MDC data center adaptive AI energy saving system for 5G edge calculation

By using the LinUCB and xgboost combination algorithm model for self-learning and self-optimization in the MDC data center, the problem of inefficient energy consumption management in the 5G edge computing environment is solved, and collaborative work between devices and uniform temperature control is achieved. Improved energy utilization efficiency and system intelligence level.

CN119012618APending Publication Date: 2024-11-22CHINA ACADEMY OF INFORMATION & COMM
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410948882.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The energy consumption management system of existing MDC data centers lacks real-time adaptability and intelligence, and cannot dynamically adjust according to highly dynamically changing loads in the 5G edge computing environment, resulting in inefficient energy management and inability to make full use of historical data and prediction information. , resulting in unnecessary waste of energy.

Method used

The cascaded algorithm model combined by LinUCB and xgboost is used for self-learning and self-optimization. Combined with the black box and white box models, the air conditioner and compressor and other equipment are controlled through AI to optimize the start position and temperature set point of the air conditioner and compressor. Parameters such as compressor capacity and fan speed realize coordinated work between equipment and uniform temperature control.

Benefits of technology

Real-time optimization of data center energy consumption in a 5G edge computing environment is achieved, energy utilization efficiency is improved, operation costs are reduced, mutual interference between equipment and temperature inhomogeneity are avoided, and the system's intelligence level is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119012618A_ABST
    Figure CN119012618A_ABST
Patent Text Reader

Abstract

The invention discloses an adaptive AI energy-saving system of an MDC (Mobile Data Center) for 5G edge computing. The first information input module is used for inputting characteristics and composite characteristics of use load power, air conditioner air supply temperature, air conditioner return air temperature, cabinet air supply temperature, cabinet return air temperature, sensor and air conditioner position information and the like. According to the MDC data center adaptive AI energy-saving system oriented to 5G edge calculation, characteristics such as load power, cabinet air supply temperature, cabinet air return temperature and compressor fan rotating speed are used as input of an algorithm; the air conditioner energy consumption, the deviation degree of the temperature and the target value and the uniformity degree of the temperature field are comprehensively considered to serve as evaluation or feedback of algorithm self-optimization. And secondly, the capacity of the compressor and the rotating speed of the fan are used as adjusting modes, so that the problem of mutual interference of control among the equipment is solved, and all components among the air conditioning equipment and in the equipment can work cooperatively, so that the energy-saving effect is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of AI energy-saving systems, and specifically to an MDC data center adaptive AI energy-saving system for 5G edge computing. Background Art

[0002] With the rapid development and popularization of 5G technology, edge computing, as a key technology supporting various real-time, low-latency applications, is gaining increasing attention. MDC, or micro data center, is an important infrastructure for edge computing. Its flexible deployment and rapid response effectively make up for the shortcomings of traditional large data centers. However, with the large-scale deployment of MDC data centers, its energy consumption problem has become increasingly prominent, which not only increases operating costs, but also puts considerable pressure on the environment. After searching, it is found that the energy-saving system in the prior art is typically such as the publication number CN117542183A, a data center energy consumption management system, including an energy consumption management platform, a switch, a 5G network and an energy consumption management module. The output end of the energy consumption management platform is connected to the switch, the switch is connected to the energy consumption management module, and the 5G network is connected to the switch; the energy consumption management module includes a first intelligent gateway, a substation multi-function meter, a distribution cabinet, a metering meter, a multi-circuit meter, a second intelligent gateway, a water meter and a cold and heat meter. The distribution cabinet is connected to a distribution cabinet metering meter, and the first intelligent gateway, the substation multi-function meter, the distribution cabinet metering meter, the metering meter, the multi-circuit meter, the metering meter, the second intelligent gateway, the water meter and the cold and heat meter are connected with an alarm mechanism. Its main features are that it can count and analyze the collected data, reasonably determine the economic indicators of building energy consumption and performance evaluation indicators in each area, and discover the patterns of energy use and energy waste. However, it has the disadvantages of limited flexibility, delayed response and low intelligence.

[0003] To sum up, although the rule-based data center energy management system can reduce energy consumption to a certain extent, it has the following shortcomings. First, it lacks real-time adaptability: the existing system mainly relies on preset static rules for energy management. These rules cannot be dynamically adjusted according to real-time load changes. In the 5G edge computing environment, the load of the data center changes highly dynamically, and the static rules cannot respond to these changes in time, resulting in inefficient energy management; second, the level of intelligence is insufficient. The existing system lacks advanced intelligent algorithm support and cannot make full use of historical data and predictive information. The system cannot self-learn and optimize energy management strategies. Long-term operation may cause unnecessary energy waste. In response to the above problems, existing equipment needs to be improved. Summary of the invention

[0004] The purpose of the present invention is to provide an MDC data center adaptive AI energy-saving system for 5G edge computing, so as to solve the problem that the existing system proposed in the above background technology mainly relies on preset static rules for energy consumption management, and these rules cannot be dynamically adjusted according to real-time load changes. In the 5G edge computing environment, the load of the data center changes highly dynamically, and the static rules cannot respond to these changes in time, resulting in inefficient energy management and the existing system lacks advanced intelligent algorithm support, and cannot make full use of historical data and predictive information. The system cannot self-learn and optimize energy management strategies, and may cause unnecessary energy waste in long-term operation.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an MDC data center adaptive AI energy-saving system for 5G edge computing, comprising:

[0006] First information input module: The first information input module is used to input characteristics such as load power, air conditioner supply air temperature, air conditioner return air temperature, cabinet supply air temperature, cabinet return air temperature, sensor and air conditioner location information and their composite characteristics;

[0007] First algorithm optimization module: The function of the first algorithm optimization module is to comprehensively consider the air conditioning power, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as evaluation or benefit feedback of the algorithm self-optimization. At the same time, the first algorithm optimization module and the first information input module belong to the MDC data center black box model solution;

[0008] Second information input module: The second information input module is used to input characteristics such as load power, cabinet supply air temperature, cabinet return air temperature, compressor fan speed or water valve opening and their combination characteristics;

[0009] Second algorithm optimization module: The function of the second algorithm optimization module is to comprehensively consider the air-conditioning energy consumption, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as evaluation or feedback for algorithm self-optimization. At the same time, the second algorithm optimization module and the second information input module belong to the MDC data center white box model solution.

[0010] Preferably, the first algorithm optimization module has adjustment means, and the adjustment means are the number of air conditioners turned on, the turn-on position, and the temperature setting point.

[0011] Preferably, the adjustment means of the first adjustment module is an algorithm-recommended action, and the adjustment means of the first adjustment module is controlled by AI.

[0012] Preferably, the algorithm used by the first adjustment module adopts a cascade algorithm model composed of LinUCB, which is the full name of Linear Upper Confidence Bound, and xgboost, wherein LinUCB can be used to process high-dimensional data and select candidate actions, and its linear model characteristics enable it to process data with linear relationships, and xgboost can be used to perform refined prediction or classification of candidate actions selected by LinUCB.

[0013] Preferably, the LinUCB is used to perform energy efficiency exploration and optimization, and the content of the LinUCB optimization exploration is expressed as the start-up position of the air conditioner and the number of air conditioners that are turned on.

[0014] Preferably, the xgboost is used for temperature regulation, which is manifested as automatically controlling the set point of an air conditioner.

[0015] Preferably, the second algorithm optimization module has an adjustment means, and the adjustment means is compressor capacity, fan speed or water valve opening.

[0016] Preferably, the adjustment mode of the second adjustment module is the recommended action of the algorithm, and the algorithm adopts a model predictive control model, namely, an MPC model.

[0017] Preferably, the MPC uses a prediction model to estimate future deviation values ​​of the process, and determines the current optimal input strategy through rolling optimization.

[0018] Preferably, the model predictive control model is optimized through Lagrange optimization, and the Lagrange optimization includes compressor speed 1, compressor speed 2 and compressor speed N, and the Lagrange optimization includes fan speed 1, fan speed 2 and fan speed N.

[0019] Compared with the prior art, the beneficial effects of the present invention are: the MDC data center adaptive AI energy-saving system for 5G edge computing,

[0020] (1) The present invention introduces a machine learning algorithm with self-learning and self-optimization capabilities, which is manifested in the use of a cascade algorithm model composed of LinUCB (full name: Linear Upper Confidence Bound) and xgboost. Among them, LinUCB can be used to process high-dimensional data and select candidate actions. Its linear model characteristics enable it to process data with linear relationships. xgboost can be used to make refined predictions or classifications of candidate actions selected by LinUCB, and process various complex nonlinear relationships through its learning ability. This combination method may enable the model to have better performance when processing tasks with high-dimensional features, nonlinear relationships, and tasks that require online learning. Considering that the server load power has large differences in different time periods, the load power is introduced as one of the input parameters. At the same time, it is difficult for many data centers to obtain outdoor temperature. Therefore, the air conditioner supply and return air temperature, cabinet supply and return air temperature, etc. are introduced so that the algorithm can indirectly learn the indoor changes caused by the outdoor temperature.

[0021] (2) Use machine learning algorithms to achieve unified group control of different air-conditioning components. This is manifested in taking load power, cabinet supply air temperature, cabinet return air temperature, compressor fan speed or water valve opening and other characteristics and their combined characteristics as the input of the algorithm, and comprehensively considering the air-conditioning energy consumption, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as the evaluation or feedback of the algorithm self-optimization. The compressor capacity, fan speed or water valve opening is used as the adjustment method, that is, the recommended action of the algorithm. This method not only solves the problem of mutual interference between the control devices, but also enables the various components between and within the air-conditioning devices to work together to achieve energy-saving effects. At the same time, energy efficiency and temperature are used as common indicators for evaluation or feedback, taking into account energy saving and temperature control to avoid the return air side temperature being too low; the algorithm adopts a model predictive control model, and MPC uses the prediction model to estimate the future deviation value of the process to determine the current optimal input strategy through rolling optimization. MPC not only considers the deviation value at the current moment and the past moment, but also predicts possible deviations in the future through the prediction model, thereby optimizing the control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the air conditioning device collaboration method of the MDC data center black box model solution of the present invention;

[0023] Figure 2 This is the core algorithm structure diagram of the MDC data center black box model solution of the present invention;

[0024] Figure 3 The air conditioning device collaboration mode of the MDC data center white box model solution of the present invention;

[0025] Figure 4This is the core algorithm structure diagram of the MDC data center white box model solution of the present invention. DETAILED DESCRIPTION

[0026] 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.

[0027] See also Figure 1-4 The present invention provides a technical solution: an MDC data center adaptive AI energy-saving system for 5G edge computing, comprising:

[0028] First information input module: The first information input module is used to input characteristics such as load power, air conditioner supply air temperature, air conditioner return air temperature, cabinet supply air temperature, cabinet return air temperature, sensor and air conditioner location information and their composite characteristics;

[0029] The first algorithm optimization module: The function of the first algorithm optimization module is to comprehensively consider the air conditioning power, the degree of deviation of the temperature from the target value, and the uniformity of the temperature field as evaluation or benefit feedback for algorithm self-optimization. At the same time, the first algorithm optimization module and the first information input module belong to the MDC data center black box model solution, which solves the existing solution. Due to the day and night differences in online business activity, the server working power in the MDC data center computer room is also significantly different, and the heat exchange efficiency on the condenser side varies in different time periods. At the same time, due to the business characteristics of the server, its heat dissipation requirements will also change with different time periods. The existing air conditioning control strategies of most data centers are relatively rough. Although the temperature of the computer room can be controlled within the normal range, due to the differences in the air conditioning heat exchange efficiency and the server heat dissipation requirements in different time periods, the workload of the air conditioner deviates from the optimal state in certain time periods;

[0030] The first algorithm optimization module has adjustment means, and the adjustment means are the number of air conditioners turned on, the turning-on position, and the temperature setting point.

[0031] The adjustment means of the first adjustment module is an algorithm-recommended action, and the adjustment means of the first adjustment module is controlled by AI.

[0032] The algorithm used by the first adjustment module adopts a cascade algorithm model composed of LinUCB, which is the full name of Linear Upper ConfidenceBound, and xgboost. Among them, LinUCB can be used to process high-dimensional data and select candidate actions. The characteristics of its linear model enable it to process data with linear relationships. xgboost can be used to perform refined prediction or classification of candidate actions selected by LinUCB.

[0033] LinUCB is used for energy efficiency exploration and optimization, and the content of LinUCB optimization exploration is the power-on position of the air conditioner and the number of air conditioners turned on.

[0034] XGBoost is used for temperature regulation, which is manifested by automatically adjusting the set point of the air conditioner.

[0035] Second information input module: The second information input module is used to input characteristics such as load power, cabinet supply air temperature, cabinet return air temperature, compressor fan speed or water valve opening and their combination characteristics;

[0036] Second algorithm optimization module: The function of the second algorithm optimization module is to comprehensively consider the air conditioning energy consumption, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as the evaluation or feedback of the algorithm self-optimization. At the same time, the second algorithm optimization module and the second information input module belong to the MDC data center white box model solution, which solves the problem that the micro-modules of the existing MDC data center converge to periodic fluctuations due to the mutual interference between the air-conditioning and the nearly air-conditioning during operation. In addition, the micro-module adopts air supply control, but due to the changes in the natural environment temperature and load power, the temperature of the return air side will be high and low in different time periods. Under these current conditions, the computer room has the following problems: When the temperature on the return air side is low, it will lead to a waste of cooling capacity, causing the module to work in a higher pPUE state. At the same time, the air-conditioning components are controlled by the individual air-conditioning itself, and the air-conditioning interacts with each other to form competitive operation, which will also converge to periodic fluctuations. In this case, the temperature will continue to change regionally, which may bring safety hazards.

[0037] The second algorithm optimization module has an adjustment means, and the adjustment means is the compressor capacity, the fan speed or the water valve opening.

[0038] The adjustment mode of the second adjustment module is the recommended action of the algorithm, and the algorithm adopts the model predictive control model, namely the MPC model.

[0039] MPC uses predictive models to estimate future deviations of the process and uses rolling optimization to determine the current optimal input strategy.

[0040] The model predictive control model is optimized through Lagrange optimization, and the Lagrange optimization includes compressor speed 1, compressor speed 2 and compressor speed N. At the same time, the Lagrange optimization includes fan speed 1, fan speed 2 and fan speed N.

[0041] In summary, 5G edge computing requires data centers to be able to quickly respond to changing computing loads. Therefore, the patent of this invention mainly solves how to effectively reduce the energy consumption of data centers, improve energy utilization efficiency, and achieve dynamic optimization management of resources through intelligent means while ensuring performance requirements. This not only helps to reduce the operating costs of data centers, but also has the advantages of environmental protection and sustainable development.

[0042] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive AI energy-saving system for MDC data centers for 5G edge computing, characterized in that: include: First information input module: The first information input module is used to input the load power, air conditioner supply air temperature, air conditioner return air temperature, cabinet supply air temperature, cabinet return air temperature, sensor and air conditioner location information characteristics and their composite characteristics; First algorithm optimization module: The function of the first algorithm optimization module is to comprehensively consider the air conditioning power, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as evaluation or benefit feedback of the algorithm self-optimization. At the same time, the first algorithm optimization module and the first information input module belong to the MDC data center black box model solution; Second information input module: The second information input module is used to input load power, cabinet supply air temperature, cabinet return air temperature and compressor fan speed or water valve opening characteristics and their combination characteristics; Second algorithm optimization module: The function of the second algorithm optimization module is to comprehensively consider the air-conditioning energy consumption, the degree of deviation between the temperature and the target value, and the uniformity of the temperature field as evaluation or feedback for algorithm self-optimization. At the same time, the second algorithm optimization module and the second information input module belong to the MDC data center white box model solution.

2. According to claim 1, an MDC data center adaptive AI energy-saving system for 5G edge computing is characterized in that: The first algorithm optimization module has adjustment means, and the adjustment means are the number of air conditioners turned on, the turning-on position, and the temperature setting point.

3. According to the MDC data center adaptive AI energy-saving system for 5G edge computing according to claim 1, it is characterized by: The adjustment means of the first adjustment module is an algorithm recommended action, and the adjustment means of the first adjustment module is controlled by AI.

4. According to claim 1, an adaptive AI energy-saving system for MDC data centers for 5G edge computing is characterized in that: The algorithm used by the first adjustment module adopts a cascade algorithm model composed of LinUCB, which is the full name of Linear Upper Confidence Bound, and xgboost, where LinUCB is used to process high-dimensional data and select candidate actions. The characteristics of its linear model enable it to process data with linear relationships, and xgboost is used to perform refined prediction or classification of candidate actions selected by LinUCB.

5. According to claim 1, an MDC data center adaptive AI energy-saving system for 5G edge computing is characterized in that: The LinUCB is used to perform energy efficiency exploration and optimization, and the content of the LinUCB optimization exploration is represented by the start-up position of the air conditioner and the number of air conditioners that are turned on.

6. According to claim 1, an MDC data center adaptive AI energy-saving system for 5G edge computing is characterized in that: The xgboost is used for temperature regulation, which is manifested as automatically adjusting the set point of the air conditioner.

7. According to claim 1, an MDC data center adaptive AI energy-saving system for 5G edge computing is characterized in that: The second algorithm optimization module has an adjustment means, and the adjustment means is the compressor capacity, the fan speed or the water valve opening.

8. The MDC data center adaptive AI energy-saving system for 5G edge computing according to claim 1 is characterized in that: The adjustment mode of the second adjustment module is the recommended action of the algorithm, and the algorithm adopts a model predictive control model, namely, an MPC model.

9. The MDC data center adaptive AI energy-saving system for 5G edge computing according to claim 1 is characterized in that: The MPC uses a prediction model to estimate future deviation values ​​of the process and determines the current optimal input strategy through rolling optimization.

10. The MDC data center adaptive AI energy-saving system for 5G edge computing according to claim 8, characterized in that: The model predictive control model is optimized through Lagrange optimization, and the Lagrange optimization includes compressor speed 1, compressor speed 2 and compressor speed N, while the Lagrange optimization includes fan speed 1, fan speed 2 and fan speed N.