A dynamic air conditioning regulation method and system based on IT load prediction

Abstract

The application provides a dynamic air conditioner adjusting method and system based on IT load prediction. The system uses operation terminal performance parameters, computing power scheduling index characteristics, equipment comprehensive performance data, business environment correlation characteristics and corresponding IT load conditions to construct an IT load prediction model. The constructed IT load prediction model predicts real-time IT load conditions. An air conditioner dynamic adjustment control model is constructed based on the IT load conditions, air conditioner dynamic adjustment indexes and the dynamic air conditioner adjusting method. The final dynamic air conditioner adjusting method is obtained by processing the real-time IT load conditions and corresponding air conditioner dynamic adjustment indexes based on the air conditioner dynamic adjustment control model. The air conditioner state is adjusted based on the final dynamic air conditioner adjusting method. The temperature difference control range of the machine room is within 0.8 DEG C by combining double machine learning models. The two constructed models realize real-time control of IT load prediction and air conditioner dynamic adjustment, and improve the high-precision control efficiency and high-efficiency energy-saving performance of the air conditioner system.

Description

Technical Field

[0001] The energy-saving refrigeration and air conditioning equipment of the present invention belongs to the field of high-efficiency energy-saving air conditioning technology, and particularly relates to a dynamic air conditioning adjustment method and system based on IT load prediction. Background Technology

[0002] With rapid economic development, a large number of data storage server centers have emerged. The electronic equipment in these centers generates significant heat during operation, requiring appropriate environmental control equipment for cooling. The environment within the data storage server center directly impacts the reliability of the electronic equipment and the efficiency of data storage and computation. During operation, the temperature of the electronic equipment in the data storage servers needs to be constantly adjusted; poor control can easily lead to the burnout of electronic components. The varying data storage and computational power consumption of different servers within a data storage server center result in poor performance when setting constant environmental parameters, further burdening the electronic equipment in the data center, wasting energy, and increasing the operating costs of the data center.

[0003] Most existing systems use electronic monitoring of the data center environment, but fail to monitor the underlying server operation, i.e., how to predict IT load in real time and generate dynamic air conditioning control methods. As a result, environmental out-of-control situations can easily occur on the servers, which in turn affects the normal use of electronic components.

[0004] To maintain a constant indoor temperature in industrial and information technology data centers, a key feature of data center air conditioning is year-round cooling. When the data center temperature is lower than the indoor temperature, it can utilize outdoor natural cooling sources directly or indirectly. Directly utilizing natural cooling sources is limited by factors such as air quality and geographical location. Indirectly using outdoor natural cooling sources is currently a common solution for data center air conditioning. However, in practice, because the parameters for switching operating modes are relatively fixed, the energy-saving effect cannot be flexibly realized when there is partial load or load changes. Manual adjustment of operating parameters may be necessary to assist in improving energy efficiency, resulting in low control efficiency and operational difficulties.

[0005] Therefore, there is a need in this field for a dynamic air conditioning regulation method and system based on IT load prediction. The system requires optimal selection of the utilized parameter features and the construction of an IT load prediction model using IT load conditions. The choice of which IT load prediction model to construct to predict real-time IT load conditions needs further consideration. How to construct a dynamic air conditioning regulation and control model based on the predicted IT load conditions, process real-time IT load conditions and corresponding dynamic air conditioning regulation indices to obtain the final dynamic air conditioning regulation method, and adjust the air conditioning state according to the final dynamic air conditioning regulation method still faces certain difficulties in the existing technology. Furthermore, how to combine dual machine learning models to control changes in computer room temperature differences, improve the high-precision control efficiency and energy-saving performance of the air conditioning system, and further address the challenges in classifying and acquiring dynamic air conditioning regulation methods for precise temperature difference control in computer rooms, selecting appropriate machine learning models, and improving the model settings to adapt to IT load prediction parameter features for dynamic air conditioning regulation under IT load scenarios remains an issue. Finally, how to make the application model obtained from the dynamic air conditioning regulation method under IT load prediction scenarios more robust and generalizable, and how to build a green and efficient system for high-precision instrument temperature control, reducing the cost and energy consumption of manual inspections, requires further creative method design. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a dynamic air conditioning adjustment method and system based on IT load prediction.

[0007] A first aspect of the present invention provides a dynamic air conditioning adjustment method based on IT load prediction, the method comprising:

[0008] The first computing terminal performance parameters, the first computing power scheduling index characteristics during computing tasks, the first device comprehensive performance data at this time, and the first business environment correlation characteristics are obtained, along with the first IT load situation corresponding to the above characteristics.

[0009] The first IT load prediction feature is obtained by processing the first computing terminal performance parameters, the first computing power scheduling index features, the first device comprehensive performance data, and the first business environment correlation features.

[0010] An IT load prediction model is constructed using the first IT load prediction characteristics and the corresponding first IT load situation at this time.

[0011] The second computing terminal performance parameters, the second computing power scheduling index characteristics during the computing task, the comprehensive performance data of the second device at this time, and the second business environment correlation characteristics of the receiving computing terminal are processed to obtain the second IT load prediction characteristics. The IT load prediction model processes the second IT load prediction characteristics to generate the second IT load situation. Based on the second IT load situation, the air conditioner adjusts the temperature of the computing terminal environment.

[0012] Furthermore, the performance parameters of the first or second computing terminal are calculated by considering the comprehensive CPU utilization, memory usage, disk read / write volume, and network traffic characteristics of the computing terminal.

[0013] Furthermore, the first computing power scheduling index feature or the second computing power scheduling index feature is calculated based on the amount of data retrieved by the computing terminal.

[0014] Furthermore, the comprehensive performance data of the first device or the comprehensive performance data of the second device includes the comprehensive power of the computing terminal, the UPS load rate, and the operating voltage.

[0015] Furthermore, the first business environment related feature or the second business environment related feature is obtained by processing external customer data access volume, climate temperature difference coefficient and computing terminal environmental humidity.

[0016] Furthermore, the first IT load prediction feature is obtained by horizontally splicing the first computing terminal performance parameters, the first computing power scheduling index feature, and the first business environment correlation feature, and vertically splicing them with the first device comprehensive performance data.

[0017] Furthermore, the IT load prediction model is constructed using a deep learning model, and its activation function is:

[0018]

[0019] In the formula, For the activation function value, The input is the IT load prediction feature.

[0020] Furthermore, an air conditioning dynamic adjustment control model is constructed based on IT load conditions, air conditioning dynamic adjustment index, and dynamic air conditioning adjustment method. The air conditioning dynamic adjustment index is calculated from air supply speed, air supply volume, and air supply humidity.

[0021] Furthermore, the air conditioning dynamic adjustment and control model adopts an improved support vector machine model based on computing terminal performance parameters, computing power scheduling index characteristics, and business environment correlation characteristics. Its decision boundary hyperplane is:

[0022]

[0023] In the formula, This represents the product of the feature values ​​of the computing terminal performance parameter, computing power scheduling index feature, and business environment correlation feature corresponding to the nth training feature vector in the IT load prediction feature dataset used to train the model, where m is the number of IT load prediction feature datasets used for training. , Let be the normal vector and intercept of the hyperplane, respectively. This is a combined feature vector of the IT load status and the air conditioning dynamic adjustment index.

[0024] This invention provides a dynamic air conditioning control system based on IT load forecasting. The system utilizes a dynamic air conditioning control method based on IT load forecasting. The system includes a computing terminal performance acquisition module, a computing power scheduling index feature acquisition module, a comprehensive equipment performance data acquisition module, a business environment correlation feature acquisition module, an IT load forecasting model construction module, and an air conditioning dynamic control model construction module.

[0025] The computing terminal performance acquisition module is used to collect computing terminal performance parameters;

[0026] The computing power scheduling index feature acquisition module is used to obtain computing power scheduling index features when calculating tasks.

[0027] The equipment comprehensive performance data acquisition module is used to acquire equipment comprehensive performance data.

[0028] The business environment association feature acquisition module is used to process and acquire business environment association features.

[0029] The IT load prediction model construction module: obtains the IT load situation, and processes the computing terminal performance parameters, the computing power scheduling index characteristics, the equipment comprehensive performance data and the business environment correlation characteristics to obtain IT load prediction characteristics, and constructs an IT load prediction model using the IT load prediction characteristics and the IT load situation;

[0030] The air conditioning dynamic adjustment and control model construction module: uses the performance parameters of the computing terminal, the computing power scheduling index features, and the business environment correlation features to construct an improved support vector machine model, and uses the improved support vector machine model to obtain a real-time air conditioning dynamic adjustment method.

[0031] This invention utilizes system performance parameters of computing terminals, computing power scheduling index characteristics, comprehensive equipment performance data, and business environment correlation characteristics and corresponding IT load conditions to construct an IT load prediction model. This model predicts real-time IT load conditions. Based on the IT load conditions, air conditioning dynamic adjustment index, and dynamic air conditioning adjustment method, an air conditioning dynamic adjustment control model is constructed. This model processes the real-time IT load conditions and corresponding dynamic air conditioning adjustment index to obtain the final dynamic air conditioning adjustment method. The air conditioning status is adjusted according to this final method. This invention combines dual machine learning models to achieve precise temperature difference control within a range of 0.8℃ in the computer room. The combination of the two models enables real-time control of IT load prediction and dynamic air conditioning adjustment, improving the high-precision control efficiency and energy-saving performance of the air conditioning system. Furthermore, to achieve precise temperature difference control in the computer room, the support vector machine obtained through classification of the dynamic air conditioning adjustment method is improved with enhanced parameters for IT load prediction. This makes the dynamic air conditioning adjustment method in IT load prediction scenarios more robust and generalizable, enabling the construction of a green and efficient system for high-precision instrument temperature control. This reduces the cost and energy consumption of manual inspections, providing significant technological value for carbon neutrality goals. Attached Figure Description

[0032] Figure 1 This is a flowchart of the IT load prediction model construction process of the present invention;

[0033] Figure 2 This is a schematic diagram of a dynamic air conditioning control system based on IT load prediction according to the present invention.

[0034] Figure 3 This is a diagram illustrating the classification principle of the SVM classifier in this invention; Detailed Implementation

[0035] This invention proposes a dynamic air conditioning adjustment method and system based on IT load prediction. The invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0036] The air conditioner that implements the dynamic air conditioning adjustment method based on IT load prediction according to this invention belongs to the energy-saving refrigeration and air conditioning equipment, which belongs to the high-efficiency energy-saving industry.

[0037] First embodiment of the present invention:

[0038] A first aspect of the present invention provides a dynamic air conditioning adjustment method based on IT load prediction, the method comprising:

[0039] The first computing terminal performance parameters, the first computing power scheduling index characteristics during computing tasks, the first device comprehensive performance data at this time, and the first business environment correlation characteristics are obtained, along with the first IT load situation corresponding to the above characteristics.

[0040] The first IT load prediction feature is obtained by processing the first computing terminal performance parameters, the first computing power scheduling index features, the first device comprehensive performance data, and the first business environment correlation features.

[0041] An IT load prediction model is constructed using the first IT load prediction characteristics and the corresponding first IT load situation at this time.

[0042] The second computing terminal performance parameters, the second computing power scheduling index characteristics during the computing task, the comprehensive performance data of the second device at this time, and the second business environment correlation characteristics of the receiving computing terminal are processed to obtain the second IT load prediction characteristics. The IT load prediction model processes the second IT load prediction characteristics to generate the second IT load situation. Based on the second IT load situation, the air conditioner adjusts the temperature of the computing terminal environment.

[0043] Furthermore, the performance parameters of the first or second computing terminal are calculated based on the comprehensive CPU utilization, memory usage, disk read / write volume, and network traffic characteristics of the computing terminal.

[0044]

[0045] In the formula, The above refers to either the performance parameters of the first or second computing terminal, where n is the number of computing terminals (in units). Let i represent the memory usage characteristics of the i-th processing terminal. Let be the CPU utilization of the i-th computing terminal. and For percentage data, Let be the disk throughput (data transfer volume) of the i-th computing terminal. Let be the number of disk read / write operations for the i-th processing terminal. Let T be the network traffic characteristics of the i-th computing terminal, and T be the network traffic characteristics of the i-th computing terminal. The time required.

[0046] In this embodiment, CPU load refers to the level of activity of the processor when performing computational tasks, typically expressed as CPU utilization (0%-100%). High CPU load can lead to decreased processing speed or even overheating. Modern multi-core processors require individual monitoring of the load on each core to avoid an imbalance where one core is overloaded while other cores are idle.

[0047] Memory load: The ratio of physical memory to virtual memory usage in the system. When memory load is too high, the system will frequently perform memory page swapping, leading to a significant performance degradation. Memory load management is particularly critical because insufficient memory often has a more direct and severe impact on system performance than CPU overload.

[0048] Disk I / O load reflects the frequency and speed of disk read / write operations, including metrics such as read / write operations per second (IOPS) and throughput (MB / s). Existing technologies only use throughput for calculation, while this invention uses read / write operations to modify throughput to obtain disk read / write characteristics for accurate acquisition of computing terminal performance parameters. High disk load can lead to response latency, impacting the performance of data-intensive applications. Traditional hard disk drives (HDDs) and solid-state drives (SSDs) exhibit significant differences in load performance; storage media characteristics must be considered during evaluation.

[0049] Network load: This refers to the data transmission volume and bandwidth utilization of network interfaces, including throughput, latency, and packet loss rate. Excessive network load can lead to communication delays, impacting the responsiveness of distributed systems and cloud services. These are the main factors affecting IT load.

[0050] Furthermore, the first computing power scheduling index feature or the second computing power scheduling index feature is calculated based on the amount of data retrieved by the computing terminal:

[0051]

[0052] In the formula, This refers to either the first computing power scheduling index feature or the second computing power scheduling index feature. This indicates the dimension of the data retrieved by the computing terminal. The exponential correction method for retrieving data from the computing terminal is rounded up, and k represents the computing power scheduling correction coefficient.

[0053] In this embodiment, if the amount of data retrieved by the computing terminal is... ,but Values This is rounding up.

[0054] In this embodiment, to reasonably store and utilize the large amount of data, the computing terminal typically generates scheduling tasks that handle massive amounts of data. This data possesses specific dimensions and quantities, and its scheduling within the computing terminal impacts the overall IT load. The computing power scheduling correction coefficient in this embodiment is generally between 0.05 and 0.15, representing the optimal coefficient obtained experimentally in this invention. The corrected computing power scheduling index characteristics better match the input vector combination characteristics trained by the IT load prediction model for this scenario.

[0055] Furthermore, the comprehensive performance data of the first device or the comprehensive performance data of the second device includes the comprehensive power of the computing terminal, the UPS load rate, and the operating voltage.

[0056] The overall power is calculated from the average power of the computing terminals, which is the average working power of the computing terminals in the computer room. The working voltage is taken as the rated value of the equipment performance index.

[0057] UPS load factor refers to the ratio of the actual load power carried by an uninterruptible power supply (UPS) to its rated power, usually expressed as a percentage. This indicator directly reflects the UPS's operating status and efficiency, and is a key parameter for evaluating UPS operational health, optimizing configuration, and extending equipment life. Therefore, UPS systems typically also have a certain impact on IT load. Here, the UPS load factor is obtained by the ratio of the actual load power of the UPS maintaining the computing terminals to the UPS power supply itself.

[0058] The overall power, UPS load rate, and operating voltage form a three-dimensional feature vector, which represents the overall performance data of the equipment.

[0059] Furthermore, the first or second business environment-related feature is obtained by processing external customer data access volume, climate temperature difference coefficient, and ambient humidity of the computing terminal, and its calculation formula is as follows:

[0060]

[0061] Because the difference between the data access volume and other feature data after removing the dimensions is too large, the logarithmic function is used to reduce the dimensionality of the data volume and reduce the impact of the data input to the training model. In addition, since the customer data access volume affects the data storage of the computing terminal, the larger the data storage volume, the more the IT load is affected. For the computing terminal data storage, which is a precision instrument, temperature and humidity changes will affect the IT load. The smaller the deviation of the change, the smaller the impact on the IT load.

[0062] In the formula, Let m be either the first or second business environment-related feature, and m be the number of external customers whose data needs to be stored. Let j be the data access volume of the j-th external customer. This represents the highest temperature of the day. This is the lowest temperature of the day. The optimal operating temperature for a computing terminal is generally 25℃. For the ambient humidity of the computing terminal, The optimal humidity for the operation of the computing terminal is generally 50%.

[0063] Furthermore, the first IT load prediction feature is obtained by horizontally splicing the first computing terminal performance parameters, the first computing power scheduling index feature, and the first business environment correlation feature, and vertically splicing them with the first device comprehensive performance data.

[0064] To ensure that the feature vectors input to the IT load prediction model are easily processed by the model, the features are creatively concatenated according to the correct acquisition of model processing results discovered in this invention. The vector form of the IT load prediction features in this embodiment can be expressed as follows:

[0065]

[0066] Furthermore, the IT load prediction model is constructed using a deep learning model, and its activation function is:

[0067]

[0068] In the formula, For the activation function value, The input is the IT load prediction feature.

[0069] A neural network is a computational model composed of a large number of interconnected neurons. Inspired by the human nervous system, it can achieve a highly adaptive and nonlinear mapping from input data to output results through the combination and training of multiple layers of neurons.

[0070] Neural networks typically consist of multiple layers, including an input layer, hidden layers, and an output layer. The input layer receives input data, while the hidden and output layers are responsible for calculating the output. Each neuron receives multiple inputs from the previous layer, which are weighted and calculated, then a bias is applied, followed by a nonlinear transformation through an activation function to produce the final output. Connections between neurons are usually represented by weights, where the weight values ​​represent the strength of the connection. These weights can be updated during training to adjust the connection strength between neurons. Neural networks possess strong adaptability and nonlinear mapping capabilities, allowing them to adapt to diverse input data and complex problems. In practical applications, neural networks have been widely used in image recognition, speech recognition, natural language processing, and intelligent control, achieving considerable success.

[0071] In this embodiment, the IT load is defined by the technology in the art and the data values ​​calculated by the neural network according to the numerical range. This is a conventional technical setting for classification in the field of pattern recognition by those skilled in the art, and is set according to the actual IT load of the computing terminal by those skilled in the art.

[0072] Furthermore, an air conditioning dynamic adjustment control model is constructed based on IT load conditions, the air conditioning dynamic adjustment index, and the dynamic air conditioning adjustment method. The air conditioning dynamic adjustment index is calculated from the supply air velocity, supply air volume, and supply air humidity.

[0073]

[0074] Those skilled in the art know that the operation of industrial variable frequency air conditioners at different frequencies is related to their regulation efficiency under extreme temperatures. High-frequency operation can improve regulation efficiency under extreme temperatures. However, in order to save energy, starting the variable frequency can effectively solve the dual control of energy consumption and regulation efficiency, and obtain high-precision temperature control for the air conditioner of the computing terminal.

[0075] In the formula, f is the current frequency of the inverter. This is the maximum inverter frequency of the air conditioner. This is the minimum inverter frequency for the air conditioner. The air supply velocity is f. Let f be the air volume at frequency f. Since the air volume has a larger dimension compared to other feature vector data, a logarithmic function is used here to reduce its data dimensionality. The supply air humidity is at frequency f. This indicates the number of times the frequency changes from the maximum to the minimum.

[0076] In this embodiment, The range is 1.5m / s-3.5m / s. The air volume indicated is generally in the range of around 10,000 m³ / h. The range is between 40% and 50%.

[0077] The IT load status includes the first IT load prediction feature and the second IT load prediction feature. The air conditioning dynamic adjustment index includes the first air conditioning dynamic adjustment index corresponding to the first IT load prediction feature and the second air conditioning dynamic adjustment index corresponding to the second IT load prediction feature. The dynamic air conditioning adjustment method includes the first dynamic air conditioning adjustment method corresponding to the first IT load prediction feature and the second dynamic air conditioning adjustment method corresponding to the second IT load prediction feature.

[0078] Furthermore, the air conditioning dynamic adjustment and control model adopts an improved support vector machine model based on computing terminal performance parameters, computing power scheduling index characteristics, and business environment correlation characteristics. Its decision boundary hyperplane is:

[0079]

[0080] In the formula, This represents the product of the feature values ​​of the computing terminal performance parameter, computing power scheduling index feature, and business environment correlation feature corresponding to the nth training feature vector in the IT load prediction feature dataset used to train the model, where m is the number of IT load prediction feature datasets used for training. , Let be the normal vector and intercept of the hyperplane, respectively. This is a combined feature vector of the IT load status and the air conditioning dynamic adjustment index.

[0081] To make the real-time dynamic adjustment and control of air conditioning more accurate after IT load prediction using support vector machines, and to improve the high-precision control efficiency and energy-saving performance of the air conditioning system, this invention uses relevant features from the training set used in IT load prediction to correct the model. The correction results make the output of the dynamic adjustment method of air conditioning more accurate.

[0082] The air conditioning dynamic adjustment method in this embodiment is divided according to the decision interface of the support vector machine. This division can be further subdivided, and the general form is based on... The boundary line with 0 is used to set the air conditioner's airflow speed and other operating parameters, or it can be set according to... The value is used to define multiple thresholds. For example, if the value is less than -3, it can be set to output one dynamic air conditioning adjustment method. If the value is greater than or equal to -3 and less than or equal to -1, another dynamic air conditioning adjustment method can be output. Here, it can be set to the optimal division method obtained by an infinite number of experiments to obtain the corresponding output of multiple dynamic air conditioning adjustment methods.

[0083] A dynamic air conditioning control system based on IT load forecasting is also provided. The system utilizes a dynamic air conditioning control method based on IT load forecasting. The system includes a computing terminal performance acquisition module, a computing power scheduling index feature acquisition module, an equipment comprehensive performance data acquisition module, a business environment correlation feature acquisition module, an IT load forecasting model construction module, and an air conditioning dynamic control model construction module.

[0084] The computing terminal performance acquisition module is used to collect computing terminal performance parameters;

[0085] The computing power scheduling index feature acquisition module is used to obtain computing power scheduling index features when calculating tasks.

[0086] The equipment comprehensive performance data acquisition module is used to acquire equipment comprehensive performance data.

[0087] The business environment association feature acquisition module is used to process and acquire business environment association features.

[0088] The IT load prediction model construction module: obtains the IT load situation, and processes the computing terminal performance parameters, the computing power scheduling index characteristics, the equipment comprehensive performance data and the business environment correlation characteristics to obtain IT load prediction characteristics, and constructs an IT load prediction model using the IT load prediction characteristics and the IT load situation;

[0089] The air conditioning dynamic adjustment and control model construction module: uses the performance parameters of the computing terminal, the computing power scheduling index features, and the business environment correlation features to construct an improved support vector machine model, and uses the improved support vector machine model to obtain a real-time air conditioning dynamic adjustment method.

[0090] This invention utilizes system performance parameters of computing terminals, computing power scheduling index characteristics, comprehensive equipment performance data, and business environment correlation characteristics and corresponding IT load conditions to construct an IT load prediction model. This model predicts real-time IT load conditions. Based on the IT load conditions, air conditioning dynamic adjustment index, and dynamic air conditioning adjustment method, an air conditioning dynamic adjustment control model is constructed. This model processes the real-time IT load conditions and corresponding dynamic air conditioning adjustment index to obtain the final dynamic air conditioning adjustment method. The air conditioning status is adjusted according to this final method. This invention combines dual machine learning models to achieve precise temperature difference control within a range of 0.8℃ in the computer room. The combination of the two models enables real-time control of IT load prediction and dynamic air conditioning adjustment, improving the high-precision control efficiency and energy-saving performance of the air conditioning system. Furthermore, to achieve precise temperature difference control in the computer room, the support vector machine obtained through classification of the dynamic air conditioning adjustment method is improved with enhanced parameters for IT load prediction. This makes the dynamic air conditioning adjustment method in IT load prediction scenarios more robust and generalizable, enabling the construction of a green and efficient system for high-precision instrument temperature control. This reduces the cost and energy consumption of manual inspections, providing significant technological value for carbon neutrality goals.

[0091] For any module structures not specifically defined in this invention, the existing technical specifications shall prevail. The existing technical specifications mentioned in the foregoing background and specific embodiments sections are considered part of this invention and are used to understand the meaning of certain technical features or parameters. The scope of protection of this invention is determined by the actual contents of the claims.

Claims

1. A dynamic air conditioning adjustment method based on IT load forecasting, characterized in that, The method includes: The first computing terminal performance parameters, the first computing power scheduling index characteristics during computing tasks, the first device comprehensive performance data at this time, and the first business environment correlation characteristics are obtained, along with the first IT load situation corresponding to the above characteristics. The first IT load prediction feature is obtained by processing the first computing terminal performance parameters, the first computing power scheduling index features, the first device comprehensive performance data, and the first business environment correlation features. An IT load prediction model is constructed using the first IT load prediction features and the corresponding first IT load situation; the IT load prediction model is constructed using a deep learning model. The second computing terminal performance parameters, the second computing power scheduling index characteristics during the computing task, the comprehensive performance data of the second device at this time, and the second business environment correlation characteristics of the receiving computing terminal are processed to obtain the second IT load prediction characteristics. The IT load prediction model processes the second IT load prediction characteristics to generate the second IT load situation. Based on the training and prediction of the IT load situation, the air conditioning dynamic adjustment index, and the dynamic air conditioning adjustment method, an air conditioning dynamic adjustment control model is constructed. The air conditioning dynamic adjustment index is calculated from the air supply speed, air supply volume, and air supply humidity. In the formula, f is the current frequency of the inverter. This is the maximum inverter frequency of the air conditioner. This is the minimum inverter frequency for the air conditioner. The air supply velocity is f. Let f be the air supply volume at frequency f. The supply air humidity is at frequency f. This indicates the number of times the frequency changes from the maximum to the minimum frequency; The air conditioning dynamic adjustment and control model adopts an improved support vector machine model based on computing terminal performance parameters, computing power scheduling index characteristics, and business environment correlation characteristics. Its decision boundary hyperplane is: In the formula, This represents the product of the feature values ​​of the computing terminal performance parameter, computing power scheduling index feature, and business environment correlation feature corresponding to the nth training feature vector in the IT load prediction feature dataset used to train the model, where m is the number of IT load prediction feature datasets used for training. , Let be the normal vector and intercept of the hyperplane, respectively. The combined feature vector of the training and prediction of IT load and the air conditioning dynamic adjustment index; the air conditioning dynamic adjustment method is divided according to the decision interface of the air conditioning dynamic adjustment control model; The real-time dynamic air conditioning adjustment method obtained by processing the air conditioning dynamic adjustment and control model is used to adjust the air conditioning to regulate the temperature of the computing terminal environment.

2. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 1, characterized in that: The performance parameters of the first or second computing terminal are calculated by considering the comprehensive CPU utilization, memory usage, disk read / write volume, and network traffic characteristics of the computing terminal.

3. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 1, characterized in that: The first computing power scheduling index feature or the second computing power scheduling index feature is calculated by the amount of data retrieved by the computing terminal.

4. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 3, characterized in that: The comprehensive performance data of the first device or the comprehensive performance data of the second device includes the comprehensive power of the computing terminal, the UPS load rate, and the operating voltage.

5. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 1, characterized in that: The first business environment related feature or the second business environment related feature is obtained by processing external customer data access volume, climate temperature difference coefficient and computing terminal environmental humidity.

6. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 1, characterized in that: The first IT load prediction feature is obtained by horizontally splicing the first computing terminal performance parameters, the first computing power scheduling index feature, and the first business environment correlation feature, and vertically splicing them with the first device comprehensive performance data.

7. The dynamic air conditioning adjustment method based on IT load prediction as described in claim 4, characterized in that: The IT load prediction model is built using a deep learning model, and its activation function is: In the formula, For the activation function value, The input is the IT load prediction feature.

8. A dynamic air conditioning control system based on IT load prediction, the system utilizing the method as described in claim 1, the system comprising a computing terminal performance acquisition module, a computing power scheduling index feature acquisition module, an equipment comprehensive performance data acquisition module, a business environment correlation feature acquisition module, an IT load prediction model construction module, and an air conditioning dynamic control model construction module, characterized in that: The computing terminal performance acquisition module is used to collect computing terminal performance parameters; The computing power scheduling index feature acquisition module is used to obtain computing power scheduling index features when calculating tasks. The equipment comprehensive performance data acquisition module is used to acquire equipment comprehensive performance data. The business environment association feature acquisition module is used to process and acquire business environment association features. The IT load prediction model construction module: obtains the IT load situation, and processes the computing terminal performance parameters, the computing power scheduling index characteristics, the equipment comprehensive performance data and the business environment correlation characteristics to obtain IT load prediction characteristics, and constructs an IT load prediction model using the IT load prediction characteristics and the IT load situation; The air conditioning dynamic adjustment and control model construction module: uses the performance parameters of the computing terminal, the computing power scheduling index features, and the business environment correlation features to construct an improved support vector machine model, and uses the improved support vector machine model to obtain a real-time air conditioning dynamic adjustment method; The air conditioning dynamic adjustment and control model adopts an improved support vector machine model based on computing terminal performance parameters, computing power scheduling index characteristics, and business environment correlation characteristics. Its decision boundary hyperplane is: In the formula, This represents the product of the feature values ​​of the computing terminal performance parameter, computing power scheduling index feature, and business environment correlation feature corresponding to the nth training feature vector in the IT load prediction feature dataset used to train the model, where m is the number of IT load prediction feature datasets used for training. , Let be the normal vector and intercept of the hyperplane, respectively. To train and predict the combined feature vector of IT load and corresponding air conditioning dynamic adjustment index; to classify the air conditioning dynamic adjustment method according to the decision interface of the air conditioning dynamic adjustment control model; The real-time dynamic air conditioning adjustment method obtained by processing the air conditioning dynamic adjustment and control model is used to adjust the air conditioning to regulate the temperature of the computing terminal environment.

Images