Intelligent cooling method for data center storage units

Abstract

This invention relates to the field of artificial intelligence technology and provides an intelligent cooling method for data center storage units. The method includes: connecting to the storage unit and acquiring a data processing sample set; performing instruction control process analysis and data access heat analysis based on the access instructions corresponding to each sample data, outputting multiple storage processes and multiple access heats, and constructing a storage instruction tree; upon receiving a first task instruction, traversing the storage instruction tree and outputting a first matching node set; connecting to a cooling control system, integrating and learning to output first cooling control parameters, and feeding them back to the cooling control system. This solves the problem that cooling methods rely on manual intervention and cannot perform adaptive cooling control technology. It enables the transformation of hot and cold data storage in data center storage units, performs heat prediction, and uses the predicted heat temperature to convert the data center's cooling control parameters, improving the response speed and efficiency of cooling control, and achieving the effect of adaptive cooling control technology.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an intelligent cooling method for data center storage units. Background Technology

[0002] Cooling control is commonly used in electronic devices such as air conditioners and refrigerators. Traditional cooling methods include fixed temperature control, manual control, and fixed time control. Fixed temperature control does not consider changes in actual load and heat demand, which may lead to over-cooling or under-cooling, wasting energy or failing to effectively lower the temperature. Manual control involves manually adjusting the air conditioner's temperature or fan speed, but it is prone to human error and delays, and cannot monitor and adjust the cooling system in real time. Fixed time control operates according to a fixed schedule, but cannot be adjusted according to actual needs, which may lead to energy waste and poor cooling effect. In summary, traditional cooling methods have limited control precision, rely on manual intervention, and lack dynamism and adaptability, urgently requiring a more efficient cooling control method.

[0003] In summary, existing technologies suffer from the problem that cooling methods rely on manual intervention and cannot achieve adaptive cooling control. Summary of the Invention

[0004] This application provides an intelligent cooling method for data center storage units, aiming to solve the technical problem that existing cooling methods rely on manual intervention and cannot achieve adaptive cooling control.

[0005] In view of the above problems, this application provides an intelligent cooling method for data center storage units.

[0006] The first aspect disclosed in this application provides an intelligent cooling method for a data center storage unit. The method includes: connecting to a storage unit in a first data center; obtaining a data processing sample set of the storage unit, including access instructions corresponding to each sample data and access frequencies corresponding to each sample data; performing instruction control process analysis based on the access instructions corresponding to each sample data, and outputting multiple storage processes; performing data access heat analysis based on the access frequencies corresponding to each sample data, and outputting multiple access heat values; dividing tree nodes according to the multiple storage processes, and identifying the tree nodes according to the multiple access heat values ​​to build a storage instruction tree; when the storage unit receives a first task instruction, traversing the storage instruction tree according to the first task instruction, and outputting a first matching node set; connecting to a cooling control system of the storage unit, and performing integrated learning based on the first matching node set and a cooling integrated prediction model, outputting first cooling control parameters, and feeding them back to the cooling control system for control.

[0007] Another aspect of this application discloses an intelligent cooling system for a data center storage unit, wherein the system includes: a data processing sample set acquisition module, configured to connect to the storage unit of a first data center and acquire a data processing sample set of the storage unit, including access instructions corresponding to each sample data and access frequencies corresponding to each sample data; a storage process output module, configured to perform instruction control process analysis based on the access instructions corresponding to each sample data and output multiple storage processes; an access heat output module, configured to perform data access heat analysis based on the access frequencies corresponding to each sample data and output multiple access heat values; a storage instruction tree construction module, configured to divide tree nodes according to the multiple storage processes, identify the tree nodes according to the multiple access heat values, and construct a storage instruction tree; a first matching node set output module, configured to traverse the storage instruction tree according to the first task instruction when the storage unit receives a first task instruction and output a first matching node set; and a cooling control module, configured to connect to the cooling control system of the storage unit, and perform integrated learning based on the first matching node set and a cooling integrated prediction model, output a first cooling control parameter, and feed it back to the cooling control system for control.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] By employing a live-connected storage unit, a data processing sample set is acquired. The access instructions corresponding to each sample data are used for instruction control process analysis and data access heat analysis, outputting multiple storage processes, multiple access heat values, and constructing a storage instruction tree. When the storage unit receives the first task instruction, it traverses the storage instruction tree and outputs the first matching node set. The cooling control system of the storage unit is connected, and integrated learning is performed based on the first matching node set and the cooling integrated prediction model to output the first cooling control parameters, which are then fed back to the cooling control system for control. This achieves the technical effect of transforming the storage of hot and cold data in the data center storage unit, performing heat prediction, and using the predicted heat temperature to convert the data center's cooling control parameters, thereby improving the response speed and efficiency of cooling control and achieving adaptive cooling control.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] Figure 1 This application provides a possible flowchart of an intelligent cooling method for a data center storage unit.

[0012] Figure 2 This application provides a schematic diagram of a possible process for outputting the first cooling control parameter in an intelligent cooling method for a data center storage unit, as described in an embodiment of the present application.

[0013] Figure 3 This application provides a schematic diagram illustrating a possible process for outputting dynamic cooling control parameters in an intelligent cooling method for a data center storage unit, as described in this embodiment.

[0014] Figure 4 This application provides a possible structural diagram of an intelligent cooling system for a data center storage unit.

[0015] Explanation of reference numerals in the attached diagram: Data processing sample set acquisition module 100, storage process output module 200, access popularity output module 300, storage instruction tree construction module 400, first matching node set output module 500, cooling control module 600. Detailed Implementation

[0016] This application provides an intelligent cooling method for data center storage units, which solves the technical problem that cooling methods rely on manual intervention and cannot perform adaptive cooling control. It realizes the transformation of hot and cold data storage in data center storage units, performs heat prediction, and uses the predicted heat temperature to convert the cooling control parameters of the data center, thereby improving the response speed and efficiency of cooling control and achieving the technical effect of adaptive cooling control.

[0017] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0018] Example 1

[0019] like Figure 1 As shown in the figure, this application embodiment provides an intelligent cooling method for a data center storage unit, wherein the method includes:

[0020] S10: Connect to the storage unit of the first data center, and obtain the data processing sample set of the storage unit, including the access instructions corresponding to each sample data and the access frequency corresponding to each sample data;

[0021] S20: Perform instruction control process analysis based on the access instructions corresponding to each sample data, and output multiple storage processes;

[0022] S30: Perform data access popularity analysis based on the access frequency corresponding to each sample data, and output multiple access popularity values;

[0023] Step S30 includes the following steps:

[0024] S31: Perform a difference analysis based on the access frequency of each sample data to obtain a set of difference indices;

[0025] S32: Perform clustering according to the coordinate distribution of the difference index set, and output the difference index clustering results;

[0026] S33: Based on the clustering results of the difference index, the popularity of each sample data is identified, and multiple access popularity values ​​are output.

[0027] Specifically, the storage unit of the first data center refers to the unit or device in the data center used to store data, such as servers and hard drives. Connecting to the storage unit of the first data center and obtaining the data processing sample set of the storage unit, the data processing sample set includes access instructions and access frequencies corresponding to each sample data. Each sample data corresponds to one access instruction and one access frequency. The access instruction refers to the instruction to access or operate on the data in the storage unit, which can be read, write, or other operations. The access frequency refers to the frequency or number of access instructions corresponding to each sample data within a certain period of time, used to measure the data's popularity or the frequency of access. The data in the data processing sample set is collected from actual operations, and the storage unit only supports reading and writing, not data rewriting.

[0028] The access instructions corresponding to each sample data are analyzed and processed to determine the control method and sequence of the storage process. The access instructions corresponding to each sample data are analyzed in a process-oriented manner. By analyzing the relationship between the instructions, multiple storage processes are output. These multiple storage processes describe the data processing flow and the relationship between storage units. The storage process refers to the operation flow of the storage unit obtained from the process-oriented analysis of the instruction control flow, which is used to guide the cooling operation of the storage unit.

[0029] The access frequency corresponding to each sample data is analyzed to determine the popularity or frequency of access to the data. Data access popularity analysis is performed based on the access frequency corresponding to each sample data, and multiple access popularity are output. The difference index set is obtained by comparing the access frequency of each sample data, reflecting the degree of difference between the access frequencies of different data.

[0030] Based on the coordinate distribution of the difference index set, clustering algorithms (such as K-means, hierarchical clustering, etc.) can be used to cluster the data. Clustering calculates similarity to divide the data into different clusters. Bottom-up agglomerative hierarchical clustering analysis is performed on the material structure basic information and the connection and fixing process information set to form the difference index clustering results.

[0031] If some clusters in the clustering results can be considered as hot data, it means that these data are frequently accessed, while other clusters can be considered as cold data, it means that these data are accessed less often. Based on these labels, the hotness and coldness of each sample data are identified according to the clustering results of the difference index, and multiple access hotness values ​​are output for further data access adjustment and cooling control operations.

[0032] In summary, by performing differential analysis and clustering on the access frequency of sample data, the popularity of different data can be identified, thereby outputting multiple access popularity levels to support the optimization and energy-saving effect of intelligent cooling systems.

[0033] S40: Divide the tree nodes according to the multiple storage processes, identify the tree nodes according to the multiple access hotness, and build a storage instruction tree;

[0034] S50: When the storage unit receives the first task instruction, it traverses the storage instruction tree according to the first task instruction and outputs the first matching node set;

[0035] S60: Connect the cooling control system of the storage unit, and perform integrated learning based on the first matching node set and the cooling integrated prediction model, output the first cooling control parameter, and feed it back to the cooling control system for control.

[0036] Specifically, based on the FP-tree (Frequent Pattern Tree) model, the multiple storage processes are divided into multiple tree nodes according to the divided tree nodes and the access frequency of the identifiers. The multiple tree nodes are then identified according to the access frequency of the multiple nodes to build a storage instruction tree. The storage instruction tree can be used to represent the dependencies between data processing processes and nodes, transform the storage of hot and cold data in data center storage units, and help to effectively manage and organize storage instructions.

[0037] The first task instruction is an instruction used to describe the cooling operation of the storage unit, which may include specifying the cooling time, cooling method or other related parameters. When the storage unit receives the first task instruction, it traverses the storage instruction tree according to the first task instruction. The traversal process involves selecting the corresponding nodes according to the requirements and dependencies of the first task instruction, which can determine which nodes need to perform the cooling operation, thereby outputting a first matching node set. The first matching node set is used to represent the set of nodes that match the first task instruction found during the traversal of the storage instruction tree. The elements in the first matching node set may be nodes that need to be cooled.

[0038] The data center storage unit is connected to the cooling control system of the storage unit. Upon receiving the invoked instruction, the data center storage unit performs ensemble learning based on the first matching node set and the cooling integrated prediction model, outputting first cooling control parameters. These first cooling control parameters are the cooling control parameters for the data center based on the heat temperature predicted by the cooling integrated prediction model, including relevant control parameters such as cooling liquid flow control parameters. Feeding these first cooling control parameters back to the cooling control system for actual cooling control operations ensures that the cooling strategy is adjusted according to the results of the matching nodes and the ensemble learning model, providing more efficient and energy-saving cooling. By combining data characteristics and cooling requirements, and utilizing advanced technology and an automated control system, dynamic cooling adjustments are achieved to improve energy efficiency and reduce energy waste.

[0039] like Figure 2 As shown, step S60 includes the following steps:

[0040] S61: Obtain the first matching node set, wherein the first matching node set includes operation flow nodes matched after traversing the storage instruction tree based on the first task instruction;

[0041] S62: Input the first set of matching nodes into the cooling integrated prediction model, which includes a heat prediction sub-model and a heat conversion sub-model;

[0042] S63: Obtain the node popularity corresponding to each matching node in the first matching node set, and output the matching node popularity set;

[0043] S64: Based on the heat prediction sub-model, perform fusion prediction on the heat set of the matching nodes, and output the predicted heat information that identifies the first task instruction;

[0044] S65: The predicted heat information is used as input data for the heat conversion sub-model to learn and output the first cooling control parameter.

[0045] Specifically, the system performs ensemble learning based on the first matching node set and the cooling integrated prediction model to output the first cooling control parameters. This includes traversing the stored instruction tree according to the first task instruction and finding the operation flow node that matches the instruction; the resulting matching nodes form the first matching node set; inputting the first matching node set into the cooling integrated prediction model for temperature prediction, the cooling integrated prediction model including two sub-models: a heat prediction sub-model and a heat conversion sub-model; obtaining the corresponding node heat for each matching node in the first matching node set, the resulting corresponding node heat forms the matching node heat set, the node heat refers to the access heat associated with each matching node in the first matching node set, the output of the node heat set can help the system perform heat-based mapping or decision-making in subsequent processing to better meet the cooling demand target;

[0046] Based on the heat prediction sub-model, the heat set of the matching nodes is fused and predicted. The fused prediction output identifies the predicted heat information of the first task instruction. This predicted heat information is used as input data for the heat conversion sub-model for learning, outputting a first cooling control parameter. This first cooling control parameter is the cooling control parameter for the data center based on the heat temperature predicted by the integrated cooling prediction model. By using the integrated cooling prediction model, appropriate cooling control parameters can be predicted according to the heat status of the matching nodes, enabling intelligent cooling operations. This allows for targeted cooling control based on the heat status of different nodes, improving system efficiency and energy saving.

[0047] Step S64 includes the following steps:

[0048] S641: Determine whether the node heat stability of the matching node heat set is greater than or equal to the preset heat stability. If the node heat stability of the matching node heat set is greater than or equal to the preset heat stability, preset the first fusion weight layer.

[0049] S642-A: Wherein, the weight distribution difference of the first fusion weight layer is less than θ1, and 0≤θ1≤0.5;

[0050] S643-A: Perform fusion prediction on the matching node heat set according to the first fusion weight layer.

[0051] This application also includes the following embodiments:

[0052] S642-B: If the node heat stability of the matching node heat set is less than the preset heat stability, a second fusion weight layer is preset, wherein the weight distribution difference of the second fusion weight layer is greater than θ1.

[0053] S643-B performs fusion prediction on the matching node heat set according to the second fusion weight layer.

[0054] Specifically, according to the heat prediction sub-model, the matching node heat set is fused and predicted, including firstly, calculating the node heat stability of the matching node heat set to determine whether the validity of meeting the preset heat stability threshold is met. The node heat stability can be measured by calculating the variance or standard deviation of the node heat.

[0055] Determine whether the node heat stability of the matched node heat set is greater than or equal to a preset heat stability: In the first case, if the node heat stability of the matched node heat set is greater than or equal to the preset heat stability, a first fusion weight layer is preset. The first fusion weight layer refers to the weight distribution of the fused heat value obtained by weighted averaging of the heat values ​​of different nodes. The form of the weight distribution can be selected according to the specific situation, such as using a uniform distribution or a Gaussian distribution. The weight distribution difference of the first fusion weight layer is less than θ1, and 0≤θ1≤0.5. θ1 is used to characterize the preset weight distribution difference threshold, which can be customized by those skilled in the art. Perform fusion prediction on the matched node heat set according to the first fusion weight layer: The first fusion weight layer can be used to perform fusion prediction on the matched node heat set. Fusion prediction can be achieved by multiplying the heat values ​​of different nodes by their corresponding weights and then summing them. The fused heat value can be used as the output result for further control parameter calculation or other purposes.

[0056] Determine whether the node heat stability of the matched node heat set is greater than or equal to a preset heat stability: In the second case, if the node heat stability of the matched node heat set is less than the preset heat stability, a second fusion weight layer is preset. Some difference measurement methods can be used to calculate the weight distribution difference of the second fusion weight layer, such as KL (Kullback-Leibler, relative entropy) divergence, Euclidean distance, etc. The weight distribution difference of the second fusion weight layer is greater than θ1. The matched node heat set is fused and predicted according to the second fusion weight layer. By judging the stability of node heat and selecting an appropriate weight distribution, the degree to which the system meets the cooling demand target can be improved.

[0057] This application also includes the following embodiments:

[0058] S651: Connect to the cooling control system of the storage unit, and acquire cooling liquid information, cooling flow rate and heat loss data from the cooling control system;

[0059] S652: Train the model using the cooling liquid information as a fixed variable and the cooling flow rate and the heat loss data as variables, and output a heat conversion sub-model that identifies the rate-heat conversion relationship.

[0060] S653: Input the predicted heat information into the heat conversion sub-model and output the first cooling control parameter.

[0061] Specifically, data collection is performed by connecting to the cooling control system of the storage unit to obtain cooling liquid information, cooling flow rate, and heat loss data from the cooling control system. The cooling liquid information is used as a fixed variable, and the cooling flow rate and heat loss data are used as variables to train a heat conversion sub-model that identifies the rate-heat conversion relationship to identify the conversion relationship between rate and heat. Preferably, the heat conversion sub-model that identifies the rate-heat conversion relationship can be based on a regression model, a neural network, or other suitable algorithms.

[0062] The heat information to be predicted is input into the trained heat conversion sub-model. The heat conversion sub-model outputs the corresponding first cooling control parameter based on the input heat information and the learned conversion relationship. The first cooling control parameter can be the flow rate of the cooling liquid or other parameters related to the cooling control system, thereby optimizing the cooling control system.

[0063] like Figure 3 As shown, embodiments of this application also include:

[0064] S71: Determine whether the processing time of the first task instruction is greater than the preset processing time;

[0065] S72: If the processing time of the first task instruction is greater than the preset processing time, obtain the node processing timing information corresponding to the first matching node set under the first task instruction;

[0066] S73: Take the first cooling control parameter as the cooling demand target, map it to the matching node heat set corresponding to the first matching node set according to the node processing time sequence information corresponding to the first matching node set, and output the cooling control parameter based on the time sequence mapping.

[0067] S74: Smooth the cooling control parameters and output dynamic cooling control parameters based on time-series mapping.

[0068] Specifically, the control achieved by feeding back the first cooling control parameter to the cooling control system is instantaneous cooling control. If long-term cooling control is required, dynamic control is needed based on changes in process nodes. Based on this, it is determined whether the processing time of the first task instruction is greater than the preset processing time. If the processing time of the first task instruction is greater than the preset processing time, it means that the cooling control system cannot complete the cooling task within the specified time and needs to make corresponding adjustments. The node processing timing information corresponding to the first matching node set under the first task instruction is obtained. The node processing timing information can provide information about the time and order of node processing.

[0069] The first cooling control parameter is used as the cooling demand target. The first cooling control parameter is dynamically adjusted according to the node processing timing information corresponding to the first matching node set to adapt to the time change of node processing. Based on this, the cooling control system can perform dynamic control according to the node processing timing information to achieve long-term cooling control.

[0070] The system maps the node processing timing information corresponding to the first set of matched nodes to the set of matched node heats, outputting time-mapping-based cooling control parameters. These time-mapping-based cooling control parameters are then smoothed to obtain dynamic cooling control parameters. Through mapping and smoothing, the cooling control system becomes more efficient and adaptable to different operating conditions. Dynamically adjusting the cooling control parameters based on node processing timing information solves the problem of long-term cooling control. Furthermore, the dynamic cooling control parameters based on time-mapping improve the system's responsiveness.

[0071] In summary, the intelligent cooling method for a data center storage unit provided in this application has the following technical effects:

[0072] 1. By employing a live-connected storage unit, a data processing sample set is obtained; the access instructions corresponding to each sample data are used for instruction control process analysis and data access heat analysis, outputting multiple storage processes, multiple access heats, and constructing a storage instruction tree; when the storage unit receives the first task instruction, it traverses the storage instruction tree and outputs the first matching node set; it connects to the cooling control system, and performs integrated learning based on the first matching node set and the cooling integrated prediction model, outputting the first cooling control parameters, and feeding them back to the cooling control system for control. This application provides an intelligent cooling method for data center storage units, realizing the transformation of hot and cold data storage in data center storage units, performing heat prediction, and using the predicted heat temperature to convert the cooling control parameters of the data center, thereby improving the response speed and efficiency of cooling control and achieving the technical effect of adaptive cooling control.

[0073] 2. The system employs a method to determine if the processing time exceeds a preset processing time. If the processing time exceeds the preset processing time, it obtains the node processing timing information corresponding to the first matching node set under the first task instruction. The first cooling control parameter is used as the cooling demand target. Based on the node processing timing information, it is mapped to the matching node heat set, outputting cooling control parameters and performing smoothing processing. This results in dynamic cooling control parameters based on timing mapping. Through mapping and smoothing, the cooling control system becomes more efficient and adaptable to different working conditions. Dynamically adjusting the cooling control parameters based on the node processing timing information solves the problem of long-term cooling control. Simultaneously, the dynamic cooling control parameters based on timing mapping improve the system's responsiveness.

[0074] Example 2

[0075] Based on the same inventive concept as the intelligent cooling method for a data center storage unit in the foregoing embodiments, such as Figure 4 As shown in the figure, this application embodiment provides an intelligent cooling system for a data center storage unit, wherein the system includes:

[0076] The data processing sample set acquisition module 100 is used to connect to the storage unit of the first data center and acquire the data processing sample set of the storage unit, including the access instructions corresponding to each sample data and the access frequency corresponding to each sample data.

[0077] The storage process output module 200 is used to perform instruction control process analysis based on the access instructions corresponding to each sample data, and output multiple storage processes.

[0078] The access popularity output module 300 is used to perform data access popularity analysis based on the access frequency corresponding to each sample data, and output multiple access popularity values.

[0079] The storage instruction tree building module 400 is used to divide tree nodes according to the multiple storage processes, identify tree nodes according to the multiple access hotness, and build a storage instruction tree.

[0080] The first matching node set output module 500 is used to traverse the storage instruction tree according to the first task instruction and output the first matching node set when the storage unit receives the first task instruction.

[0081] The cooling control module 600 is used to connect to the cooling control system of the storage unit, and to perform integrated learning based on the first matching node set and the cooling integrated prediction model, output the first cooling control parameters, and feed them back to the cooling control system for control.

[0082] Furthermore, the system includes:

[0083] The first matching node set acquisition module is used to acquire the first matching node set, wherein the first matching node set includes operation flow nodes matched after traversing the storage instruction tree based on the first task instruction;

[0084] The first matching node set input module is used to input the first matching node set into the cooling integrated prediction model, which includes a heat prediction sub-model and a heat conversion sub-model.

[0085] The matching node heat set output module is used to obtain the node heat corresponding to each matching node in the first matching node set and output the matching node heat set;

[0086] The predicted heat information output module is used to perform fusion prediction on the heat prediction sub-model of the matching node heat set and output the predicted heat information that identifies the first task instruction.

[0087] The first cooling control parameter output module is used to learn the predicted heat information as input data for the heat conversion sub-model and output the first cooling control parameters.

[0088] Furthermore, the system includes:

[0089] A cooling-related index acquisition module is used to connect to the cooling control system of the storage unit and acquire cooling liquid information, cooling flow rate and heat loss data in the cooling control system.

[0090] The heat conversion sub-model output module is used to train the model with the cooling liquid information as a fixed variable and the cooling flow rate and the heat loss data as variables, and output a heat conversion sub-model that identifies the rate-heat conversion relationship.

[0091] The first cooling control parameter output module is used to input the predicted heat information into the heat conversion sub-model and output the first cooling control parameter.

[0092] Furthermore, the system includes:

[0093] The difference analysis module is used to perform difference analysis based on the access frequency of each sample data to obtain a set of difference indices.

[0094] The difference index clustering result output module is used to perform clustering according to the coordinate distribution of the difference index set and output the difference index clustering result;

[0095] The access popularity output module is used to identify the popularity of each sample data according to the clustering results of the difference index and output multiple access popularity values.

[0096] Furthermore, the system includes:

[0097] The first judgment module is used to determine whether the node heat stability of the matching node heat set is greater than or equal to the preset heat stability. If the node heat stability of the matching node heat set is greater than or equal to the preset heat stability, a first fusion weight layer is preset.

[0098] The first case analysis module is used in which the weight distribution difference of the first fusion weight layer is less than θ1, and 0≤θ1≤0.5;

[0099] The first fusion prediction module is used to perform fusion prediction on the matching node heat set according to the first fusion weight layer.

[0100] Furthermore, the system includes:

[0101] The second case analysis module is used to preset a second fusion weight layer if the node heat stability of the matching node heat set is less than the preset heat stability. The weight distribution difference of the second fusion weight layer is greater than θ1.

[0102] The second fusion prediction module is used to perform fusion prediction on the matching node heat set according to the second fusion weight layer.

[0103] Furthermore, the system includes:

[0104] The second judgment module is used to determine whether the processing time of the first task instruction is greater than the preset processing time.

[0105] The node processing timing information acquisition module is used to acquire the node processing timing information corresponding to the first matching node set under the first task instruction if the processing time of the first task instruction is greater than the preset processing time.

[0106] The cooling control parameter output module is used to take the first cooling control parameter as the cooling demand target, map it to the matching node heat set corresponding to the first matching node set according to the node processing time sequence information corresponding to the first matching node set, and output the cooling control parameter based on the time sequence mapping.

[0107] The smoothing module is used to smooth the cooling control parameters and output dynamic cooling control parameters based on time-series mapping.

[0108] In summary, any step of the method described above can be stored as a computer instruction or program in an unrestricted computer memory, and can be called and identified by an unrestricted computer processor to implement any method in the embodiments of this application, without any additional restrictions.

[0109] Furthermore, the "first" or "second" mentioned above may not only represent a sequential relationship, but may also represent a specific concept, and / or refer to the individual or collective selection of multiple elements. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A smart cooling method for a data center storage unit, characterized in that, The method includes: The storage unit connected to the first data center obtains the data processing sample set of the storage unit, including the access instructions corresponding to each sample data and the access frequency corresponding to each sample data. The access instructions corresponding to each sample data are analyzed in a flow-oriented manner to output multiple storage processes. Data access popularity analysis is performed based on the access frequency corresponding to each sample data, and multiple access popularity values ​​are output. Tree nodes are divided according to the multiple storage processes, and tree nodes are identified according to the multiple access frequencies to build a storage instruction tree; When the storage unit receives the first task instruction, it traverses the storage instruction tree according to the first task instruction and outputs the first matching node set; The cooling control system connected to the storage unit performs integrated learning based on the first matching node set and the cooling integrated prediction model, outputs the first cooling control parameters, and feeds them back to the cooling control system for control. The cooling integrated prediction model includes a heat prediction sub-model and a heat conversion sub-model.

2. The method as described in claim 1, characterized in that, The method involves performing ensemble learning based on the first set of matching nodes and the cooling ensemble prediction model to output the first cooling control parameter, including: Obtain a first set of matching nodes, wherein the first set of matching nodes includes operation flow nodes matched after traversing the storage instruction tree based on the first task instruction; Input the first set of matching nodes into the integrated cooling prediction model; Obtain the node popularity corresponding to each matching node in the first matching node set, and output the matching node popularity set; Based on the heat prediction sub-model, the heat set of the matching nodes is fused and predicted, and the predicted heat information that identifies the first task instruction is output. The predicted heat information is used as input data for the heat conversion sub-model to learn and output the first cooling control parameter.

3. The method as described in claim 2, characterized in that, The method further includes: The cooling control system connected to the storage unit acquires information on the cooling liquid, cooling flow rate, and heat loss from the cooling control system. The system is trained using the cooling liquid information as a fixed variable and the cooling flow rate and the heat loss data as variables, and outputs a heat conversion sub-model that identifies the rate-heat conversion relationship. The predicted heat information is input into the heat conversion sub-model, and the first cooling control parameter is output.

4. The method as described in claim 1, characterized in that, Data access popularity analysis is performed based on the access frequency corresponding to each sample data, and the method includes: A difference analysis was performed based on the access frequency of each sample data to obtain a set of difference indices; Clustering is performed according to the coordinate distribution of the difference index set, and the difference index clustering results are output. The popularity of each sample data is identified based on the clustering results of the difference index, and multiple access popularity values ​​are output.

5. The method as described in claim 2, characterized in that, Based on the heat prediction sub-model, the heat set of the matching nodes is fused and predicted, including: Determine whether the node heat stability of the matching node heat set is greater than or equal to the preset heat stability. If the node heat stability of the matching node heat set is greater than or equal to the preset heat stability, a first fusion weight layer is preset. Wherein, the weight distribution difference of the first fusion weight layer is less than ,and ; The matching node heat set is fused and predicted according to the first fusion weight layer.

6. The method as described in claim 5, characterized in that, The method further includes: If the node heat stability of the matched node heat set is less than the preset heat stability, a second fusion weight layer is preset, wherein the weight distribution difference of the second fusion weight layer is greater than... ; The matching node heat set is fused and predicted according to the second fusion weight layer.

7. The method as described in claim 1, characterized in that, The method further includes: Determine whether the processing time of the first task instruction is greater than the preset processing time; If the processing time of the first task instruction is greater than the preset processing time, obtain the node processing timing information corresponding to the first matching node set under the first task instruction; The first cooling control parameter is used as the cooling demand target. According to the node processing time sequence information corresponding to the first matching node set, it is mapped with the matching node heat set corresponding to the first matching node set, and the cooling control parameter based on the time sequence mapping is output. The cooling control parameters are smoothed to output dynamic cooling control parameters based on time-series mapping.

8. An intelligent cooling system for a data center storage unit, characterized in that, A smart cooling method for implementing a data center storage unit according to any one of claims 1-7 includes: The data processing sample set acquisition module is used to connect to the storage unit of the first data center and acquire the data processing sample set of the storage unit, including the access instructions corresponding to each sample data and the access frequency corresponding to each sample data. The storage process output module is used to perform instruction control process analysis based on the access instructions corresponding to each sample data, and output multiple storage processes. The access popularity output module is used to perform data access popularity analysis based on the access frequency corresponding to each sample data, and output multiple access popularity values. The storage instruction tree building module is used to divide tree nodes according to the multiple storage processes, identify tree nodes according to the multiple access frequencies, and build a storage instruction tree. The first matching node set output module is used to traverse the storage instruction tree according to the first task instruction when the storage unit receives the first task instruction and output the first matching node set. A cooling control module is used to connect to the cooling control system of the storage unit, and to perform integrated learning based on the first matching node set and the cooling integrated prediction model, outputting the first cooling control parameters and feeding them back to the cooling control system for control. The cooling integrated prediction model includes a heat prediction sub-model and a heat conversion sub-model.

Images