;

[0035] Useful feature data are extracted from the electricity spot market price data and renewable energy supply data collected in step 1, including time series features, meteorological features, and supply and demand features. The type of prediction model is then selected. The selected prediction model is then trained with the extracted useful feature data, and the model parameters are adjusted to obtain the best performance. The prediction performance of the model is evaluated using the validation set and the test set. According to the results of the validation and test, the model is tuned to obtain the price prediction model of the electricity spot market. The model formula is as follows:

[0036] y t =c+φ 1 y t-1 +φ 2 y t-2 +...+φ p y t-p +θ 1 ∈ t-1 +θ 2 ∈t-2 +...+θ q ∈ t-q +∈ t

[0037] Among them, y t is the electricity price at time t, c is a constant term, φ i and θ j is the model parameter, ∈ t is the error term;

[0038] Step 3: Data prediction

[0039] In this embodiment, the thermodynamic model is iteratively solved by the finite element method until the convergence condition is met, and then the results obtained by the solution are analyzed to predict the thermodynamic state of the cabinet-level computing power server under different conditions, including temperature distribution and heat flux density;

[0040] Prepare the latest data for forecasting in advance, including historical electricity spot market prices and renewable energy production historical data, and forecast the prepared data through the price forecasting model to obtain electricity spot market prices and renewable energy supply conditions;

[0041] Step 4: Output the optimal refrigeration system control plan

[0042] Based on the current and predicted electricity spot market prices, renewable energy supply, and the thermodynamic state of the cabinet-level computing server, a multi-objective optimization algorithm is used to output the optimal refrigeration system control plan. The MPC (model predictive control) algorithm is then used to modify the refrigeration system control plan in combination with the actual operating state of the system. In each control cycle, the first control action generated by the MPC algorithm is executed, that is, adjusting the output of the refrigeration system.

[0043] The output optimal refrigeration system control plan includes:

[0044] Refrigeration system start and stop schedule: decide when to start or stop the refrigeration system based on cost and demand;

[0045] Refrigeration system operating parameters: Set the refrigeration system operating parameters, including temperature set point and fan speed, taking into account the load of the internal equipment of the cabinet-level computing server, and dynamically adjust the output of the refrigeration system to match the actual cooling needs;

[0046] Renewable energy scheduling: According to the supply of renewable energy, adjust the energy use strategy of the cooling system. When the supply of renewable energy is sufficient, give priority to using these energy sources to meet the cooling needs of the cabinet-level computing servers to reduce dependence on traditional energy.

[0047] Consider price fluctuations in the electricity spot market: During periods when the price difference between peak and valley electricity is large, rationally arrange the operation of the refrigeration system to reduce energy costs;

[0048] Consider external ambient temperature parameters: adjust the temperature settings inside the cabinet-level computing server according to changes in the external ambient temperature to improve the efficiency of the cooling system;

[0049] Step 5: Execute the refrigeration system control plan

[0050] According to the cooling system control plan output by the MPC algorithm in step 4, the ambient temperature of the cabinet-level computing server and the output of the cooling system are adjusted in real time to achieve dynamic temperature control and load optimization of the cabinet-level computing server. The deviation between the actual operation of the cabinet-level computing server and the cooling system control plan is continuously monitored and corrected in the next control cycle. The specific control is as follows:

[0051] When electricity prices are at their peak or renewable energy supply is lower than a preset value, the temperature setting of the data center's cabinet-level computing servers is increased, and the heat capacity of the data center's main building and server cluster is used to absorb heat to reduce the energy consumption of the cooling system;

[0052] When electricity prices are low or the supply of renewable energy is higher than the preset value, the output of the cooling system is increased, the ambient temperature of the cabinet-level computing server is reduced, and cold energy is reserved to cope with high-load periods.

[0053] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for dynamic temperature control and load optimization of a cabinet-level computing server, characterized in that: The following steps are involved: Step 1: Collect electricity spot market price data, cabinet-level computing server energy consumption data, and renewable energy supply data, and perform data cleaning and preprocessing on the collected data; Step 2: Based on the data collected in step 1, a thermodynamic model of the cabinet-level computing server and a price prediction model of the electricity spot market are established respectively; Step 3: Predict the thermodynamic state of the cabinet-level computing server through the thermodynamic model, and predict the spot electricity market price and renewable energy supply through the price prediction model; Step 4: Based on the current and predicted electricity spot market prices, renewable energy supply, and the thermodynamic state of the cabinet-level computing server, a multi-objective optimization algorithm is used to output the optimal cooling system control plan. The MPC algorithm is then used to modify the cooling system control plan in combination with the actual operating status of the system. Step 5: According to the cooling system control plan output by the MPC algorithm in step 4, the ambient temperature of the cabinet-level computing server and the output of the cooling system are adjusted dynamically in real time.

2. According to claim 1, a method for dynamic temperature control and load optimization of a cabinet-level computing server is characterized in that: In step one, the electricity spot market price data includes electricity spot market price fluctuation data and electricity spot market price peak-valley price difference data, the cabinet-level computing server energy consumption data includes ambient temperature data, equipment heat generation data and refrigeration system energy consumption data, and the renewable energy supply data includes wind energy supply data and solar energy supply data.

3. According to claim 1, a method for dynamic temperature control and load optimization of a cabinet-level computing server is characterized in that: In the step 1, the specific steps of data cleaning and preprocessing for the collected data are: firstly, using a missing value processing algorithm to process missing values ​​in the data, then using an outlier detection algorithm to process outliers in the data, and finally standardizing the data.

4. According to claim 1, a method for dynamic temperature control and load optimization of a cabinet-level computing server is characterized in that: In step 2, the specific steps for establishing the thermodynamic model of the cabinet-level computing server are: first determine the scope that the model needs to cover and clarify the model objectives, then establish a physical model based on the collected energy consumption data of the cabinet-level computing server, then use thermodynamic principles to analyze the thermal behavior of the server, and then convert the thermodynamic analysis into a mathematical model, that is, the thermodynamic model of the cabinet-level computing server.

5. According to claim 1, a method for dynamic temperature control and load optimization of a cabinet-level computing server is characterized in that: In the step 2, the specific steps for establishing a price prediction model for the electricity spot market are: first extract useful feature data from the collected electricity spot market price data and renewable energy supply data, then select the type of prediction model, and then train the selected prediction model through the extracted useful feature data, adjust the model parameters to obtain the best performance, and obtain the price prediction model for the electricity spot market.

6. A method for dynamic temperature control and load optimization of a cabinet-level computing server according to claim 1, characterized in that: In step five, when the electricity price is at a peak or the renewable energy supply is lower than a preset value, the temperature setting of the cabinet-level computing server of the data center is increased, and the heat capacity of the data center building and the server cluster is used to absorb heat to reduce the energy consumption of the refrigeration system.

7. A method for dynamic temperature control and load optimization of a cabinet-level computing server according to claim 1, characterized in that: In step five, when electricity prices are low or the supply of renewable energy is higher than a preset value, the output of the refrigeration system is increased, the ambient temperature of the cabinet-level computing server is reduced, and cold energy is reserved to cope with high-load periods.

8. The method for dynamic temperature control and load optimization of a cabinet-level computing server according to claim 1, characterized in that: In step five, the deviation between the actual operation status of the cabinet-level computing server and the refrigeration system control plan is continuously monitored and corrected in the next control cycle.

Images