Liquid cooling system energy consumption data management platform based on artificial intelligence

Abstract

The invention relates to the technical field of liquid cooling system management, in particular to a liquid cooling system energy consumption data management platform based on artificial intelligence, and the platform comprises a data obtaining module which is used for obtaining the actual energy consumption data of each server in a server cluster under each historical liquid cooling cycle; the first division module is used for acquiring an initial clustering cluster; the second division module is used for redistributing the energy consumption data points in the initial clustering cluster to obtain each target clustering cluster; and the liquid cooling system management module is used for carrying out redistribution management on the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target cluster. In addition, energy consumption waste during cooling of the centralized liquid cooling system can be reduced, the energy consumption utilization rate and the energy utilization rate of the centralized liquid cooling system are improved, and balanced management of energy consumption and resources of the centralized liquid cooling system is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid cooling system management, and in particular to an artificial intelligence-based liquid cooling system energy consumption data management platform. Background Art

[0002] For servers in data centers, a large amount of heat is generated during operation. If the heat cannot be dissipated in time, the increased temperature of the equipment will not only cause the chip to down-clock, reduce the computing speed, trigger a protective shutdown, but also accelerate the aging of electronic components and increase the risk of hardware failure. Therefore, in order to ensure the stable operation of the data center, a liquid cooling system is usually used to dissipate the heat of the servers in the data center. In order to consider the technical difficulty and overall cost, the liquid cooling system used in the data center is usually a centralized liquid cooling system. The centralized liquid cooling system refers to the servers in the same cabinet or multiple cabinets in the data center. They are usually controlled by a CDU in the centralized liquid cooling system, and the servers controlled by the same CDU will be connected in parallel in the same water cycle, and then large refrigerators and pumps will be used to uniformly regulate the water temperature and flow of the entire water cycle.

[0003] However, in the prior art, when determining the server controlled or served by the CDU in the liquid cooling system, the temperature difference between the servers is not taken into account, that is, the difference in energy consumption required for heat dissipation or cooling of the server is not taken into account, and the temperature of the circulating liquid is determined by the server with the highest temperature among all parallel servers. Therefore, in the actual operation process, energy waste and resource waste will occur, resulting in low energy utilization and resource utilization of the liquid cooling system. For example, if only one server among the servers controlled by a CDU needs to be cooled, the other servers do not need it. However, at this time, in order to cool the servers that need to be cooled, all water circulation branches controlled by the CDU will be cooled. However, this process will cause the branches corresponding to those servers that do not need to be cooled to waste energy and resources during the cooling process. Therefore, how to manage the servers controlled by the CDU of the liquid cooling system to improve the energy utilization and resource utilization of the liquid cooling system has become an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above problems, the present invention provides an energy consumption data management platform for a liquid cooling system based on artificial intelligence. The technical solutions adopted are as follows: An embodiment of the present invention provides an artificial intelligence-based liquid cooling system energy consumption data management platform, the power grid business data management service terminal, including: A data acquisition module, used to acquire actual energy consumption data of each server in a server group under each historical liquid cooling cycle, where the liquid cooling cycle is completed by a liquid cooling system, and the server group is a server group served by the liquid cooling system; The first partitioning module is used to record the sequence composed of the actual energy consumption data of the server under all historical liquid cooling cycles as the actual energy consumption data sequence of the corresponding server, and obtain the energy consumption data points corresponding to the server according to the mean value of the actual energy consumption data sequence and the number of data in the actual energy consumption data sequence that is greater than a preset energy consumption threshold, and cluster the energy consumption data points according to the distance between the energy consumption data points to obtain an initial clustering cluster; A second partitioning module is used to redistribute the energy consumption data points in the initial clustering cluster according to the total number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system to obtain various target clustering clusters; The liquid cooling system management module is used to reallocate and manage the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target cluster.

[0005] Beneficial effects: The present invention includes a data acquisition module for acquiring the actual energy consumption data of each server in the server group under each historical liquid cooling cycle; a first partitioning module for obtaining the energy consumption data points corresponding to the server according to the mean of the actual energy consumption data sequence and the number of data in the actual energy consumption data sequence that is greater than the preset energy consumption threshold, and clustering the energy consumption data points according to the distance between the energy consumption data points to obtain the initial clustering cluster; a second partitioning module for redistributing the energy consumption data points in the initial clustering cluster according to the total number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system to obtain each target clustering cluster; a liquid cooling system management module for redistributing and managing the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target clustering cluster. The present invention can reduce the energy consumption waste and resource waste when the centralized liquid cooling system is cooled, improve the energy consumption utilization rate and energy utilization rate of the centralized liquid cooling system, and realize the balanced management of energy consumption and resources of the centralized liquid cooling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0007] Figure 1This is a structural block diagram of an artificial intelligence-based liquid cooling system energy consumption data management platform of the present invention; Figure 2 Schematic diagram of the liquid cooling system of the present invention. DETAILED DESCRIPTION

[0008] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the embodiments of the present invention.

[0009] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0010] This embodiment provides an artificial intelligence-based liquid cooling system energy consumption data management platform, which is described in detail as follows: like Figure 1 As shown, this embodiment provides an artificial intelligence-based liquid cooling system energy consumption data management platform, including: The data acquisition module 01 is used to obtain the actual energy consumption data of each server in the server group under each historical liquid cooling cycle.

[0011] The main purpose of this embodiment is to reallocate the servers controlled by the CDU (cold capacity distribution unit) of the liquid cooling system according to the energy consumption required by the servers during multiple historical cycles, so as to improve the energy utilization rate of the centralized liquid cooling system and realize the balanced management of the energy consumption of the centralized liquid cooling system. In order to facilitate understanding, this embodiment will be described in detail with the reallocation process of the servers controlled by the CDU of the centralized liquid cooling system used in any data center, that is, all the servers that appear in this embodiment belong to the same data center, and the CDUs that appear in the following are all CDUs of the same centralized liquid cooling system. The CDU of the centralized liquid cooling system is a key component of the centralized liquid cooling system, and the CDU of the centralized liquid cooling system serves multiple servers. The schematic diagram of the centralized liquid cooling system is shown in FIG. Figure 2 As shown; in addition, the data center in this embodiment is an equalized cooling data center or a symmetrical heat dissipation architecture data center, and the number of servers covered by the CDU of the liquid cooling system for cooling the equalized cooling data center or the symmetrical heat dissipation architecture data center is the same, such as the application scenarios of ultra-large-scale cloud data centers and data centers for AI training clusters, which are equalized cooling data centers or symmetrical heat dissipation architecture data centers.

[0012] In this embodiment, the cluster composed of all servers controlled by the centralized liquid cooling system in the data center is first recorded as a server group, and then the actual energy consumption data of each server in the server group under each historical liquid cooling cycle is obtained. The actual energy consumption data is also the actual refrigeration machine liquid cooling energy consumption required for cooling the circulation branch of each server during each historical liquid cooling cycle. The circulation branch of the server refers to a specific pipeline branch in the liquid cooling system for the flow of coolant and the completion of heat transfer. Its core function is to efficiently export the heat generated by the server through liquid circulation, and the liquid cooling cycles performed before the current moment are all called historical liquid cooling cycles, and a liquid cooling cycle of the server refers to a liquid cooling cycle of the CDU that controls the corresponding server. The loop process or the server's primary liquid cooling cycle refers to the closed circulation path formed by the coolant between the external cooling equipment (such as cooling towers, chillers) and the CDU, and the CDU's primary liquid cooling cycle also refers to the closed circulation path of the coolant between the external cooling equipment and the CDU. The core function is to transfer the heat generated inside the data center to the external environment through liquid-liquid heat exchange to complete the final heat dissipation. From this, it can be seen that a liquid cooling cycle of a server controlled by a CDU refers to the cooling liquid completing the entire process of "flow-heat absorption-heat dissipation-reflux" in the primary side circuit of the corresponding CDU, that is, a liquid cooling cycle of a server controlled by a CDU refers to the process in which the coolant starts from the corresponding CDU, flows through the external cooling equipment to complete the heat dissipation, and then returns to the CDU. In addition, it should be noted that in order to avoid energy waste and to ensure the cooling effect, when designing a centralized liquid cooling system for a data center, factors such as server power consumption density, cooling requirements, and system architecture design are generally combined to comprehensively determine the number of servers covered by each CDU of the centralized liquid cooling system. That is, in actual applications, it is necessary to set the number of servers covered by the CDU according to the configuration scenario of the actual business scenario. If the application scenario of the data center in this embodiment is an AI training cluster, which mainly performs large model training or scientific computing, then the number of servers covered by the CDU in this embodiment is generally set to 8, but when designing the CDU interface, redundancy reservation is generally required, and the redundant reserved interface is generally reserved as a backup loop. If the number of servers covered by a CDU is 8, then the CDU is generally configured with 10 interfaces, and the remaining 2 interfaces are reserved for expansion or redundancy.

[0013] For ease of understanding, this embodiment will be described below by taking the specific acquisition process of the actual energy consumption data of the bth server in the server group under the ath historical liquid cooling cycle as an example, wherein the ath historical liquid cooling cycle also refers to the ath historical liquid cooling cycle of the CDU that controls or serves the bth server during the centralized liquid cooling system operation historical time period, and the centralized liquid cooling system operation historical time period refers to the operation time period before the current moment, then the specific acquisition process of the actual energy consumption data of the bth server under the ath historical liquid cooling cycle is: First, the cold night flow rate, cold night density, and cold night specific heat capacity in the circulation branch of the b-th server during the a-th historical liquid cooling cycle are obtained. The cold night refers to the cooling medium or the liquid used for cooling in the liquid cooling system. The cold night in this embodiment is water, and the specific heat capacity of water is 4.2×10³ joules / kg·degrees Celsius. The density of water is 1×10³ kg / m3. The cold night flow rate in the circulation branch refers to the water flow rate in the circulation branch. The water flow rate in the circulation branch is collected by a flow sensor installed on the circulation branch pipeline; then, the CDU serving the b-th server during the a-th historical liquid cooling cycle is obtained. The temperature value of the bth server before the cycle starts is recorded as Z1. The temperature value of the server can be collected by the temperature sensor configured inside the corresponding server or the cold plate temperature of the server can be used as the temperature value of the server. The cold plate temperature of the server is usually collected by a temperature sensor installed on the cold plate of the server. Then the CPU safety temperature value of the bth server is obtained and recorded as Z0. The CPU safety temperature value of the server refers to the maximum allowable temperature value. Once the CPU safety temperature value of the server is exceeded or reached, it is necessary to cool down and determine whether Z1 is greater than Z0. If so, Then the result of Z1-Z0 is used as the temperature that needs to be lowered for the b-th server during the a-th historical liquid cooling cycle. Otherwise, 0 is used as the temperature that needs to be lowered for the b-th server during the a-th historical liquid cooling cycle. Then the overall cycle time of the CDU serving the b-th server during the a-th historical liquid cooling cycle is obtained, and recorded as the overall cycle time of the b-th server during the a-th historical liquid cooling cycle. Then the energy efficiency ratio of the refrigerator of the centralized liquid cooling system is obtained, and the inverse of the energy efficiency ratio of the refrigerator of the centralized liquid cooling system is recorded as the characteristic ratio. The energy efficiency ratio of the refrigerator is generally the heat removed by the refrigerator. The energy efficiency ratio of the refrigerator is determined by dividing it by the energy consumed by the refrigerator, and in this embodiment, the refrigerator energy efficiency ratio is set to 3.6 according to the prior art; then, when the b-th server is in the a-th historical liquid cooling cycle, the cold night flow rate in the circulation branch of the b-th server, the cold night density, the cold night specific heat capacity, the temperature that the b-th server needs to reduce in the a-th historical liquid cooling cycle, the characteristic ratio, and the overall cycle time of the b-th server in the a-th historical liquid cooling cycle are multiplied as the actual energy consumption data of the b-th server in the a-th historical liquid cooling cycle, that is, the actual energy consumption data of the b-th server in the a-th historical liquid cooling cycle is ,in, is the cold night specific heat capacity in the circulation branch of the bth server, is the temperature that needs to be lowered for the bth server during the ath historical liquid cooling cycle, f is the cold night flow rate in the circulation branch of the bth server, t is the overall cycle time of the bth server during the ath historical liquid cooling cycle, is the cold night density in the loop branch of the b-th server, is the energy efficiency ratio of the refrigerator in the liquid cooling system; and the calculation process of the actual energy consumption data is determined based on the energy calculation process.

[0014] In addition, in this embodiment, the purpose of obtaining the actual energy consumption data is to allocate servers with similar actual energy consumption characteristics to the same CDU during subsequent server reallocation, so as to improve energy utilization.

[0015] Therefore, this embodiment can obtain the actual energy consumption data of each server in the server group under each historical liquid cooling cycle through the above process.

[0016] The first partitioning module 02 is used to record the sequence composed of the actual energy consumption data of the server under all historical liquid cooling cycles as the actual energy consumption data sequence of the corresponding server, and obtain the energy consumption data points corresponding to the server according to the mean of the actual energy consumption data sequence and the number of data in the actual energy consumption data sequence that is greater than a preset energy consumption threshold, and cluster the energy consumption data points according to the distance between the energy consumption data points to obtain an initial clustering cluster.

[0017] Since the actual energy consumption data under a single cycle is difficult to characterize the energy consumption characteristics of the server, this embodiment will then characterize the energy consumption characteristics of the server based on the actual energy consumption data of the server under multiple historical cycles, and perform clustering to obtain an initial clustering cluster, that is, this embodiment first sorts the actual energy consumption data of each server under all historical liquid cooling cycles in the order of liquid cooling cycle time, and records the sorted sequence as the actual energy consumption data sequence of the corresponding server; and as another real-time method, this embodiment can also record the sequence composed of the actual energy consumption data of each server under the most recent N historical liquid cooling cycles as the actual energy consumption data sequence of the corresponding server, and the value of N generally needs to be set by the implementer according to the actual number of historical cycles and experimental statistics. Generally, the value of N is set to 50, that is, the actual energy consumption demand under 50 historical liquid cooling cycles can better describe the changes in the energy consumption demand of the server in the near future.

[0018] After obtaining the actual energy consumption data sequence of the server, this embodiment then obtains the energy consumption data point corresponding to each server according to the mean of the actual energy consumption data sequence of the server and the number of data in the actual energy consumption data sequence of the server that is greater than the preset energy consumption threshold, and the specific process of obtaining the energy consumption data point corresponding to the server is: First, the mean of the actual energy consumption data sequence of each server is obtained and recorded as the historical actual energy consumption data of the corresponding server. Then, in the actual energy consumption data sequence of each server, the total number of actual energy consumption data greater than the preset energy consumption threshold of the corresponding server is counted and recorded as the actual cooling frequency of the corresponding server. For example, if in the actual energy consumption data sequence of a server, there are n actual energy consumption data greater than the preset energy consumption threshold of the server, then the actual cooling frequency of the server is n. In addition, the preset energy consumption threshold in this embodiment is 0, because when the temperature that the server needs to reduce during the historical liquid cooling cycle is 0, it indicates that the actual energy consumption value required during the cycle is also 0, so the calculated actual energy consumption data is 0.

[0019] Next, a two-dimensional mapping space is constructed, in which the horizontal axis is the actual cooling frequency and the vertical axis is the historical actual energy consumption data; then the historical actual energy consumption data and the actual energy consumption frequency of each server are mapped to the two-dimensional mapping space to obtain the energy consumption data points corresponding to the corresponding servers, and the horizontal axis value of the energy consumption data points corresponding to the servers is the actual cooling frequency of the corresponding servers, and the vertical axis value is the historical actual energy consumption data of the corresponding servers, that is, one server corresponds to one energy consumption data point.

[0020] After obtaining the energy consumption data points corresponding to the server, the distance between the energy consumption data points is calculated, and for the i-th energy consumption data point and the j-th energy consumption data point in the two-dimensional mapping space, the distance between the i-th energy consumption data point and the j-th energy consumption data point refers to the Euclidean distance between the i-th energy consumption data point and the j-th energy consumption data point, and i is not equal to j; then, according to the distance between the energy consumption data points, K-means clustering is performed on all energy consumption data points in the two-dimensional mapping space, and the clustering clusters obtained by clustering are all recorded as initial clustering clusters, and one energy consumption data point in the initial clustering cluster represents one server; and when performing K-means clustering, clustering The metric distance is the Euclidean distance between energy consumption data points. In this embodiment, the number of cluster centers during K-means clustering is consistent with the number of CDUs in the centralized liquid cooling system, and the number of CDUs needs to be set by the total number of servers covered by the centralized liquid cooling system and the number of servers covered by one CDU. The total number of servers covered by the liquid cooling system is the number of servers in the server group in this embodiment. For example, if the number of servers covered by one CDU is A1 and the number of servers in the server group is A0, then the number of CDUs in the centralized liquid cooling system is generally the rounded-up value of the result obtained by dividing A0 by A1.

[0021] Therefore, this embodiment can obtain the initial clustering cluster through the above process.

[0022] The second partitioning module 03 is used to redistribute the energy consumption data points in the initial clusters according to the total number of energy consumption data points in the initial clusters and the number of servers covered by the CDU in the liquid cooling system to obtain various target clusters.

[0023] Since the data center in this embodiment is an equalized cooling data center or a symmetrical heat dissipation architecture data center, it is required that the number of servers covered or served by each CDU of the centralized liquid cooling system in this embodiment is consistent, but the total number of energy consumption data points in the initial clustering cluster obtained above may be greater than or less than the number of servers covered by the set CDU. Therefore, this embodiment needs to redistribute the energy consumption data points in the initial clustering cluster according to the total number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system to obtain each target clustering cluster, and redistribute the energy consumption data points in the initial clustering cluster according to the total number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system. The specific process of obtaining each target clustering cluster is as follows: First, the number of servers covered by a single CDU of the centralized liquid cooling system is used as the quantity judgment threshold; then, it is determined whether the total number of energy consumption data points in the initial clustering cluster is greater than the quantity judgment threshold. If so, it indicates that there are redundant energy consumption data points in the corresponding initial clustering cluster, and it is necessary to allocate the redundant energy consumption data points in this type of initial clustering cluster to other initial clustering clusters. Then, this type of initial clustering cluster is recorded as a surplus clustering cluster; then, it is determined whether the total number of energy consumption data points in the initial clustering cluster is less than the quantity judgment threshold. If so, it indicates that the corresponding initial clustering cluster has a surplus energy consumption data point. There are some missing energy consumption data points in the initial clustering clusters. Such initial clustering clusters need to receive the allocated energy consumption data points. Then such initial clustering clusters are recorded as lacking clustering clusters. Then, it is determined whether the total number of energy consumption data points in the initial clustering clusters is equal to the quantity judgment threshold. If so, it indicates that the number of energy consumption data points in the corresponding initial clustering clusters is appropriate. The corresponding initial clustering clusters do not need to receive other energy consumption data points, nor do they need to redistribute the energy consumption data points in the corresponding initial clustering clusters to other clusters. Then such initial clustering clusters are recorded as target clustering clusters.

[0024] Next, this embodiment needs to reallocate the excess clusters and the lack clusters obtained above to obtain the target clusters, and this embodiment reallocates the excess clusters and the lack clusters according to the distance between each energy consumption data point in the excess clusters and the cluster center point of the lack clusters and the actual energy consumption data sequence of the server corresponding to each energy consumption data point in the excess clusters. Then, in this embodiment, according to the distance between each energy consumption data point in the excess clusters and the cluster center point of the lack clusters and the actual energy consumption data sequence of the server corresponding to each energy consumption data point in the excess clusters, the process of reallocating the excess clusters and the lack clusters to obtain the target clusters is as follows: First, according to the distance between each energy consumption data point in the excess cluster and the cluster center point of the deficiency cluster and the difference between adjacent actual energy consumption data in the actual energy consumption data sequence of the server corresponding to each energy consumption data point in the excess cluster, the degree of allocation corresponding to each energy consumption data point in each excess cluster is obtained; then, according to the degree of allocation corresponding to each energy consumption data point in each excess cluster and the quantity judgment threshold, the sequence of data points to be allocated corresponding to each excess cluster is obtained; then, according to the sequence of data points to be allocated corresponding to each excess cluster, the excess cluster and the deficiency cluster are reallocated again to obtain the allocated cluster corresponding to each excess cluster and the received cluster corresponding to each deficiency cluster, and all allocated clusters and all received clusters are recorded as target clusters.

[0025] In this embodiment, according to the distance between each energy consumption data point in the excess cluster and the cluster center point of the lacking cluster and the difference between adjacent actual energy consumption data in the actual energy consumption data sequence of the server corresponding to each energy consumption data point in the excess cluster, the specific process of obtaining the degree to be allocated corresponding to each energy consumption data point in each excess cluster is as follows: For the g-th energy consumption data point in any excess cluster G: first, obtain the server corresponding to the g-th energy consumption data point, and record the actual energy consumption data sequence of the server corresponding to the g-th energy consumption data point as the sequence to be analyzed, then obtain the difference sequence of the sequence to be analyzed, and record the cumulative sum of all differences in the difference sequence as the comprehensive difference, perform negative correlation mapping on the comprehensive difference, and record the mapping result as the credibility of the g-th energy consumption data point, and the h-th difference in the difference sequence is the absolute value of the difference between the h-th data and the h+1-th data in the sequence to be analyzed. The smaller the comprehensive difference, the higher the credibility of the sequence to be analyzed, and the more credible the initial clustering cluster to which the g-th energy consumption data point obtained by clustering belongs; then obtain the Euclidean distance between the cluster center point of each lacking cluster and the cluster center point of the excess cluster G, and record it as the corresponding lacking cluster The measured distance between the cluster and the excess cluster G is calculated, and the cluster center point of the lacking cluster corresponding to the minimum measured distance is used as the nearest neighbor cluster center point of the excess cluster G, that is, the cluster center point of the lacking cluster corresponding to the minimum measured distance is the cluster center point of the lacking cluster closest to the excess cluster G; then the Euclidean distance between the g-th energy consumption data point and the nearest neighbor cluster center point of the excess cluster G is calculated, and recorded as the nearest neighbor cluster distance of the g-th energy consumption data point; then the inverse of the credibility of the g-th energy consumption data point and the inverse of the nearest neighbor cluster distance of the g-th energy consumption data point are obtained, and the result of multiplying the inverse of the credibility of the g-th energy consumption data point by the inverse of the nearest neighbor cluster distance of the g-th energy consumption data point is recorded as the degree of allocation corresponding to the g-th energy consumption data point in the excess cluster G, that is, the degree of allocation corresponding to the g-th energy consumption data point is , R0 is the credibility of the g-th energy consumption data point, R1 is the nearest neighbor cluster distance of the g-th energy consumption data point, the credibility of the g-th energy consumption data point is exp(-D0), D0 is the comprehensive difference, and exp() is an exponential function with the constant e as the base; and when the degree of to-be-allocated corresponding to the g-th energy consumption data point is larger, it indicates that the corresponding energy consumption data point is more suitable to be allocated to the lacking clustering cluster, on the contrary, when the degree of to-be-allocated corresponding to the g-th energy consumption data point is smaller, it indicates that the corresponding energy consumption data point is more suitable to be retained in the initial clustering cluster where it is at this time, the larger the R0 is, the greater the credibility of the g-th energy consumption data point is, then the corresponding energy consumption data point is more suitable to be retained in the initial clustering cluster where it is at this time, the larger the R1 is, the farther the distance between the g-th energy consumption data point and the lacking clustering cluster is, then the corresponding energy consumption data point is more suitable to be retained in the initial clustering cluster where it is at this time.

[0026] In this embodiment, according to the to-be-allocated degree and quantity judgment threshold corresponding to each energy consumption data point in each excess clustering cluster, the specific process of obtaining the sequence of to-be-allocated data points corresponding to each excess clustering cluster is as follows: for any excess clustering cluster G: first, sort all energy consumption data points in the excess clustering cluster G in descending order according to the to-be-allocated degree, and record the sorted sequence as the energy consumption data point sequence corresponding to the excess clustering cluster G; then obtain the result of subtracting the quantity judgment threshold from the total number of energy consumption data points in the excess clustering cluster G, and record it as the number value of excess data points corresponding to the excess clustering cluster G. If the total number of energy consumption data points in the excess clustering cluster G is The quantity is 10 and the quantity judgment threshold is 8, then the number of excess data points corresponding to the excess cluster G is 2; then, in the energy consumption data point sequence corresponding to the excess cluster G, the front M continuous energy consumption data points are obtained, and the sequence formed by the obtained front M continuous energy consumption data points is recorded as the sequence of data points to be allocated corresponding to the excess cluster G, M is the number of excess data points corresponding to the excess cluster G, all data points in the sequence of data points to be allocated need to be reallocated to the lacking clusters, and the data points in the sequence of data points to be allocated corresponding to the excess cluster G are all the data points to be allocated in the excess cluster G.

[0027] And the allocation principle in this embodiment is: determine the allocation order of the excess clusters according to the number of data points to be allocated, determine the order of the lacking clusters to receive the data points to be allocated according to the distance from the excess clusters, determine the number of data points to be allocated that the lacking clusters can receive according to the difference between the total number of energy consumption data points in the lacking clusters and the quantity judgment threshold, determine the number of data points that the excess clusters need to allocate according to the number of data points to be allocated in the excess clusters, and determine the allocation order of the data points to be allocated in the excess clusters according to the sequence of data points to be allocated corresponding to the excess clusters, that is, the more the number of data points to be allocated, the earlier the corresponding excess clusters are allocated. The closer the shortage cluster is to the surplus cluster, the earlier it receives the to-be-allocated data points in the corresponding surplus cluster. The greater the difference between the total number of energy consumption data points in the shortage cluster and the quantity judgment threshold, the more data points the corresponding shortage cluster needs to receive. The more to-be-allocated data points in the surplus cluster, the more data points the corresponding surplus cluster needs to allocate. The data points in the front of the to-be-allocated data point sequence corresponding to the surplus cluster are allocated first. In addition, as other real-time methods, other allocation principles can also be set, for example, the fewer the number of to-be-allocated data points, the earlier the to-be-allocated data points in the corresponding surplus cluster are allocated, or the allocation order of the surplus cluster is determined according to the principle of random selection.

[0028] In this embodiment, according to the set allocation principle, the excess clusters and the lacking clusters are reallocated according to the sequence of data points to be allocated corresponding to each excess cluster, and the specific process of obtaining the allocated clusters corresponding to each excess cluster and the received clusters corresponding to each lacking cluster is as follows: First, the lack cluster sequence corresponding to each excess cluster is obtained, and the lack clusters in the lack cluster sequence corresponding to the excess cluster are arranged according to the distance from the corresponding excess cluster, that is, the more front-ranked lack clusters in the lack cluster sequence corresponding to any excess cluster are closer to the excess cluster, or the more front-ranked lack clusters in the lack cluster sequence corresponding to any excess cluster have a smaller measured distance from the excess cluster; then the starting reception number of each lack cluster is obtained, and the starting reception number of the lack cluster refers to the number of data points missing in the corresponding cluster when the corresponding lack cluster has not yet started to receive the data points to be allocated, that is, the starting reception number of any lack cluster is the result of the quantity judgment threshold minus the total number of original data points in the lack cluster, and the original data points in the lack cluster refer to the energy consumption data points that are divided into the corresponding lack cluster by K-means clustering.

[0029] Then determine whether the total number of data points to be allocated in the first excess cluster in the cluster sequence is less than or equal to the starting number of the first lacking cluster in the lacking cluster sequence corresponding to the first excess cluster. If so, allocate all the data points to be allocated in the first excess cluster to the first lacking cluster in the lacking cluster sequence corresponding to the first excess cluster, and allocate all the data points to be allocated in the first excess cluster to the lacking cluster corresponding to the first excess cluster. The first excess cluster after the first lacking cluster in the sequence is recorded as the allocated cluster corresponding to the first excess cluster, that is, the allocated cluster corresponding to the first excess cluster refers to the cluster after all the to-be-allocated data points in the first excess cluster are allocated to the lacking cluster, and then the starting receiving quantity of each lacking cluster is updated according to the number of to-be-allocated data points belonging to the first excess cluster received by the lacking cluster to obtain the first updated receiving quantity of each lacking cluster.

[0030] Then continue to determine whether the total number of data points to be allocated in the second excess cluster in the cluster sequence is less than or equal to the first updated received number of the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster; if not, determine whether the result of adding the first updated received number of the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster and the first updated received number of the second lacking cluster in the lacking cluster sequence corresponding to the second excess cluster is greater than or equal to the total number of data points to be allocated in the second excess cluster; if so, the first V1 data points to be allocated in the data point sequence to be allocated corresponding to the second excess cluster are allocated to the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster, and all data points to be allocated after the V1th data point to be allocated in the data point sequence to be allocated corresponding to the second excess cluster are allocated to the V2th data point to be allocated in the data point sequence to be allocated corresponding to the second excess cluster The first V1 data points to be allocated in the sequence of data points to be allocated corresponding to the second excess clustering cluster are allocated to the first lacking clustering cluster in the sequence of lacking clustering clusters corresponding to the second excess clustering cluster, and all the data points to be allocated after the V1th data point to be allocated in the sequence of data points to be allocated corresponding to the second excess clustering cluster are allocated to the second excess clustering cluster after the second lacking cluster in the sequence of lacking clustering clusters corresponding to the second excess clustering cluster, which is recorded as the allocated clustering cluster of the second excess clustering cluster, that is, the allocated clustering cluster corresponding to the second excess clustering cluster refers to the clustering cluster after all the data points to be allocated in the second excess clustering cluster are allocated to the lacking clustering cluster, and then the first updated reception quantity of each lacking clustering cluster is updated according to the number of data points to be allocated belonging to the second excess clustering cluster received by the lacking clustering cluster to obtain the second updated reception quantity of each lacking clustering cluster.

[0031] Then, continue to determine whether the total number of data points to be allocated in the third excess cluster in the cluster sequence is less than or equal to the first updated received number of the first lacking cluster in the lacking cluster sequence corresponding to the third excess cluster; if not, determine whether the sum of the second updated received number of the first lacking cluster in the lacking cluster sequence corresponding to the third excess cluster and the second updated received number of the second lacking cluster in the lacking cluster sequence corresponding to the third excess cluster is greater than or equal to the total number of data points to be allocated in the third excess cluster; if not, continue to determine whether the sum of the second updated received number of the first lacking cluster in the lacking cluster sequence corresponding to the third excess cluster is greater than or equal to the total number of data points to be allocated in the third excess cluster; Whether the result of accumulating the newly received number, the second updated received number of the second lacking cluster in the lacking cluster sequence corresponding to the third surplus cluster, and the second updated received number of the third lacking cluster in the lacking cluster sequence corresponding to the third surplus cluster is greater than or equal to the total number of data points to be allocated in the third surplus cluster; if so, the first Q1 data points to be allocated in the sequence of data points to be allocated corresponding to the third surplus cluster are allocated to the first lacking cluster in the sequence of lacking clusters corresponding to the third surplus cluster, and the Q1+1th data point to be allocated and the Q2th data point to be allocated in the sequence of data points to be allocated corresponding to the third surplus cluster are allocated. All the data points to be allocated between are allocated to the second lacking cluster in the lacking cluster sequence corresponding to the third excess cluster, the data points allocated to the second lacking cluster include the Q1+1th data point to be allocated and the Q2th data point to be allocated, all the data points to be allocated after the Q2th data point to be allocated in the data point sequence to be allocated corresponding to the third excess cluster are allocated to the third lacking cluster in the lacking cluster sequence corresponding to the third excess cluster, Q1 is the second updated received quantity of the first lacking cluster in the lacking cluster sequence corresponding to the third excess cluster, Q2 is the second lacking cluster in the lacking cluster sequence corresponding to the second excess cluster. The second updated reception quantity of the lacking cluster, the Q2th to-be-allocated data point belongs to the to-be-allocated data point sequence corresponding to the third excess cluster, and the cluster after all the to-be-allocated data points in the third excess cluster are allocated to the lacking cluster is recorded as the allocated cluster corresponding to the third excess cluster, and then the second updated reception quantity of each lacking cluster is updated according to the number of to-be-allocated data points belonging to the third excess cluster received by the lacking cluster to obtain the third updated reception quantity of each lacking cluster; and the allocation of all excess clusters is completed according to the above process, and after obtaining the allocated cluster corresponding to the last excess cluster in the cluster sequence, it is determined that the allocation is completed.

[0032] In addition, after the allocation is completed, the cluster formed by all the data points to be allocated received by each lacking cluster during the allocation process and all the original data points in the corresponding lacking cluster is recorded as the received cluster corresponding to the corresponding lacking cluster, that is, for any lacking cluster, the cluster formed by all the data points to be allocated received by the lacking cluster during the allocation process and all the original data points in the lacking cluster is the received cluster corresponding to the lacking cluster.

[0033] In this embodiment, the specific process of updating the initial reception number of the lacking cluster to obtain the first updated reception number corresponding to the lacking cluster, updating the first updated reception number of the lacking cluster to obtain the second updated reception number corresponding to the lacking cluster, and updating the second updated reception number of each lacking cluster to obtain the third updated reception number of each lacking cluster is: For any lacking cluster F: if the lacking cluster F receives the to-be-allocated data points allocated by the first excess cluster, then the result of subtracting the number of to-be-allocated data points received by the lacking cluster F from the number of to-be-allocated data points belonging to the first excess cluster received by the lacking cluster F is recorded as the first updated received number of the lacking cluster F. If the lacking cluster F has not received the to-be-allocated data points belonging to the first excess cluster, then the initial received number of the lacking cluster F is recorded as the first updated received number of the lacking cluster F. The first updated number of receptions; if the lacking cluster F receives the to-be-allocated data points allocated from the second excess cluster, then the result of deducting the number of to-be-allocated data points received by the lacking cluster F from the first updated number of receptions is recorded as the second updated number of receptions of the lacking cluster F. If the lacking cluster F has not received the to-be-allocated data points belonging to the second excess cluster, then the first updated number of receptions of the lacking cluster F is recorded as the second updated number of receptions of the lacking cluster F. Similarly, if the lacking cluster F receives the to-be-allocated data points allocated from the third excess cluster, then the result of subtracting the number of to-be-allocated data points received by the lacking cluster F from the second updated received number of the lacking cluster F and belonging to the third excess cluster is recorded as the third updated received number of the lacking cluster F. If the lacking cluster F has never received the to-be-allocated data points belonging to the third excess cluster, then the second updated received number of the lacking cluster F is recorded as the third updated received number of the lacking cluster F. The third updated reception quantity of cluster F; for example, if the initial reception quantity of the lacking cluster F is 3, and in the process of allocating the to-be-allocated data points in the first excess cluster, the lacking cluster F receives 2 to-be-allocated data points allocated from the first excess cluster, then the first updated reception quantity of the lacking cluster F is 1, and the purpose of updating the reception quantity of the lacking cluster is to ensure that the data points received by all the lacking clusters are not greater than the maximum limit value that the lacking clusters can receive.

[0034] For example: if the number of servers covered by CDU is 8, that is, the number judgment threshold is 8, if the number of excess clusters is 3, namely excess cluster W1, excess cluster W2 and excess cluster W3, if the number of lack clusters is 3, namely lack cluster U1, lack cluster U2 and lack cluster U3, if the total number of data points to be allocated in excess cluster W1 is 6, the total number of data points to be allocated in excess cluster W2 is 4, and the total number of data points to be allocated in excess cluster W3 is 2, the starting number of received lack cluster U1 is is 5, the starting number of received data points of the lacking cluster U2 is 4, the starting number of received data points of the lacking cluster U3 is 3, the lacking cluster sequence corresponding to the excess cluster W1 is {U2,U1,U3}, the lacking cluster sequence corresponding to the excess cluster W2 is {U1,U3,U2}, and the lacking cluster sequence corresponding to the excess cluster W3 is {U3,U2,U1}. Then the allocation process of the excess cluster W1 is: first allocate the first 4 to-be-allocated data points in the sequence of to-be-allocated data points corresponding to the excess cluster W1 to U2, and then allocate the first 4 to-be-allocated data points in the sequence of to-be-allocated data points corresponding to the excess cluster W1 to U2. The last two data points to be allocated in the sequence of data points to be allocated are allocated to U1, and then the first updated reception number of the lacking clusters is obtained according to the process of allocating the data points to be allocated in the excess cluster W1. At this time, the first updated reception number of the lacking cluster U1 is 3, the first updated reception number of the lacking cluster U2 is 0, and the first updated reception number of the lacking cluster U3 is 3; the allocation process of the excess cluster W2 is: first allocate the first three data points to be allocated in the sequence of data points to be allocated corresponding to the excess cluster W2 to U1, and then allocate the excess cluster W2 to U1. The last data point to be allocated in the sequence of data points to be allocated corresponding to cluster W2 is allocated to U3, and then the second updated reception number of the lacking clusters is obtained according to the process of allocating the data points to be allocated in the excess cluster W2. At this time, the second updated reception number of the lacking cluster U1 is 0, the second updated reception number of the lacking cluster U2 is 0, and the first updated reception number of the lacking cluster U3 is 2; the allocation process of the excess cluster W3 is: allocating all the data points to be allocated in the sequence of data points to be allocated corresponding to the excess cluster W3 to U3;The allocated clusters corresponding to the excess cluster W1, the excess cluster W2 and the excess cluster W3 are all clusters after allocating all the to-be-allocated data points in the corresponding excess cluster to the lacking cluster. The received cluster corresponding to the lacking cluster U1 is composed of all the original data points in the lacking cluster U1, the first three to-be-allocated data points in the to-be-allocated data point sequence corresponding to the excess cluster W2 and the last two to-be-allocated data points in the to-be-allocated data point sequence corresponding to the excess cluster W1. The received cluster corresponding to the lacking cluster U2 is composed of the first four to-be-allocated data points in the to-be-allocated data point sequence corresponding to the excess cluster W1 and all the original data points in the lacking cluster U2. The received cluster corresponding to the lacking cluster U3 is composed of all the original data points in the lacking cluster U3, the last one to-be-allocated data point in the to-be-allocated data point sequence corresponding to the excess cluster W2 and all the to-be-allocated data points in the to-be-allocated data point sequence corresponding to the excess cluster W3. ;

[0035] Therefore, this embodiment obtains all target clusters through the above process.

[0036] The liquid cooling system management module 04 is used to reallocate and manage the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target cluster.

[0037] After obtaining the target cluster, the servers corresponding to all energy consumption data points in the target cluster are connected to the CDU of the same liquid cooling system; that is, the servers corresponding to all energy consumption data points in the same target cluster are connected to the same CDU of the centralized liquid cooling system using CDU quick connectors. This method of connecting servers of the same target cluster to the same CDU can reduce the energy waste and energy waste of the centralized liquid cooling system during cooling, improve the energy consumption utilization rate and energy utilization rate of the centralized liquid cooling system, and realize the balanced management of energy consumption and energy of the centralized liquid cooling system.

[0038] At this point, this embodiment has completed the reallocation of servers controlled by the CDU, reduced the energy waste and energy waste during cooling of the centralized liquid cooling system, improved the energy utilization rate and energy utilization rate of the centralized liquid cooling system, and achieved balanced management of the energy consumption and energy of the centralized liquid cooling system.

[0039] In summary, this embodiment includes a data acquisition module for acquiring the actual energy consumption data of each server in the server group under each historical liquid cooling cycle; a first partitioning module for obtaining the energy consumption data points corresponding to the server according to the mean of the actual energy consumption data sequence and the number of data in the actual energy consumption data sequence that is greater than the preset energy consumption threshold, and clustering the energy consumption data points according to the distance between the energy consumption data points to obtain the initial clustering cluster; a second partitioning module for redistributing the energy consumption data points in the initial clustering cluster according to the number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system to obtain each target clustering cluster; a liquid cooling system management module for redistributing and managing the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target clustering cluster. Moreover, this embodiment can reduce the energy consumption waste and energy waste of the centralized liquid cooling system during cooling, improve the energy consumption utilization rate and energy utilization rate of the centralized liquid cooling system, and realize the balanced management of energy consumption and energy of the centralized liquid cooling system.

[0040] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based liquid cooling system energy consumption data management platform, characterized in that: The liquid cooling system energy consumption data management platform includes: A data acquisition module, used to acquire actual energy consumption data of each server in a server group under each historical liquid cooling cycle, where the liquid cooling cycle is completed by a liquid cooling system, and the server group is a server group served by the liquid cooling system; The first partitioning module is used to record the sequence composed of the actual energy consumption data of the server under all historical liquid cooling cycles as the actual energy consumption data sequence of the corresponding server, and obtain the energy consumption data points corresponding to the server according to the mean value of the actual energy consumption data sequence and the number of data in the actual energy consumption data sequence that is greater than a preset energy consumption threshold, and cluster the energy consumption data points according to the distance between the energy consumption data points to obtain an initial clustering cluster; A second partitioning module is used to redistribute the energy consumption data points in the initial clustering cluster according to the total number of energy consumption data points in the initial clustering cluster and the number of servers covered by the CDU in the liquid cooling system to obtain various target clustering clusters; The liquid cooling system management module is used to reallocate and manage the servers controlled by the CDU of the liquid cooling system according to the servers corresponding to the energy consumption data points in the target cluster.

2. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 1, characterized in that: The method for acquiring the actual energy consumption data of each server in each historical liquid cooling cycle includes: For any server: obtain the temperature that needs to be lowered for the server during the ath historical liquid cooling cycle, and record it as the temperature change; obtain the inverse of the refrigerator energy efficiency ratio of the liquid cooling system, and record it as the characteristic ratio; multiply the cold night flow rate, cold night density, cold night specific heat capacity, the temperature change, the overall cycle time of the server in the ath historical liquid cooling cycle, and the characteristic ratio in the circulation branch during the ath historical liquid cooling cycle of the server, and record the result as the actual energy consumption data of the server under the ath historical liquid cooling cycle.

3. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 1, characterized in that: The method for acquiring the energy consumption data points corresponding to the server includes: The mean of the actual energy consumption data sequence of the server is recorded as the historical actual energy consumption data of the corresponding server, and the total number of actual energy consumption data greater than the preset energy consumption threshold in the actual energy consumption data sequence of the server is recorded as the actual cooling frequency of the corresponding server; a two-dimensional mapping space is constructed, the horizontal axis of the two-dimensional mapping space is the actual cooling frequency, and the vertical axis is the historical actual energy consumption data; the historical actual energy consumption data and the actual energy consumption frequency of the server are mapped to the two-dimensional mapping space to obtain the energy consumption data points corresponding to the corresponding server.

4. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 3, characterized in that: The method for obtaining the initial clustering clusters includes: The Euclidean distance between any two energy consumption data points in the two-dimensional mapping space is recorded as the metric distance between the corresponding two energy consumption data points; according to the Euclidean distance between any two energy consumption data points in the two-dimensional mapping space, all energy consumption data points in the two-dimensional mapping space are clustered using a clustering algorithm, and the cluster clusters obtained by clustering are recorded as initial cluster clusters.

5. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 1 is characterized in that: The method of redistributing the energy consumption data points in the initial clusters to obtain each target cluster includes: The number of servers covered by a single CDU is used as the quantity judgment threshold; if the total number of energy consumption data points in the initial clustering cluster is greater than the quantity judgment threshold, the corresponding initial clustering cluster is recorded as a surplus clustering cluster; if the total number of energy consumption data points in the initial clustering cluster is less than the quantity judgment threshold, the corresponding initial clustering cluster is recorded as a lacking clustering cluster; if the total number of energy consumption data points in the initial clustering cluster is equal to the quantity judgment threshold, the corresponding initial clustering cluster is recorded as a target clustering cluster; According to the distance between each energy consumption data point in the excess cluster and the cluster center point of the deficiency cluster and the difference between adjacent actual energy consumption data in the actual energy consumption data sequence of the server corresponding to each energy consumption data point in the excess cluster, the degree to be allocated corresponding to each energy consumption data point in the excess cluster is obtained; according to the degree to be allocated corresponding to each energy consumption data point in the excess cluster and the quantity judgment threshold, the sequence of data points to be allocated corresponding to each excess cluster is obtained; according to the sequence of data points to be allocated corresponding to the excess cluster, the excess cluster and the deficiency cluster are reallocated to obtain the allocated cluster corresponding to each excess cluster and the received cluster corresponding to each deficiency cluster, and all allocated clusters and all received clusters are recorded as target clusters, and the data points in the sequence of data points to be allocated are the data points to be allocated.

6. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 5, characterized in that: The method for obtaining the degree of allocation corresponding to each energy consumption data point in the excess cluster includes: For the g-th energy consumption data point in any excess cluster G: record the actual energy consumption data sequence of the server corresponding to the g-th energy consumption data point as the sequence to be analyzed, obtain the difference sequence of the sequence to be analyzed, record the cumulative sum of all differences in the difference sequence as the comprehensive difference, record the negative correlation mapping value of the comprehensive difference as the credibility of the g-th energy consumption data point, and the h-th difference in the difference sequence is the absolute value of the difference between the h-th data and the h+1-th data in the sequence to be analyzed; obtain the cluster center point of the lacking cluster closest to the excess cluster G, and record it as the nearest neighbor cluster center point of the excess cluster G, and record the Euclidean distance between the g-th energy consumption data point and the nearest neighbor cluster center point as the nearest neighbor cluster distance of the g-th energy consumption data point; and record the result of multiplying the inverse of the credibility by the inverse of the nearest neighbor cluster distance as the degree to be allocated corresponding to the g-th energy consumption data point in the excess cluster G.

7. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 5, characterized in that: The method for obtaining the sequence of data points to be allocated corresponding to the excess clusters includes: For any excess cluster: sort the energy consumption data points in the excess cluster in descending order of the degree to be allocated to obtain a sequence of energy consumption data points corresponding to the excess cluster; record the result of subtracting the quantity judgment threshold from the total number of energy consumption data points in the excess cluster as the number of excess data points corresponding to the excess cluster; in the sequence of energy consumption data points corresponding to the excess cluster, record the sequence composed of the front M energy consumption data points as the sequence of data points to be allocated corresponding to the excess cluster, where M is the number of excess data points of the excess cluster.

8. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 5, characterized in that: The method for acquiring the allocated clusters corresponding to the excess clusters and the received clusters corresponding to the lacking clusters includes: Obtain the excess cluster sequence and the deficiency cluster sequence corresponding to the excess cluster, wherein the deficiency clusters that are arranged in front in the deficiency cluster sequence corresponding to the excess cluster are closer to the corresponding excess cluster, and the excess cluster sequence is arranged according to the number of data points to be allocated in the cluster; obtain the initial reception quantity of the deficiency cluster, wherein the initial reception quantity of the deficiency cluster is the result of subtracting the total number of original data points in the corresponding deficiency cluster from the quantity judgment threshold, and the original data points in the deficiency cluster refer to the energy consumption data points that are divided into the deficiency cluster through clustering; Determine whether the total number of data points to be allocated in the first excess cluster in the cluster sequence is less than or equal to the starting number of data points received in the first lacking cluster in the lacking cluster sequence corresponding to the first excess cluster. If so, allocate all data points to be allocated in the first excess cluster to the first cluster after the first lacking cluster in the lacking cluster sequence corresponding to the first excess cluster, record it as the allocated cluster corresponding to the first excess cluster, and allocate the first excess cluster to the first cluster. The starting reception quantity of the lacking cluster is updated to obtain the first updated reception quantity of the lacking cluster; continue to determine whether the total number of data points to be allocated in the second excess cluster in the cluster sequence is less than or equal to the first updated reception quantity of the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster; if not, determine whether the sum of the first updated reception quantity of the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster and the first updated reception quantity of the second lacking cluster in the lacking cluster sequence corresponding to the second excess cluster is greater than or equal to the total number of data points to be allocated in the second excess cluster; if so, allocate the front V1 data points to be allocated in the sequence of data points to be allocated corresponding to the second excess cluster to the first lacking cluster in the sequence of lacking clusters corresponding to the second excess cluster, and allocate the V1th data points to be allocated in the sequence of data points to be allocated corresponding to the second excess cluster All data points to be allocated after the allocated data point are allocated to the second excess cluster after the second lacking cluster in the lacking cluster sequence corresponding to the second excess cluster, and are recorded as the allocated cluster corresponding to the second excess cluster, and so on, until the allocated cluster corresponding to the last excess cluster in the cluster sequence is obtained, and then the allocation is determined to be complete; V1 is the first updated reception quantity of the first lacking cluster in the lacking cluster sequence corresponding to the second excess cluster; The cluster formed by all the to-be-allocated data points received by the lacking cluster in the allocation process and all the original data points in the corresponding lacking cluster is recorded as the received cluster corresponding to the corresponding lacking cluster.

9. The artificial intelligence-based liquid cooling system energy consumption data management platform according to claim 8, characterized in that: The method of updating the initial reception quantity of the lacking cluster to obtain the first updated reception quantity of the lacking cluster comprises: For any deficient cluster, if the deficient cluster receives the to-be-allocated data points allocated from the first excess cluster, then the result of subtracting the number of to-be-allocated data points belonging to the first excess cluster received by the deficient cluster from the initial reception number of the deficient cluster is recorded as the first updated reception number of the deficient cluster; if the deficient cluster has not received the to-be-allocated data points belonging to the first excess cluster, then the initial reception number of the deficient cluster is used as the first updated reception number of the deficient cluster.

10. The liquid cooling system energy consumption data management platform based on artificial intelligence according to claim 1, characterized in that: The method for reallocating and managing servers controlled by the CDU of the liquid cooling system according to servers corresponding to energy consumption data points in the target cluster includes: For any target cluster, all servers corresponding to the energy consumption data points in the target cluster are connected to the CDU of the same liquid cooling system.

Images