Data center energy-saving scheduling method based on energy consumption correlation graph

Abstract

This invention discloses a data center energy-saving scheduling method based on energy consumption correlation graphs, belonging to the field of data center energy-saving technology. It addresses the problems of lacking a quantitative model of inter-rack thermal interference during the cooling release window of a cold storage device and the inability to predict cooling costs during task scheduling. By reading the remaining cooling capacity of the cold storage device and combining it with the rack inlet air temperature and airflow velocity to generate an overcooling risk index, risk racks are marked. The response sensitivity is calculated by retrieving the fan speed baseline. A thermal interference diffusion graph is constructed with the overcooled risk racks as source nodes. The cooling cost of each rack is calculated to generate a task landing point avoidance set. The heat dissipation tasks are migrated to the rack with the minimum cooling cost to complete the landing point pre-adjustment. After the window ends, the actual cooling cost is collected and compared with the predicted value to generate a weighted deviation rate and correct the graph edge weights, realizing continuous adaptive updating of the graph and reducing the overall energy consumption of the data center.

Description

Technical Field

[0001] This invention relates to the field of data center energy-saving technology, and more specifically, to a data center energy-saving scheduling method based on energy consumption correlation maps. Background Technology

[0002] Data centers are a core component of information infrastructure, and their energy consumption has long been a concern in the industry. As a peak-shaving and valley-filling cooling auxiliary means, cold storage devices store cold energy during off-peak hours and release it during peak hours, which can effectively reduce the instantaneous cooling pressure on air conditioning systems. However, when cold storage devices release a large amount of cold energy during the release window, it can cause a sudden drop in the local temperature of the rack. This can cause the server BMC system to misjudge the environment as a high-temperature environment and reduce the fan speed, resulting in airflow failure and heat transfer interference to adjacent racks, ultimately causing global energy consumption fluctuations.

[0003] The existing technology has the following shortcomings: Currently, existing technologies lack quantitative modeling methods for the propagation of inter-rack thermal interference during the cooling release window of cold storage devices. They cannot predict energy consumption correlation weights based on rack spatial topology and fan response characteristics. This results in the task scheduling system being unable to predict the actual cooling cost of each rack when formulating landing plans, causing high heat dissipation tasks to be assigned to areas with superimposed thermal interference, which in turn leads to excessive cooling costs and increased overall energy consumption. Therefore, an energy-saving scheduling method for data centers based on energy consumption correlation maps is proposed.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a data center energy-saving scheduling method based on energy consumption correlation graphs. This method addresses the problems mentioned in the background art by employing methods such as overcooling risk index quantification, thermal interference diffusion graph construction, cooling cost prediction and task landing point pre-adjustment, and graph weight closed-loop correction.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data center energy-saving scheduling method based on energy consumption correlation maps, comprising the following steps: Step S1: Read the remaining cooling capacity data of the cold storage device, combine the current air inlet temperature of each rack with the airflow velocity in the rack to generate the overcooling risk index of each rack, and determine whether to mark the current rack as overcooling risk based on the overcooling risk index. Step S2: Retrieve the server fan speed baseline, combine the overcooling risk index to evaluate the response sensitivity of each overcooling risk rack during the cooling release window, obtain the spatial position relationship data of each rack, take the overcooling risk rack as the source node and the spatial position relationship as the edge, and calculate the energy consumption correlation weight of each adjacent rack in combination with the response sensitivity to construct a thermal interference diffusion map. Step S3: Collect the heat dissipation intensity of each heat dissipation task in the current task queue, calculate the cooling cost of each rack undertaking the heat dissipation task during the cooling release window, generate a task landing point avoidance set, filter heat dissipation tasks based on the task landing point avoidance set, and complete the pre-adjustment of task landing points during the cooling release window. Step S4: After the cooling window ends, collect the actual cooling cost data of each rack, compare it with the cooling cost in the thermal interference diffusion map before the task landing point is pre-adjusted, generate the weight deviation rate, and correct and update the weight of the corresponding edge in the thermal interference diffusion map.

[0007] In a preferred embodiment, in step S1, before the cold storage device release window is opened, the remaining cold energy data of the cold storage device sensor is read, and the remaining cold energy data is the total amount of cold energy that the cold storage device has not yet released; the air inlet temperature of each rack temperature sensor is collected, and the air inlet temperature refers to the current air temperature on the air inlet side of the rack; the airflow velocity inside the rack is collected from each rack wind speed sensor, and the airflow velocity inside the rack is the average velocity of the airflow inside the rack. After normalizing the remaining cooling capacity, the inlet air temperature of each rack, and the airflow velocity in each rack, the overcooling risk index of the rack is obtained by weighted summation.

[0008] In a preferred embodiment, in step S1, the overcooling risk index of each rack is compared with a preset overcooling risk judgment threshold: when the overcooling risk index is greater than or equal to the overcooling risk judgment threshold, the rack is marked as an overcooling risk rack. When the overcooling risk index is less than the overcooling risk judgment threshold, the rack is marked as a normal rack.

[0009] In a preferred embodiment, in step S2, the BMC equipment speed baseline of the server corresponding to each rack with overcooling risk is retrieved. The BMC equipment speed baseline is the rated speed reference value of the server fan in the rack under normal temperature environment. The fan speed reduction ratio is calculated based on the BMC equipment speed baseline, and the response sensitivity of the overcooled risk rack is calculated by combining the overcooled risk index of the overcooled risk rack.

[0010] In a preferred embodiment, in step S2, the spatial location relationship data of the data center topology database is accessed, and the spatial location relationship data includes the three-dimensional spatial coordinates of each rack; Each rack at risk of excessive cooling is taken as a source node. For each source node, other racks in the data center are traversed, and the spatial distance between the source node and the target rack is calculated using three-dimensional spatial coordinates. When the spatial distance is not greater than the preset spatial influence radius, a directed edge is established from the source node to the target rack, and the energy consumption association weight of the directed edge is calculated. When the spatial distance is greater than the preset spatial influence radius, no directed edge is established; all directed edges and their weights are summarized to form a thermal interference diffusion map.

[0011] In a preferred embodiment, in step S3, the heat dissipation intensity of each heat dissipation task in the current task queue of the task scheduling system is read. The heat dissipation intensity is the heat generated by the heat dissipation task per unit time when it is running at full load. For each rack to be scheduled in the task queue, the energy consumption correlation weight of its corresponding directed edge is extracted from the thermal interference diffusion map. The energy consumption correlation weight is summed to obtain the thermal interference superposition intensity of the rack to be scheduled, and the cooling cost when the task landing point is the rack to be scheduled is calculated. The cooling cost of each rack is compared with the preset upper limit threshold of cooling cost: when the cooling cost of the rack to be scheduled is greater than or equal to the upper limit threshold of cooling cost, the rack to be scheduled is added to the task landing point avoidance set. When the cooling cost of the rack to be scheduled is less than the upper limit threshold of cooling cost, the rack to be scheduled will be retained in the available landing point candidate set.

[0012] In a preferred embodiment, in step S3, tasks with heat dissipation intensity greater than or equal to a preset heat dissipation intensity threshold in the task queue are determined to be heat dissipation tasks. Their original landing point is moved from the rack in the task landing point avoidance set to the rack with the minimum cooling cost in the available landing point candidate set. The task landing point pre-adjustment is completed, and a cooling release window scheduling scheme is generated. For tasks with a heat dissipation power value less than the preset heat dissipation intensity threshold, the original landing point will remain unchanged, and no migration operation will be performed.

[0013] In a preferred embodiment, in step S4, at the end of the cooling release window, the actual cooling cost data of each rack during this window period is collected by the cooling cost interface of the energy management system database. The actual cooling cost data is the actual electrical power consumed by the cooling equipment in the corresponding rack service area; for each directed edge in the thermal interference map, the actual cooling cost data of the rack corresponding to the directed edge and the rack cooling cost are taken to calculate the weighted deviation rate.

[0014] In a preferred embodiment, in step S4, when the absolute value of the weight deviation rate is greater than or equal to the preset deviation trigger threshold, weight correction is performed, and the corrected energy consumption correlation weight is calculated by combining the weight deviation rate and the energy consumption correlation weight. When the absolute value of the weight deviation rate is less than the preset deviation trigger threshold, the current weight error is determined to be within an acceptable range, and the energy consumption associated weight remains unchanged. After correcting all edges in the current version of the thermal interference map, the updated map is stored in the energy management system to generate the corrected thermal interference diffusion map.

[0015] Technical effects and advantages of the invention: This invention quantifies the energy consumption correlation weights between racks by constructing a thermal interference diffusion map, achieving accurate modeling of the thermal interference propagation relationship between racks within the cooling release window, and providing a reliable basis for predicting cooling costs for scheduling decisions. Through a task landing point pre-adjustment mechanism, high heat dissipation tasks are migrated from racks with excessive cooling costs to racks with the lowest costs in the candidate set, effectively reducing the overall energy consumption within the cooling release window. By collecting actual cooling cost data and performing closed-loop correction on the map weights, continuous adaptive updating of the thermal interference diffusion map is achieved, gradually improving the prediction accuracy over time. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of the data center energy-saving scheduling method based on energy consumption correlation graphs according to the present invention.

[0017] Figure 2 This is a schematic diagram illustrating the steps of the data center energy-saving scheduling method based on energy consumption correlation graphs according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention quantifies the energy consumption correlation between racks by constructing a thermal interference diffusion map, combines cooling cost prediction and task landing point pre-adjustment mechanism, and uses actual collected data to perform closed-loop correction of map weights, thereby reducing the overall energy consumption of the data center during the cooling release window and improving energy-saving scheduling efficiency.

[0020] Example 1: Please refer to Figures 1 to 2 The data center energy-saving scheduling method based on energy consumption correlation graphs has the following specific operation process: Step S1: Read the remaining cooling capacity data of the cold storage device, combine the current air inlet temperature of each rack with the airflow velocity in the rack to generate the overcooling risk index of each rack, and determine whether to mark the current rack as overcooling risk based on the overcooling risk index. Step S2: Retrieve the server fan speed baseline, combine the overcooling risk index to evaluate the response sensitivity of each overcooling risk rack during the cooling release window, obtain the spatial position relationship data of each rack, take the overcooling risk rack as the source node and the spatial position relationship as the edge, and calculate the energy consumption correlation weight of each adjacent rack in combination with the response sensitivity to construct a thermal interference diffusion map. Step S3: Collect the heat dissipation intensity of each heat dissipation task in the current task queue, calculate the cooling cost of each rack to undertake the heat dissipation task during the cooling release window, generate a task landing point avoidance set, filter heat dissipation tasks based on the task landing point avoidance set, and complete the pre-adjustment of task landing points during the cooling release window. Step S4: After the cooling window ends, collect the actual cooling cost data of each rack, compare it with the cooling cost in the thermal interference diffusion map before the task landing point is pre-adjusted, generate the weight deviation rate, and correct and update the weight of the corresponding edge in the thermal interference diffusion map.

[0021] The specific implementation is as follows: In step S1, before the cold storage device releases its cooling window, the risk of overcooling in each rack is quantitatively assessed. Read the remaining cold energy data from the sensors of the cold storage device. The remaining cold energy data is the total amount of cold energy that the cold storage device has not yet released. The higher the remaining cold energy data, the greater the cold energy impact on the rack during the cold release window. The system collects the inlet air temperature from the temperature sensors of each rack. The inlet air temperature refers to the current air temperature on the air inlet side of the rack, which is used to reflect the current state of the thermal environment in which the rack is located. The system also collects the airflow velocity inside the rack from the wind speed sensors of each rack. The airflow velocity inside the rack is the average velocity of the airflow inside the rack, which reflects the effectiveness of the current airflow organization.

[0022] The remaining cooling capacity, the inlet air temperature of each rack, and the airflow velocity in each rack were normalized respectively. The larger the difference between the inlet air temperature and the rack design inlet air temperature baseline, the closer it is to the subcooled state; the direction is taken by the complementary value. The lower the ratio of airflow velocity to rated airflow velocity, the more fragile the airflow organization. The direction is also taken as the complementary value to ensure that all three components have the same value in the direction. The larger the value, the higher the risk.

[0023] Based on the three normalized components, the overcooling risk index of each rack is calculated by weighting and summing according to the weight coefficients of each component: ;in, For the first The overcooling risk index of each rack The remaining cold energy of the cold storage device The rated total cooling capacity of the cold storage device. For the first Normalized value of rack inlet air temperature For the first The normalized value of the airflow velocity within each rack, where α, β, and γ are the weighting coefficients of each component, satisfying α+β+γ=1.

[0024] It should be noted that the baseline values ​​for the rack design inlet air temperature and the rated airflow velocity are given by professionals in the field based on actual conditions, and will not be elaborated here.

[0025] Compare the overcooling risk index of each rack with the preset overcooling risk judgment threshold: When the overcooling risk index is greater than or equal to the overcooling risk judgment threshold, the rack is marked as an overcooling risk rack, included in the risk rack identifier set, and processed in subsequent steps. When the overcooling risk index is less than the overcooling risk judgment threshold, the rack is marked as a normal rack and will not participate in subsequent processing steps.

[0026] Through the above-mentioned quantification and labeling of supercooling risk, a supercooling risk index sequence and a set of risk rack identifiers are generated, providing a basis for the range and intensity of risk racks to be constructed in the subsequent construction of thermal interference diffusion maps.

[0027] It should be noted that the weighting coefficients of α, β, and γ are calibrated using the least squares fitting method based on the statistical data of rack overcooling events in historical cooling release windows; the overcooling risk judgment threshold is determined by taking the lower quartile value of the overcooling risk index distribution of airflow short circuits in historical cooling release windows; the normalization of inlet air temperature and airflow velocity adopts interval scaling linear transformation, and the value range is determined by the maximum and minimum values ​​of rack operation records throughout the year.

[0028] In step S2, after the risk rack marking is completed, a thermal interference diffusion map reflecting the energy consumption correlation between racks is constructed for each overcooled risk rack in the risk rack identification set.

[0029] Retrieve the BMC equipment speed baseline of the server corresponding to each rack at risk of overcooling. The BMC equipment speed baseline is the rated speed reference value of the server fan in each rack under normal temperature conditions, reflecting the starting point of fan speed reduction when receiving a temperature misjudgment signal.

[0030] The overcooling risk index of each rack at risk of overcooling is combined with the fan speed reduction ratio to calculate the response sensitivity of each rack at risk of overcooling. The calculation formula is as follows: ;in, For the first The responsiveness of an overcooled risk rack. The supercooling risk index from step S1, For the first The baseline speed value of the BMC equipment corresponds to the rack with the risk of overcooling. For the first The current actual rotation speed of the rack server fan at the risk of overcooling. This determines the percentage by which the fan speed is reduced.

[0031] The higher the response sensitivity value, the more likely the rack is to experience airflow failure due to fan speed reduction during the cooling window, and the stronger the energy consumption interference transmitted to adjacent racks.

[0032] Access the spatial location relationship data of the data center topology database, which includes the three-dimensional spatial coordinates of each rack.

[0033] Using each overcooled risk rack in the risk rack identifier set as the source node, for each source node, traverse other racks in the data center and calculate the spatial distance between the source node and the target rack using three-dimensional spatial coordinates; When the spatial distance is not greater than the preset spatial influence radius, a directed edge is established from the source node to the target rack, and the energy consumption association weight of this edge is calculated using the following formula: ;in, The energy consumption correlation weight between the overcooled risk rack i and the target rack j is defined. For the response sensitivity of rack i, Let be the Euclidean distance between the three-dimensional coordinates of rack i and rack j. This is the spatial attenuation coefficient; When the spatial distance is greater than the preset spatial influence radius, no directed edge is established; All directed edges and their weights are aggregated to form a thermal interference diffusion map.

[0034] Through the above response sensitivity calculation and graph construction process, a rack sensitivity rating set and a thermal interference diffusion graph are generated, providing a basis for the energy consumption correlation weight of each rack for subsequent calculation of the cooling cost during the cooling release window.

[0035] It should be noted that the preset spatial influence radius is determined based on actual airflow organization data in the computer room, taking the maximum transmission distance at which the thermal interference signal from adjacent racks attenuates to the background noise level; the spatial attenuation coefficient... Based on the airflow impedance characteristics of the closed aisle structure or open rack layout of the computer room, the BMC equipment speed baseline is calibrated through actual measurement during the construction phase; the average fan speed under rated operating conditions over the past thirty days is collected periodically through the server management interface.

[0036] In step S3, after the thermal interference diffusion map is constructed, the cooling cost of each rack during the cooling release window is calculated based on the map weights, and the landing point of the heat dissipation task is pre-adjusted. Read the heat dissipation intensity of each heat dissipation task in the current task queue of the task scheduling system. The heat dissipation intensity is the heat generated by the heat dissipation task per unit time when it is running at full load, which is estimated by the task scheduling system based on historical running data. For each rack to be scheduled in the task queue, the energy consumption correlation weights of all directed edges pointing to that rack are extracted from the thermal interference diffusion map. The summation of the weights is used to obtain the thermal interference superposition intensity of that rack, and then the cooling cost when the task lands on that rack is calculated: ;in, The cost of cooling when the mission lands at rack j. The heat dissipation intensity of the tasks to be scheduled. It is the sum of the energy consumption correlation weights of all directed edges pointing to rack j in the thermal interference diffusion map.

[0037] Compare the cooling cost of each rack with the preset upper limit threshold for cooling cost: When the cooling cost of rack j is greater than or equal to the upper limit threshold of cooling cost, rack j will be added to the task landing point avoidance set. When the cooling cost of rack j is less than the upper limit threshold of cooling cost, rack j is retained in the available landing point candidate set.

[0038] It should be explained that when the task scheduling system allocates tasks, it records the heat dissipation intensity of each task. In this example, it is used to read the racks to be scheduled for each task in the current task queue. The upper limit threshold of cooling cost is based on the rated cooling power of the data center cooling system and the current available cooling capacity. Eighty percent of the current cooling capacity is taken as the upper limit of the cost to trigger migration.

[0039] Record the cooling cost of each rack and generate a predicted cooling cost record for each rack. For tasks in the task queue whose heat dissipation intensity is greater than or equal to the preset heat dissipation intensity threshold, they are identified as heat dissipation tasks. Their original landing point is moved from the rack in the task landing point avoidance set to the rack with the minimum cooling cost in the available landing point candidate set. This completes the task landing point pre-adjustment and generates a cooling window period scheduling scheme. For tasks with heat dissipation intensity less than the preset heat dissipation intensity threshold, the original landing point remains unchanged, and no migration operation is performed.

[0040] It should be noted that the preset heat dissipation intensity threshold is determined based on the median value of the rack's rated heat dissipation capacity; the rack with the lowest cooling cost in the available landing point candidate set is determined by taking the first rack after sorting the candidate set in ascending order of cooling cost.

[0041] In step S4, after the cooling release window ends, the actual cooling cost data of each rack is collected and compared with the predicted cooling cost record of each rack. The edge weights that exceed the deviation in the current version of the thermal interference map are corrected to achieve closed-loop update of the map. At the end of the cooling release window, the actual cooling cost data of each rack during this window period is collected by the cooling cost interface of the energy management system database. The actual cooling cost data is the actual electrical power consumed by the cooling equipment in the corresponding rack service area, reflecting the actual cooling burden borne by the rack during this cooling release window.

[0042] For each directed edge in the thermal interference map, the actual cooling cost data of the rack corresponding to the directed edge and the corresponding rack cooling cost are taken, and the weighted deviation rate is calculated. The calculation formula is as follows: ;in, Let i be the weight deviation rate from j. This refers to the actual cooling cost data for rack j. The cost of rack cooling for rack j; The absolute value of the weight deviation rate of each side is compared with the preset deviation trigger threshold: when the absolute value of the weight deviation rate is greater than or equal to the preset deviation trigger threshold, it is determined that the current weight of that side deviates significantly from the actual value, and weight correction is performed. The correction formula is as follows: ;in, The corrected energy consumption correlation weights, The energy consumption correlation weights before correction. To correct the step size coefficient, the value range is (0,1); the corrected weights are written into the corresponding edges of the current version of the thermal interference map; When the absolute value of the weight deviation rate is less than the preset deviation trigger threshold, the current weight error is determined to be within an acceptable range, and the original weight remains unchanged.

[0043] After completing the above corrections to all edges in the current version of the thermal interference map, the updated map will be stored in the energy management system to generate a corrected thermal interference diffusion map for scheduling decisions before the next cold storage device release window opens.

[0044] It should be noted that the preset deviation trigger threshold is determined by taking the upper quartile value of the historical deviation rate distribution based on the measurement error range and historical prediction error statistics of the computer room cooling system; the larger the value of the correction step size coefficient λ, the more sensitive the correction response to single observation data, and the smaller the value, the more conservative and stable the correction process. It can be set in the range of 0.1 to 0.5 according to the historical convergence speed of the spectrum; the acquisition time window of the actual cooling power consumption is triggered when the remaining cooling capacity of the cold storage device drops to less than five percent of the rated total cooling capacity, so as to ensure that the acquisition range covers the complete cooling release process.

[0045] By collecting actual cooling costs and correcting the map weights as described above, the thermal interference diffusion map is continuously and adaptively updated. This allows the map weights to gradually approach the actual energy consumption correlation with the real observation data of each cooling release window, thereby improving the prediction accuracy of subsequent scheduling decisions.

[0046] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0047] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0049] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0050] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data center energy-saving scheduling method based on energy consumption correlation graphs, characterized in that: Includes the following steps: Step S1: Read the remaining cooling capacity data of the cold storage device, combine the current air inlet temperature of each rack with the airflow velocity in the rack to generate the overcooling risk index of each rack, and determine whether to mark the current rack as overcooling risk based on the overcooling risk index. Step S2: Retrieve the server fan speed baseline, combine the overcooling risk index to evaluate the response sensitivity of each overcooling risk rack during the cooling release window, obtain the spatial position relationship data of each rack, take the overcooling risk rack as the source node and the spatial position relationship as the edge, and calculate the energy consumption correlation weight of each adjacent rack in combination with the response sensitivity to construct a thermal interference diffusion map. Step S3: Collect the heat dissipation intensity of each heat dissipation task in the current task queue, calculate the cooling cost of each rack undertaking the heat dissipation task during the cooling release window, generate a task landing point avoidance set, filter heat dissipation tasks based on the task landing point avoidance set, and complete the pre-adjustment of task landing points during the cooling release window. Step S4: After the cooling window ends, collect the actual cooling cost data of each rack, compare it with the cooling cost in the thermal interference diffusion map before the task landing point is pre-adjusted, generate the weight deviation rate, and correct and update the weight of the corresponding edge in the thermal interference diffusion map.

2. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 1, characterized in that: In step S1, before the cold storage device opens its cold release window, the remaining cold energy data of the cold storage device's sensor is read. The remaining cold energy data is the total amount of cold energy that the cold storage device has not yet released. Collect the inlet air temperature of each rack temperature sensor. The inlet air temperature refers to the current air temperature on the air inlet side of the rack. Collect the airflow velocity inside the rack of each rack wind speed sensor. The airflow velocity inside the rack is the average velocity of the airflow inside the rack. After normalizing the remaining cooling capacity, the inlet air temperature of each rack, and the airflow velocity in each rack, the overcooling risk index of the rack is obtained by weighted summation.

3. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 1, characterized in that: In step S1, the overcooling risk index of each rack is compared with the preset overcooling risk judgment threshold: When the overcooling risk index is greater than or equal to the overcooling risk judgment threshold, the rack will be marked as an overcooling risk rack; When the overcooling risk index is less than the overcooling risk judgment threshold, the rack is marked as a normal rack.

4. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 1, characterized in that: In step S2, the BMC equipment speed baseline of the server corresponding to each rack with excessive cooling risk is retrieved. The BMC equipment speed baseline is the rated speed reference value of the server fan in each rack under normal temperature environment. The fan speed reduction ratio is calculated based on the BMC equipment speed baseline, and the response sensitivity of the overcooled risk rack is calculated by combining the overcooled risk index of the overcooled risk rack.

5. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 4, characterized in that: In step S2, the spatial location relationship data of the data center topology database is accessed. The spatial location relationship data includes the three-dimensional spatial coordinates of each rack. Each rack at risk of excessive cooling is taken as a source node. For each source node, other racks in the data center are traversed, and the spatial distance between the source node and the target rack is calculated using three-dimensional spatial coordinates. When the spatial distance is not greater than the preset spatial influence radius, a directed edge is established from the source node to the target rack, and the energy consumption association weight of the directed edge is calculated. When the spatial distance is greater than the preset spatial influence radius, no directed edge is established; All directed edges and their weights are aggregated to form a thermal interference diffusion map.

6. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 1, characterized in that: In step S3, the heat dissipation intensity of each heat dissipation task in the current task queue of the task scheduling system is read. The heat dissipation intensity is the heat generated by the heat dissipation task per unit time when it is running at full load. For each rack to be scheduled in the task queue, the energy consumption correlation weight of the corresponding directed edge is extracted from the thermal interference diffusion map. The energy consumption correlation weights are summed to obtain the thermal interference superposition intensity of the rack to be scheduled, and the cooling cost when the task landing point is the rack to be scheduled is calculated. Compare the cooling cost of each rack with the preset upper limit threshold for cooling cost: When the cooling cost of the rack to be scheduled is greater than or equal to the upper limit threshold of the cooling cost, the rack to be scheduled will be added to the task landing point avoidance set. When the cooling cost of the rack to be scheduled is less than the upper limit threshold of cooling cost, the rack to be scheduled will be retained in the available landing point candidate set.

7. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 6, characterized in that: In step S3, tasks with heat dissipation intensity greater than or equal to a preset heat dissipation intensity threshold in the task queue are identified as heat dissipation tasks. Their original landing point is moved from the rack in the task landing point avoidance set to the rack with the minimum cooling cost in the available landing point candidate set. This completes the pre-adjustment of the task landing point and generates a cooling window scheduling scheme. For tasks with a heat dissipation power value less than the preset heat dissipation intensity threshold, the original landing point will remain unchanged, and no migration operation will be performed.

8. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 1, characterized in that: In step S4, at the end of the cooling release window, the actual cooling cost data of each rack during this window period is collected by the cooling cost interface of the energy management system database. The actual cooling cost data is the actual electrical power consumed by the cooling equipment in the corresponding rack service area; For each directed edge in the thermal interference map, the actual cooling cost data of the rack corresponding to the directed edge and the rack cooling cost are taken, and the weighted deviation rate is calculated.

9. The data center energy-saving scheduling method based on energy consumption correlation graph as described in claim 8, characterized in that: In step S4, when the absolute value of the weight deviation rate is greater than or equal to the preset deviation trigger threshold, weight correction is performed, and the corrected energy consumption correlation weight is calculated by combining the weight deviation rate and the energy consumption correlation weight. When the absolute value of the weight deviation rate is less than the preset deviation trigger threshold, the current weight error is determined to be within an acceptable range, and the energy consumption associated weight remains unchanged. After correcting all edges in the current version of the thermal interference map, the updated map is stored in the energy management system to generate the corrected thermal interference diffusion map.

Images