A method for dynamically adjusting data center cooling power based on server load changes

Abstract

This invention relates to the field of power regulation technology, specifically to a method for dynamically adjusting the cooling power of a data center based on server load changes. The method includes: S1: collecting server load change data, server fan speed change data, rack exhaust temperature data, and cold aisle intake temperature data to generate a true load heat release marker; S2: jointly determining the persistence of the exhaust temperature of the corresponding rack, changes in cold aisle intake recirculation, and disturbances to the intake air of adjacent racks; identifying the delay transmission chain and generating the original cold aisle thermal boundary intrusion result; S3: performing directional air supply compensation on the rack area where thermal boundary intrusion occurs and adjusting the air supply priority; if the cold aisle thermal boundary intrusion result continues to expand after directional air supply compensation, increasing the cooling output power of the corresponding cooling equipment, and generating a dynamic adjustment result for the data center cooling power. This invention can reduce unnecessary increases in cooling power and lower the overall energy consumption of the data center.

Description

Technical Field

[0001] This invention relates to the field of power regulation technology, and in particular to a method for dynamically adjusting the cooling power of a data center based on changes in server load. Background Technology

[0002] With the rapid development of cloud computing, big data, artificial intelligence training, and high-performance computing, the deployment density of servers in data center server rooms is constantly increasing, and the power of a single rack is continuously growing. Server room cooling systems have become an important infrastructure to ensure the stable operation of servers. Currently, server room cooling control is usually adjusted based on server load information, rack temperature information, or cold aisle temperature information. When an increase in load or a temperature exceeding the set range is detected, cooling is achieved by increasing the air conditioning air volume, decreasing the air supply temperature, or increasing the output power of the cooling equipment. However, there are significant time lags and transmission differences between server load changes, server fan speed adjustment processes, heat accumulation processes inside racks, and cold aisle temperature response processes. Especially in high-density server room, a sudden increase in server load often triggers a rapid increase in fan speed first, while the exhaust air temperature may be affected by the airflow dilution effect in a short period of time and cannot accurately reflect the heat accumulation situation. This makes it difficult for existing technologies to identify the real heat load change process in a timely and accurate manner.

[0003] Furthermore, most existing data center cooling control solutions rely on the temperature of a single rack or a localized area as the basis for adjustment, assuming that temperature anomalies originate from changes in the rack's own load. They lack the ability to identify the heat propagation process within the cold aisle. When some racks generate significant heat and exceed their own heat exchange boundaries, the high-temperature exhaust airflow can gradually affect subsequent racks along the cold aisle, leading to cross-rack heat transfer and induced temperature rises. This can cause abnormally high intake air temperatures in subsequent racks even when their own loads have not changed significantly. Because current technology cannot identify the heat propagation path, the location of the inducing source, and the affected area, it often resorts to increasing the overall cooling capacity. This not only easily leads to localized overcooling and energy waste but also makes it difficult to precisely suppress heat diffusion at its source. Summary of the Invention

[0004] This invention provides a method for dynamically adjusting the cooling power of a data center based on changes in server load. It can identify the actual heat release process, track the heat propagation path across server racks, and adjust cooling resources in a targeted and precise manner based on changes in server load.

[0005] A method for dynamically adjusting data center cooling power based on server load changes includes the following steps:

[0006] S1: Collect server load change data, server fan speed change data, rack exhaust temperature data, and cold aisle intake temperature data; identify the rack exhaust dilution state caused by server fan adaptive speed increase after server load increases; and generate a load heat release marker. The load heat release marker is used to distinguish between the actual heat release process caused by server load increase and the short-term exhaust disturbance process caused by server fan speed increase.

[0007] S2: Using the actual heat release marker as input, jointly determine the continuity of exhaust temperature of the corresponding cabinet, the change of cold aisle airflow return, and the disturbance of airflow to adjacent cabinets to identify whether the actual heat release process has broken through the local heat exchange boundary of the cabinet; track the time sequence of airflow temperature response of each cabinet in the same cold aisle, identify the delayed heat transfer chain in the cold aisle, and determine whether there is a cross-cabinet induced heating phenomenon induced by a locally high-load cabinet, generating the original cold aisle thermal boundary intrusion result; the cross-cabinet induced heating phenomenon is that the disturbed hot airflow discharged from the front-end cabinet does not directly cause its own temperature runaway, but causes the back-end cabinet to show an abnormal increase in airflow temperature in advance;

[0008] S3: Using the original cold aisle thermal boundary intrusion result as input, first perform directional air supply compensation on the cabinet area where thermal boundary intrusion occurs, and adjust the air supply priority of the corresponding cabinet for the cold aisle area where cross-cabinet induced heating phenomenon exists; if the cold aisle thermal boundary intrusion result is still detected to continue to expand after the directional air supply compensation, increase the cooling output power of the corresponding cooling equipment to generate the dynamic adjustment result of the data center cooling power.

[0009] Optionally, S1 specifically includes:

[0010] S11: Temperature sensors are deployed on the internal exhaust side and the front cold aisle side of each server rack to continuously collect the exhaust temperature sequence and cold aisle inlet temperature sequence of each server rack in the first sampling period; the load occupancy sequence and fan speed sequence of each server are obtained in real time through the server's baseboard management controller.

[0011] S12: Compare the load change rate of two adjacent sampling points in the load occupancy rate sequence with a preset load increase threshold. When the load change rate exceeds the load increase threshold, mark the corresponding server as entering the load increase candidate state, and record the fan speed sequence of the server within a preset short window after the start time of the load increase candidate state.

[0012] S13: Compare the fan acceleration value with a preset fan adaptive acceleration threshold. If the fan acceleration value exceeds the fan adaptive acceleration threshold, it is determined that there is a cabinet exhaust dilution state caused by fan adaptive speed increase on the exhaust side of the server rack.

[0013] S14: After determining that the cabinet exhaust dilution state exists, extract the actual temperature change rate of the exhaust temperature sequence of the server cabinet within the preset short-time window, and simultaneously extract the reference temperature change rate of the cold aisle inlet air temperature sequence of the server cabinet within the same window. Compare the actual temperature change rate with the reference temperature change rate. If the actual temperature change rate is less than the sum of the reference temperature change rate and the preset dilution offset, generate a first intermediate marker; otherwise, generate a second intermediate marker.

[0014] S15: Use the first intermediate mark as the actual release mark of the load heat, and the second intermediate mark as the short-term exhaust disturbance mark.

[0015] Optionally, the exhaust dilution state includes the physical phenomenon where, after the server fan speeds up, a large amount of cold air is forcibly drawn through the inside of the server and discharged at high speed from the exhaust side, resulting in an increase in the local airflow velocity on the exhaust side and a decrease or a smaller increase in the exhaust temperature in a short period of time.

[0016] Optionally, the load heat release marker indicates that the heat generated by the current increase in server load has not been fully released to the exhaust side and is still accumulated in the form of heat inside the server or in a local area of ​​the rack; the short-term exhaust disturbance marker indicates that the current exhaust temperature change is mainly dominated by the airflow dilution effect caused by the fan speed increase.

[0017] Optionally, S2 specifically includes:

[0018] S21: Using the load heat release mark as a trigger signal, for the corresponding cabinet marked as having a real heat release process, extract the exhaust temperature change persistence parameter of the cabinet within the continuous monitoring window after the load heat release mark is generated. The exhaust temperature change persistence parameter includes the continuous rise slope of the exhaust temperature and the duration of the continuous rise slope. When the continuous rise slope is greater than a preset temperature rise slope threshold and the duration of ...

[0019] S22: Simultaneously extract the intake air temperature sequence of the cold aisle where the corresponding cabinet is located within the continuous monitoring window, including the positive deviation of the cold aisle intake air temperature from the reference intake air temperature of the cold aisle within the continuous monitoring window and the fluctuation amplitude of the positive deviation. When the positive deviation is greater than a preset recirculation temperature threshold and the fluctuation amplitude is less than a preset stable fluctuation threshold, it is determined that the corresponding cold aisle has intake air recirculation change characteristics caused by cabinet exhaust recirculation.

[0020] S23: Extract the intake air temperature sequence of adjacent cabinets within the continuous monitoring window as intake air disturbance parameters. The intake air disturbance parameters include the rate of increase of the intake air temperature of the adjacent cabinet and the delay time relative to the rise of the exhaust air temperature of the corresponding cabinet. When the rate of increase of the intake air temperature of the adjacent cabinet exceeds a preset disturbance rate threshold and the delay time is within a preset disturbance delay interval, it is determined that the adjacent cabinet is experiencing intake air disturbance.

[0021] Optionally, S2 further includes:

[0022] The characteristics of continuous exhaust temperature, changes in air inlet recirculation, and disturbances to air inlet in adjacent cabinets are jointly determined. The joint determination includes determining that if the continuous temperature rise, changes in air inlet recirculation, and disturbances to air inlet in at least one adjacent cabinet are simultaneously satisfied, the actual heat release process has broken through the local heat exchange boundary of the cabinet, and a first boundary intrusion mark is generated; otherwise, a boundary-not-broken mark is generated.

[0023] After the first boundary intrusion marker is generated, the air intake temperature response time sequence of each rack arranged sequentially along the airflow direction in the same cold aisle is tracked. The response time when the air intake temperature of each rack first exceeds the preset abnormal air intake temperature threshold is recorded in order from the air intake end of the cold aisle to the far end. The response times of each rack are arranged in chronological order to form a delayed heat transfer chain in the cold aisle. When the difference in response time between adjacent racks in the delayed transfer chain is greater than the preset minimum transfer delay interval, it is determined that there is an effective heat delayed transfer path. Causal analysis is performed on the delayed transfer chain.

[0024] Optionally, the causal analysis specifically includes:

[0025] Extract the exhaust temperature change history of the first responding cabinet in the delayed transmission chain. If the first responding cabinet does not show a real release of its own load heat within the preset time window before its own response time, but there is a record that the front cabinet has broken through the local heat exchange boundary, then it is determined that the abnormal intake temperature of the first responding cabinet is caused by the disturbance of hot air flow discharged from the front cabinet through the cold aisle, and it is marked as cross-cabinet induced heating phenomenon.

[0026] All abnormal air intake temperatures of subsequent response cabinets in the delayed propagation chain are attributed to the same induction source, generating the original cold aisle thermal boundary intrusion result, including the induction source cabinet identifier, the induction propagation direction, and the induced cabinet set.

[0027] Optionally, S3 specifically includes:

[0028] S31: Analyze the original cold aisle thermal boundary intrusion results, extract the induction source cabinet identifier, the induction propagation direction, and the set of cabinets affected by the induction; mark the cabinets corresponding to the induction source cabinet identifier and all cabinets in the set of cabinets affected by the induction as the set of cabinets to be compensated, and use the induction propagation direction as the preferred direction of directional air supply;

[0029] S32: For each cabinet in the set of cabinets to be compensated, generate directional air supply compensation parameters for the corresponding cabinet based on the cabinet's position coordinates in the cold aisle and the response order in the delay transmission chain; the directional air supply compensation parameters include the deflection angle of the air supply nozzle toward the air inlet of the cabinet, the local air supply velocity increment for the cabinet, and the air supply duration.

[0030] S33: The generated directional air supply compensation parameters for each cabinet are sent to the variable air supply unit actuator above the corresponding cold aisle. The actuator adjusts the nozzle direction according to the deflection angle and increases the air supply volume according to the local air supply velocity increment. During the air supply duration, the first round of directional air supply compensation is performed on the cabinet set to be compensated.

[0031] S34: While performing the first round of directional air supply compensation, for cold aisle areas where cross-rack induced heating occurs, the air supply priority of the corresponding rack is dynamically adjusted according to the response order of each rack in the delay transmission chain.

[0032] S35: In the first evaluation window after the first round of directional air supply compensation is completed, the exhaust air temperature data and cold aisle inlet air temperature data of each cabinet in the set of cabinets to be compensated are collected again, and the updated cold aisle thermal boundary intrusion results are regenerated; the updated cold aisle thermal boundary intrusion results are compared with the original cold aisle thermal boundary intrusion results to determine whether the thermal boundary intrusion range continues to expand.

[0033] S36: If the thermal boundary intrusion range does not continue to expand or shrinks, maintain the current directional air supply compensation parameters and gradually reduce the air supply velocity increment to zero according to the preset attenuation step, ending this adjustment; if the thermal boundary intrusion range continues to expand, based on the first round of directional air supply compensation, increase the fan speed or chilled water valve opening of the corresponding refrigeration equipment according to the preset power increase step, generate a refrigeration output power increase command, and at the same time keep the air supply priority unchanged, and execute the second round of joint adjustment.

[0034] Optionally, the airflow velocity increment of the inducing source cabinet is greater than the airflow velocity increment of any induced cabinet, and the airflow velocity increment of the cabinets further back in the induction propagation direction decreases step by step; the airflow priority includes the weight of airflow resource allocation for each cabinet when the total airflow capacity of the variable airflow unit above the cold aisle is limited; the inducing source cabinet is assigned the highest priority weight, and the cabinets in the induced cabinets that respond earlier are assigned a higher priority weight, and the airflow priority decreases step by step according to the induction propagation direction.

[0035] Optionally, S3 further includes:

[0036] Within the second evaluation window following the completion of the second round of joint regulation, the cold aisle thermal boundary intrusion results are regenerated;

[0037] If the thermal boundary intrusion range stops expanding, the current cooling output power is maintained until the exhaust temperature of the cabinet assembly to be compensated returns to the normal range, after which it is gradually reduced.

[0038] If the thermal boundary intrusion range continues to expand, repeat step S36 to increase the cooling output power again until the thermal boundary intrusion range stops expanding or the cooling output power reaches the preset maximum safe power threshold. Finally, the dynamic adjustment result of the computer room cooling power is generated. The dynamic adjustment result of the computer room cooling power includes the sequence of directional air supply compensation parameters executed in each round and the record of cooling output power changes.

[0039] The beneficial effects of this invention are:

[0040] This invention generates a true heat release marker by identifying fan speed-up behavior after load increases, exhaust dilution status, and the offset relationship between exhaust temperature and cold aisle inlet temperature, thereby decoupling fan airflow disturbance from the actual heat release process. By combining the continuous characteristics of exhaust temperature, the characteristics of inlet air recirculation changes, and the disturbance of air intake in adjacent cabinets, it identifies whether heat has exceeded the local heat exchange boundary of the cabinet, enabling early perception of the initial stage of heat diffusion. It can identify potential thermal risks before heat has formed a large-scale temperature rise, improving the accuracy of heat load identification and the foresight of cooling control.

[0041] This invention records the response time when the inlet air temperature of each rack first exceeds the abnormal inlet air temperature threshold, constructs a delayed heat transfer chain in the cold aisle, and performs causal analysis in conjunction with the actual heat release markers of the load. This accurately identifies the cross-rack induced temperature rise phenomenon caused by heat propagation from the preceding rack, and generates the original cold aisle thermal boundary intrusion result, including the identifier of the inducing source rack, the direction of induced propagation, and the set of racks affected by the intrusion. This invention performs directional air supply compensation for the inducing source rack and the set of racks affected by the intrusion, and dynamically allocates air supply priority according to the delayed transfer chain. This ensures that the cooling capacity is preferentially applied to the heat diffusion source and key nodes of the propagation path. Only when the thermal boundary intrusion range continues to expand does the output power of the cooling equipment gradually increase. This not only effectively blocks heat diffusion and improves the temperature control accuracy of the computer room, but also reduces unnecessary cooling power increases, lowers the overall energy consumption of the computer room, and improves the operating efficiency of the cooling system. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram illustrating the heat release determination in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0046] like Figures 1-2 As shown, a method for dynamically adjusting the cooling power of a data center based on changes in server load includes the following steps:

[0047] S1: Collect server load change data, server fan speed change data, rack exhaust temperature data, and cold aisle intake temperature data; identify the rack exhaust dilution state caused by server fan adaptive speed increase after server load increases; and generate a load heat release marker. The load heat release marker is used to distinguish between the actual heat release process caused by server load increase and the short-term exhaust disturbance process caused by server fan speed increase.

[0048] S1 specifically includes:

[0049] S11: Temperature sensors are deployed on the internal exhaust side and the corresponding front cold aisle side of each server rack to collect real-time data on the changes in exhaust temperature and cold aisle intake temperature during server rack operation. The temperature sensors are preferably digital temperature sensors or thermistor temperature sensors, and are synchronously sampled according to a unified first sampling period to ensure the time consistency of temperature data between different racks.

[0050] The first sampling period is set to 1s-5s. After a sudden increase in server load, the server fan usually completes the first round of adaptive speed increase within 1s-3s. If the sampling period is too long, it will be difficult to capture the correspondence between the rapid change of fan speed and the short-term dilution distortion of exhaust temperature in time, which will easily lead to a lag in the identification of the rack exhaust dilution state. Therefore, the first sampling period needs to be less than the typical response period of the server fan control loop. The change of rack exhaust temperature has a certain thermal inertia compared with the internal chip temperature of the server. It usually gradually forms a continuous change trend within a few seconds. If the sampling period is too short, for example, less than 0.5s, the temperature sensor is easily affected by local airflow pulsation, fan blade disturbance and transient turbulence, which will cause high-frequency fluctuations in the exhaust temperature sequence, thereby reducing the stability of the judgment of the actual release of load heat. The cold aisle inlet temperature is a slowly changing temperature field after spatial diffusion. Its effective change is usually not faster than 1s. Therefore, using a first sampling period of 1s-5s can ensure sufficient time synchronization accuracy between the cold aisle inlet temperature sequence and the exhaust temperature sequence, while avoiding excessive data collection and increasing the processing burden on the edge controller.

[0051] Meanwhile, the server's baseboard management controller obtains the operating status information of each server in real time. The operating status information includes at least the server load utilization sequence and the server fan speed sequence. The server load utilization can be CPU utilization, GPU utilization, memory utilization, or overall resource utilization.

[0052] set up Taiwan server at all times The load utilization rate is The corresponding fan speed is The exhaust temperature of the server rack is [temperature value missing]. The corresponding cold aisle inlet side temperature is ;

[0053] in, Indicates the first Taiwan server at any time Under load utilization, Indicates the first Taiwan server at any time The fan speed is reduced. This indicates the temperature on the exhaust side of the corresponding server rack. This indicates the temperature on the inlet side of the corresponding cold aisle.

[0054] S12: The system performs differential calculations on adjacent sampling points in the load occupancy rate sequence to obtain the server load change rate; assuming... The load occupancy rates of the servers at the current sampling time and the previous sampling time are respectively and The load change rate corresponding to this server is expressed as:

[0055] ;

[0056] in, Indicates the first The load change rate of the server. Indicates the time interval between adjacent sampling times;

[0057] Differential calculation continuously reads the server's load utilization data at two adjacent sampling times according to the first sampling period. By calculating the change in load utilization between the later and earlier sampling times and combining it with the time interval between the two sampling times, the rate of change of server load per unit time is obtained, thus representing the upward or downward trend of the server's current load. When the load change between adjacent sampling points is large, it indicates that the server has experienced a significant load surge in a short period of time; when the load change between adjacent sampling points is small, it indicates that the server load is in a relatively stable state. Through this differential calculation method, the transient change process of server load can be quickly identified, providing a basis for subsequent identification of candidate states of increased load and rack exhaust dilution status.

[0058] Load change rate Compared with the preset load increase threshold Compare, when satisfied When this happens, the server is determined to be a candidate for increased load.

[0059] in, The load increase threshold is used to determine whether the server has entered a phase of rapid load increase. Its value is set to 8% / s-20% / s of the server's rated load. Server load changes caused by fluctuations in normal business requests, background cache refreshes, and regular process switching are usually relatively gradual, with a load change rate generally below 5% / s per unit time. If the load increase threshold is set too low, the system may easily misidentify a large number of normal business fluctuations as candidate states for load increase, leading to frequent triggering of rack exhaust dilution state judgments. When the server executes GPU training tasks, batch data calculations, virtual machine migrations, or high-concurrency storage access, the server load usually increases rapidly within seconds. At this time, the load change rate is usually over 10% / s, accompanied by the server fan control system entering a rapid acceleration phase. Therefore, setting the load increase threshold within the range of 8% / s-20% / s can effectively distinguish between the actual rapid increase in heat load and the normal business fluctuation process. The load increase threshold also needs to match the response characteristics of the server fan control strategy. If the load increase threshold is too high, the system may only identify the candidate state for load increase after the fans have already adaptively accelerated, thus missing the optimal identification window for rack exhaust dilution state.

[0060] After entering the load increase candidate state, record the fan speed changes of the corresponding server within a preset short window; let the fan speed sequence be... The corresponding fan acceleration value is expressed as:

[0061] ;

[0062] in, This indicates the server fan acceleration value. This indicates the fan speed at the end of the window. This indicates the fan speed at the start of the window.

[0063] S13: Set fan acceleration value With preset fan adaptive acceleration threshold Compare, when satisfied At that time, it was determined that the server rack where the server was located was in a rack exhaust dilution state.

[0064] in, The fan adaptive acceleration threshold is used to determine whether the server fan has entered a rapid adaptive speed-up phase triggered by a sudden load surge. Its value is between 800 RPM / s and 2500 RPM / s. During normal server operation, the fan control system dynamically adjusts the fan speed based on CPU, GPU, or motherboard temperatures. This adjustment is typically a gradual speed correction process, with a fan speed change rate generally below 500 RPM / s. If the fan adaptive acceleration threshold is set too low, normal cooling speed adjustments may be mistakenly identified as a rack exhaust dilution state. When the server experiences high-load computing tasks, concentrated GPU computing tasks, or sudden storage read / write tasks, the internal heat source of the server will rapidly increase in a short period. The fan control system will typically enter a rapid response mode, causing the fan speed to increase rapidly within 1-3 seconds to enhance air heat exchange. At this time, the fan acceleration value... Typically exceeding 1000 RPM / s, setting the fan adaptive acceleration threshold within the range of 800 RPM / s to 2500 RPM / s effectively distinguishes between normal fan speed regulation and the fan adaptive speed increase induced by a sudden load surge. The fan adaptive acceleration threshold also needs to match the airflow inertial characteristics inside the server rack. When the fan speed increases rapidly in a short period, the airflow velocity inside the server increases rapidly, and hot air is exhausted at high speed before it has time to accumulate sufficiently, which can easily lead to exhaust temperature distortion. In summary, for high-density GPU servers, due to their higher rated fan speeds and more drastic changes in thermal power, the fan adaptive acceleration threshold is set at 1800 RPM / s to 2500 RPM / s; for ordinary CPU servers, the fan adaptive acceleration threshold is set at 800 RPM / s to 1500 RPM / s.

[0065] Rack exhaust dilution state refers to the situation where, due to a rapid increase in server load, the fans trigger adaptive speed increase, resulting in a large amount of cool air being rapidly drawn through the server's interior and quickly expelled. This leads to a significant increase in the local airflow velocity on the rack exhaust side, causing the exhaust temperature to temporarily decrease or even decrease in the rate of increase. A significant increase is defined as an increase of more than 20% in the local airflow velocity on the rack exhaust side relative to the average exhaust airflow velocity during the stable load phase. When the local airflow velocity increase is less than 20%, the disturbance to the exhaust temperature field from the increased airflow is relatively weak, and the exhaust temperature can still accurately reflect the heat changes inside the server. Normally, no significant exhaust dilution effect is formed; however, when the local airflow velocity increases by more than 20%, the high-speed air will significantly shorten the residence time of hot air inside the server, causing local heat to be quickly carried away from the server before it can accumulate sufficiently. This results in the exhaust temperature decreasing instead of increasing, delayed heating, or a significantly reduced heating rate. At this point, a significant deviation begins to appear between the exhaust temperature and the actual heat load of the server. When the local airflow velocity increases to 30%-60%, the rack exhaust dilution state is most obvious, which can easily cause the cooling system to mistakenly believe that the current heat load has not increased, thus delaying the timing of cooling power adjustment.

[0066] In this state, the exhaust temperature cannot accurately reflect the degree of heat accumulation inside the server, which can easily cause the cooling system to misjudge changes in the server's heat load.

[0067] S14: After determining that there is a cabinet exhaust dilution state, further extract the exhaust temperature change of the corresponding server cabinet within a preset short time window;

[0068] The preset short window is set to 3s-15s. The preset short window needs to cover the main response phase of the fan adaptive speed increase after the server load increases. Setting the preset short window to 3s-15s can simultaneously cover the three continuous processes of fan speed increase, airflow dilution, and exhaust temperature distortion. For high-density GPU server rooms, the load surge and fan response are faster, so the preset short window is 3s-6s. For ordinary CPU server rooms, the thermal response is relatively slow, so the preset short window is 6s-10s. For server rooms with complex rack airflow organization or slow sensor response, the preset short window can be appropriately extended to 10s-15s.

[0069] Let the exhaust temperature sequence be... The actual rate of change of temperature on the exhaust side is expressed as:

[0070] ;

[0071] Simultaneously extract the corresponding cold aisle inlet air temperature sequence. The rate of change of the cold aisle reference temperature is obtained and expressed as:

[0072] ;

[0073] in, This indicates the actual rate of temperature change on the exhaust side. Indicates the rate of change of the cold aisle reference temperature;

[0074] The actual temperature change rate Rate of change with reference temperature Compare, when satisfied If the condition is met, generate the first intermediate tag; otherwise, generate the second intermediate tag.

[0075] in, This represents a preset dilution offset, used to characterize the degree of deviation between the actual temperature change rate on the exhaust side caused by the dilution state of the server fan's adaptive speed increase and the allowable temperature change rate of the cold aisle reference. It establishes a boundary between the airflow dilution effect and the actual heat release effect. The value range is 0.03℃ / s-0.10℃ / s. During stable server load operation, the rack exhaust temperature and the cold aisle inlet temperature typically maintain a relatively stable relationship, with the difference in their temperature change rates generally small, not exceeding 0.02℃ / s. When the fan speed has not yet increased, the exhaust side temperature change rate can... This more accurately reflects the server's heat release process. When the server enters a load increase candidate state, if the fan rapidly accelerates, the high-speed airflow will speed up the passage of cold air through the server's interior, causing hot air on the exhaust side to be prematurely carried away from the heat exchange area. This results in the exhaust temperature change rate being lower than the rate of temperature increase that should occur according to the actual heat load growth. Numerous data center thermal management tests show that during the rapid fan speed-up phase, the exhaust temperature change rate typically exhibits a decrease of 0.03℃ / s to 0.10℃ / s relative to the normal thermal response state. Therefore, this range can effectively reflect the typical impact range of the exhaust dilution effect. If the setting is too small, such as less than 0.03℃ / s, then fluctuations in ambient temperature, sensor measurement errors, and local airflow disturbances may all cause the system to frequently generate the first intermediate marker, thereby increasing the probability of misjudgment; if If the setting is too high, such as exceeding 0.10℃ / s, only extremely strong exhaust dilution phenomena can be identified, which may easily miss some real cabinet exhaust dilution states, thereby reducing the identification sensitivity of the load heat release marker.

[0076] S15: When the system generates the first intermediate marker, use it as the marker for the actual release of load heat.

[0077] The load heat release indicator is used to characterize the fact that the heat generated after the server load increases has not been fully transferred to the exhaust side, but is still accumulated in the form of heat in the server heat sink area, motherboard area, power supply area, or local airflow stagnation area of ​​the rack.

[0078] When the system generates a second intermediate marker, it is used as a short-term exhaust disturbance marker.

[0079] The short-term exhaust disturbance marker indicates that the current exhaust temperature change is mainly affected by the dilution effect of the high-speed airflow generated by the server fan speed increase, rather than being dominated by the actual increase in server heat load. By distinguishing between the actual load heat release marker and the short-term exhaust disturbance marker, the cooling system can avoid misinterpreting a momentary fan speed increase as a real heat load surge, thereby reducing the risk of incorrect cooling power adjustment.

[0080] S2: Using the actual heat release marker as input, it jointly judges the continuity of exhaust temperature of the corresponding cabinet, the change of cold aisle airflow recirculation, and the disturbance of airflow to adjacent cabinets to identify whether the actual heat release process has broken through the local heat exchange boundary of the cabinet; it tracks the time sequence of airflow temperature response of each cabinet in the same cold aisle, identifies the delayed heat transfer chain in the cold aisle, and judges whether there is cross-cabinet induced heating phenomenon induced by local high-load cabinets, generating the original cold aisle thermal boundary intrusion result; cross-cabinet induced heating phenomenon is when the disturbed hot airflow discharged from the front-end cabinet does not directly cause its own temperature runaway, but causes the back-end cabinet to show abnormal increase in airflow temperature in advance.

[0081] S2 specifically includes:

[0082] S21: Using the generated load heat release marker as a trigger signal, for the corresponding cabinet marked as having a real heat release process, extract the exhaust temperature sequence of that cabinet at the time the load heat release marker is generated. Subsequent continuous monitoring window The parameter indicating the duration of temperature variation in the exhaust air;

[0083] Let the exhaust temperature sequence within this time window be:

[0084] ;

[0085] The slope of the continuous rise in exhaust air temperature Duration of maintenance They are represented as follows:

[0086] ;

[0087] ;

[0088] When satisfied At that time, it was determined that the exhaust temperature of the cabinet had a continuous temperature rise characteristic;

[0089] in, This marks the end of the continuous rise in exhaust air temperature. The slope of the continuous rise in exhaust air temperature. The duration for which the exhaust air temperature rises. The preset temperature rise slope threshold is used to determine whether the rack exhaust temperature has entered the continuous heat release phase. Essentially, it distinguishes between normal temperature fluctuations and a continuous temperature rise process caused by actual heat accumulation. The value range is 0.03℃ / s-0.12℃ / s. Under normal data center operation, even with slight fluctuations in server load, the rack exhaust temperature will be affected by air conditioning fluctuations, sensor measurement errors, and local airflow disturbances, resulting in small random changes. The rate of change of exhaust temperature corresponding to such fluctuations is usually no more than 0.02℃ / s. When the server enters the actual heat release phase, the rack exhaust temperature will show a continuous upward trend due to the continuous heating of the CPU, GPU, memory, and power modules. Actual data center test results show that the rate of change of exhaust temperature corresponding to this heat accumulation process is usually between 0.04℃ / s and 0.15℃ / s. Therefore, setting the temperature rise slope threshold within the range of 0.03℃ / s-0.12℃ / s can effectively distinguish between normal temperature fluctuations, short-term disturbances caused by fan speed increases, and the actual continuous heat release process. The preset temperature rise duration threshold is used to determine whether the exhaust temperature rise is continuous. The value is 10s-60s. The server fan adaptive speed increase process is usually completed within a few seconds. The temperature rise phenomenon that occurs within 10s may still be in the airflow adjustment stage, which does not necessarily mean that the actual heat has exceeded the local heat exchange boundary of the rack. The heat inside the server needs a certain thermal inertia process to be conducted from the chip to the heat sink and then to the rack exhaust side. Setting the temperature rise duration threshold to 10s-60s can effectively distinguish between short-term thermal disturbances, fan speed adjustment disturbances and continuous heat release processes. The continuous monitoring window length is related to the thermal inertial response and ranges from 3s to 15s.

[0090] S22: Synchronously extract the inlet air temperature sequence of the corresponding cabinet's cold aisle. Calculate the inlet air temperature relative to the cold aisle reference temperature. The positive deviation is expressed as:

[0091] ;

[0092] And extract its fluctuation range , is represented as:

[0093] ;

[0094] When satisfied At that time, it was determined that the cold aisle exhibited characteristics of air intake recirculation changes caused by exhaust air recirculation from the server racks;

[0095] in, This represents the positive deviation of the cold aisle inlet air temperature. This refers to the fluctuation range of the cold aisle inlet air temperature. The preset recirculation temperature threshold is used to determine whether the cold aisle inlet air temperature deviates significantly in the positive direction due to the recirculation of cabinet exhaust air. The value range is 0.2℃-0.8℃. In data center operation experiments, local cabinet exhaust air recirculation usually causes the inlet air temperature of the same cold aisle to rise by 0.2℃-0.8℃ within a few seconds. Temperature deviations below 0.2℃ are often covered by environmental airflow disturbances, sensor errors, or fan fine-tuning. Deviations above 0.8℃ usually indicate serious recirculation or cold aisle design defects. A preset stable fluctuation threshold is used to exclude high-frequency temperature fluctuations caused by cold aisle airflow turbulence, fan fine-tuning, or sensor noise, ensuring that only continuous temperature deviations truly caused by rack exhaust are identified. The value range is 0.05℃-0.15℃. Cold aisle airflow turbulence or slight fan speed adjustment in the data center usually causes temperature fluctuations between 0.05℃ and 0.15℃. When the fluctuation amplitude is below 0.05℃, the signal is easily masked by noise; when the fluctuation amplitude is above 0.15℃, it may include transient large disturbances, which are not suitable for judging continuous backflow. The reference inlet air temperature for the cold aisle is usually the average value of the previous stable sample.

[0096] S23: Extract the inlet air temperature sequence of adjacent cabinets to the corresponding cabinet. Calculate the rate of increase of its inlet air temperature. The delay time relative to the corresponding rack exhaust temperature rise , is represented as:

[0097] ;

[0098] When satisfied At that time, it was determined that there was an air intake disturbance in the adjacent cabinet;

[0099] in, The rate of temperature rise of the air intake air in adjacent cabinets. To delay the time of air intake disturbance, The disturbance rate threshold is used to determine whether the rise in the inlet air temperature of an adjacent rack has exceeded the normal cold aisle temperature fluctuation range, thus identifying that the rack is affected by the hot airflow discharged from the preceding rack. The value range is 0.015℃ / s-0.050℃ / s. The rate of change in inlet air temperature caused by normal air conditioning supply adjustment, environmental airflow disturbance, and sensor error is usually no more than 0.010℃ / s. However, after the heat from the preceding rack breaks through the local heat exchange boundary, the rate of increase in inlet air temperature caused by the hot airflow invading the inlet air area of ​​the adjacent rack usually reaches more than 0.015℃ / s. Therefore, setting the disturbance rate threshold within the above range can effectively identify the actual heat transfer process. At the same time, since the temperature rise of the adjacent rack inlet air is a decay response after heat transfer, its threshold should be lower than the temperature rise slope threshold in S21, and is taken as 30% to 60% of the temperature rise slope threshold to conform to the physical law of gradual decay during heat transfer. The time interval of the disturbance delay. This indicates the moment when the intake air temperature of the adjacent cabinet began to become abnormal. This corresponds to the moment when the exhaust temperature of the corresponding server rack began to abnormal.

[0100] S24: Combine the characteristics of the exhaust temperature persistence in step S21, the characteristics of the intake air recirculation change in step S22, and the disturbance of the intake air of adjacent cabinets in step S23 to make a joint judgment:

[0101] If the continuous heating characteristic, the air intake and recirculation change characteristic, and the air intake disturbance of at least one adjacent cabinet are all met at the same time, it is determined that the actual heat release process of the cabinet has broken through the local heat exchange boundary, and a first boundary intrusion mark is generated.

[0102] Otherwise, generate a boundary-breaking marker.

[0103] S25: After the first boundary intrusion marker is generated, continue to track the intake air temperature response sequence of the racks arranged along the airflow direction within the same cold aisle, and record the first time the intake air temperature of each rack exceeds the abnormal intake air temperature threshold in order from the intake end of the cold aisle to the far end. Response time This forms a delayed propagation chain, represented as:

[0104] ;

[0105] When the response time difference between adjacent racks When the time is right, it is determined that there is an effective heat transfer delay path;

[0106] in, The abnormal inlet air temperature threshold is used to determine whether the inlet air temperature of a cabinet in the cold aisle is abnormally high due to the intrusion of heat from the front-end cabinet, thus forming a delayed transfer chain. The value is 1.0℃-2.0℃. When the exhaust air from the front-end cabinet breaks through the local heat exchange boundary and is transferred along the cold aisle, the heat from the front-end cabinet will usually cause the inlet air temperature to rise by 1℃-2℃ when it reaches the far-end cabinet. This is in line with the laws of physical transfer. Therefore, setting the threshold in this range can effectively capture the actual heat impact. The minimum interval for transmission delay is used to determine whether the temperature response interval of the air intake of adjacent cabinets in the delay transmission chain has a causal relationship with actual heat transfer, and to prevent misjudgment of the delay path due to transient disturbances or high-frequency noise. The value is 2s-6s. If the interval is less than 2s, high-frequency airflow disturbances may cause misjudgment; if the interval is too large, the rapid heat propagation path may be missed. This represents the total number of server racks within the cold aisle.

[0107] S26: Perform causal analysis on delayed propagation chains, including:

[0108] S261: Extract the exhaust temperature change history of the first responding cabinet;

[0109] S262: If the first responding cabinet does not show a true load heat release marker within the preset time window before its own response time, but there is a record that the front cabinet has exceeded the local heat exchange boundary, then it is determined that the abnormal air intake temperature of the first responding cabinet is caused by the disturbance hot air flow discharged from the front cabinet being transferred through the cold aisle.

[0110] The preset time window is 10s-60s, which matches the thermal inertia of the rack and is consistent with the delay propagation chain logic. It can cover most of the real delay while filtering short-term transient disturbances.

[0111] S263: Mark the first responding cabinet as the inducing source cabinet, and attribute subsequent responding cabinets in the delayed propagation chain to the inducing source, generating the original cold aisle thermal boundary intrusion result including the inducing source cabinet identifier, the induction propagation direction, and the set of induced cabinets.

[0112] S3: Taking the original cold aisle thermal boundary intrusion result as input, first perform directional air supply compensation on the rack area where thermal boundary intrusion occurs, and adjust the air supply priority of the corresponding rack for cold aisle areas with cross-rack induced temperature rise; if the cold aisle thermal boundary intrusion result continues to expand after directional air supply compensation, increase the cooling output power of the corresponding cooling equipment to generate dynamic adjustment result of data center cooling power.

[0113] S3 specifically includes:

[0114] S31: Analyze the original cold aisle thermal boundary intrusion results to obtain the recorded induction source cabinet identifier, induction propagation direction, and set of induced cabinets; let the induction source cabinet be... The induced cabinets are grouped as The set of cabinets to be compensated is represented as:

[0115] ;

[0116] in, Indicates the source cabinet of the induction process. Indicates the first One induced cabinet, Indicates the number of racks affected by the inducement. This indicates the collection of cabinets to be compensated.

[0117] Meanwhile, the induced propagation direction in the original cold aisle thermal boundary intrusion result is defined as the priority direction for subsequent directional air supply compensation, which is used to guide the priority delivery path of cold energy.

[0118] S32: For each rack in the set of racks to be compensated, generate corresponding directional airflow compensation parameters based on the rack's position coordinates in the cold aisle and its response order in the delay propagation chain; let the first... The location coordinates of each rack are The response order in the delayed propagation chain is as follows: Then, the directional air supply compensation parameters are generated, expressed as:

[0119] ;

[0120] in, Indicates the deflection angle of the air supply nozzle. This indicates the increase in local air supply velocity. Indicates the duration of air supply. This indicates the directional air supply compensation parameters for the corresponding cabinet;

[0121] To ensure that the cooling capacity is preferentially applied to the induction source cabinet, let the corresponding air velocity increment for the induction source cabinet be... Then it satisfies At the same time, it satisfies the following along the direction of induced propagation. This creates a cooling compensation gradient that gradually decreases along the heat propagation path.

[0122] S33: Send the directional air supply compensation parameters to the actuator of the variable air supply unit above the corresponding cold aisle; the actuator then... Adjust the direction of the air supply nozzles according to Adjust the local air supply volume; based on Maintaining the air supply compensation time creates a directional cooling compensation airflow distributed along the induced propagation direction in the area of ​​the cabinets to be compensated, thus achieving the first round of directional air supply compensation.

[0123] S34: During the first round of directional air supply compensation, a priority allocation mechanism for air supply is further constructed for cold aisle areas where cross-rack induced temperature rise occurs; Let the first... The air supply priority weight for each server rack is: The priority weight of the source cabinet is then expressed as: ,satisfy ;

[0124] in, This indicates the priority weight of the induction source cabinet, with a value range of 0.50-0.70. The induction source cabinet is the starting point for cold aisle thermal boundary intrusion, and the hot airflow discharged from it will continue to spread to downstream cabinets. Therefore, the heat source should be suppressed first. In actual operation, when about 50%-70% of the adjustable air supply resources are preferentially allocated to the induction source cabinet, the heat output intensity can be quickly reduced and the further diffusion of heat can be blocked. This indicates the priority weight of the induced cabinet, with a value range of 0.50-0.30. The induced cabinet belongs to the affected area after heat propagation, and its thermal risk is usually lower than that of the induced source cabinet. Therefore, a lower air supply weight can meet the local cooling needs. At the same time, as the cabinet moves further down the delay transmission chain, its weight should decrease step by step.

[0125] When the total air supply capacity of the variable air supply unit is limited, air supply resources are allocated according to their weights, so that more cooling capacity is preferentially allocated to the front-end area of ​​heat propagation, thereby blocking further heat diffusion.

[0126] S35: In the first evaluation window after the first round of directional air supply compensation is completed, re-collect the exhaust temperature data and cold aisle inlet temperature data of the set of cabinets to be compensated, and regenerate the updated cold aisle thermal boundary intrusion results according to the method in step S2.

[0127] Let the number of affected cabinets corresponding to the original cold aisle thermal boundary intrusion result be . The number of affected racks after the update is The change in the thermal boundary intrusion range is expressed as:

[0128] ;

[0129] in, Indicates the original intrusion range. Indicates the updated intrusion range. This indicates the change in the intrusion range, and the change is used to determine whether the thermal boundary intrusion range continues to expand.

[0130] S36: When satisfied When the thermal boundary intrusion range does not expand or shrinks, the current directional air supply compensation parameters are maintained, and the wind speed increment is gradually reduced according to the preset attenuation step size, as shown in the following:

[0131] ;

[0132] Until This round of adjustments will end.

[0133] in, This indicates the wind speed attenuation step size, used to gradually reduce the local airflow velocity increment when the thermal boundary intrusion range has not continued to expand or has already shrunk. This avoids a sudden withdrawal of airflow compensation that could cause a temperature rebound. The value range is 0.05m / s-0.20m / s. If the attenuation step size is too small, the airflow compensation withdrawal process will be too slow, which may easily cause unnecessary waste of cooling capacity. If the attenuation step size is too large, the local cooling capacity will decrease rapidly, which may cause the intake air temperature of the cabinet to be compensated to rise again. Therefore, setting the attenuation step size to 0.05m / s-0.20m / s can reduce excessive cooling while ensuring a stable temperature drop.

[0134] When satisfied If this indicates that the thermal boundary intrusion range continues to expand, then the power of the refrigeration equipment should be increased while maintaining the current air supply priority.

[0135] Assume the current output level of the refrigeration equipment is The output level after the upgrade is represented as:

[0136] ;

[0137] in, Indicates the power increase step size. Indicates the current cooling output level. This indicates the upgraded cooling output level.

[0138] Generate a command to increase cooling output power and execute the second round of joint regulation.

[0139] S37: During the second evaluation window after the second round of joint regulation is completed, the cold aisle thermal boundary intrusion results are regenerated;

[0140] If the updated intrusion range stops expanding, maintain the current cooling output power until the exhaust temperature in the cabinet assembly to be compensated returns to the normal range;

[0141] If the intrusion continues to expand, repeat step S36 to continue increasing the cooling output level.

[0142] Let the maximum allowable safe power threshold of the system be... Then it satisfies ;

[0143] The value range is 80%-95% of the rated maximum power. If the setting is too low, it may not be able to prevent local overheating of the cabinet under extreme heat load conditions, affecting server safety. If the setting is too high, although it can quickly suppress heat boundary intrusion, long-term high-load operation of the equipment can easily accelerate aging, increase energy consumption and failure risk.

[0144] when When the increase stops, the final result of the dynamic adjustment of the computer room cooling capacity is generated, which is represented as:

[0145] ;

[0146] in, This indicates the dynamic adjustment result of the computer room's cooling capacity. This represents the sequence of directional air supply compensation parameters executed in each round. This represents the record of changes in cooling output power throughout the entire adjustment process.

[0147] Through the above process, a progressive adjustment mechanism from directional air supply compensation to cooling power enhancement is achieved. Priority is given to using airflow organization optimization to solve the problem of local thermal boundary intrusion. Cooling output power is only increased when the thermal boundary intrusion continues to expand, thereby reducing the energy consumption of the cooling system and improving the efficiency of cooling capacity utilization.

[0148] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0149] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for dynamically adjusting the cooling power of a data center based on changes in server load, characterized in that, Includes the following steps: S1: Collect server load change data, server fan speed change data, rack exhaust temperature data, and cold aisle intake temperature data; identify the rack exhaust dilution state caused by server fan adaptive speed increase after server load increases; and generate a load heat release marker. The load heat release marker is used to distinguish between the actual heat release process caused by server load increase and the short-term exhaust disturbance process caused by server fan speed increase. S2: Using the actual heat release marker as input, jointly determine the continuity of exhaust temperature of the corresponding cabinet, the change of cold aisle airflow return, and the disturbance of airflow to adjacent cabinets to identify whether the actual heat release process has broken through the local heat exchange boundary of the cabinet; track the time sequence of airflow temperature response of each cabinet in the same cold aisle, identify the delayed heat transfer chain in the cold aisle, and determine whether there is a cross-cabinet induced heating phenomenon induced by a locally high-load cabinet, generating the original cold aisle thermal boundary intrusion result; the cross-cabinet induced heating phenomenon is that the disturbed hot airflow discharged from the front-end cabinet does not directly cause its own temperature runaway, but causes the back-end cabinet to show an abnormal increase in airflow temperature in advance; S3: Using the original cold aisle thermal boundary intrusion result as input, first perform directional air supply compensation on the cabinet area where thermal boundary intrusion occurs, and adjust the air supply priority of the corresponding cabinet for the cold aisle area where cross-cabinet induced heating phenomenon exists; if the cold aisle thermal boundary intrusion result is still detected to continue to expand after the directional air supply compensation, increase the cooling output power of the corresponding cooling equipment to generate the dynamic adjustment result of the data center cooling power.

2. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 1, is characterized in that... S1 specifically includes: S11: Temperature sensors are deployed on the internal exhaust side and the front cold aisle side of each server rack to continuously collect the exhaust temperature sequence and cold aisle inlet temperature sequence of each server rack in the first sampling period; the load occupancy sequence and fan speed sequence of each server are obtained in real time through the server's baseboard management controller. S12: Compare the load change rate of two adjacent sampling points in the load occupancy rate sequence with a preset load increase threshold. When the load change rate exceeds the load increase threshold, mark the corresponding server as entering the load increase candidate state, and record the fan speed sequence of the server within a preset short window after the start time of the load increase candidate state. S13: Compare the fan acceleration value with a preset fan adaptive acceleration threshold. If the fan acceleration value exceeds the fan adaptive acceleration threshold, it is determined that there is a cabinet exhaust dilution state caused by fan adaptive speed increase on the exhaust side of the server rack. S14: After determining that the cabinet exhaust dilution state exists, extract the actual temperature change rate of the exhaust temperature sequence of the server cabinet within the preset short-time window, and simultaneously extract the reference temperature change rate of the cold aisle inlet air temperature sequence of the server cabinet within the same window. Compare the actual temperature change rate with the reference temperature change rate. If the actual temperature change rate is less than the sum of the reference temperature change rate and the preset dilution offset, generate a first intermediate marker; otherwise, generate a second intermediate marker. S15: Use the first intermediate mark as the actual release mark of the load heat, and the second intermediate mark as the short-term exhaust disturbance mark.

3. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 2, is characterized in that... The exhaust dilution state refers to the physical phenomenon where, after the server fan speeds up, a large amount of cold air is forcibly drawn through the inside of the server and discharged at high speed from the exhaust side, resulting in an increase in the local airflow velocity on the exhaust side and a decrease or a smaller increase in the exhaust temperature in a short period of time.

4. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 2, is characterized in that... The load heat release indicator indicates that the heat generated by the current increase in server load has not been fully released to the exhaust side and is still accumulated in the form of heat inside the server or in a local area of ​​the rack; the short-term exhaust disturbance indicator indicates that the current exhaust temperature change is mainly dominated by the airflow dilution effect caused by the fan speed increase.

5. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 1, is characterized in that... S2 specifically includes: S21: Using the load heat release mark as a trigger signal, for the corresponding cabinet marked as having a real heat release process, extract the exhaust temperature change persistence parameter of the cabinet within the continuous monitoring window after the load heat release mark is generated. The exhaust temperature change persistence parameter includes the continuous rise slope of the exhaust temperature and the duration of the continuous rise slope. When the continuous rise slope is greater than a preset temperature rise slope threshold and the duration of ... S22: Simultaneously extract the intake air temperature sequence of the cold aisle where the corresponding cabinet is located within the continuous monitoring window, including the positive deviation of the cold aisle intake air temperature from the reference intake air temperature of the cold aisle within the continuous monitoring window and the fluctuation amplitude of the positive deviation. When the positive deviation is greater than a preset recirculation temperature threshold and the fluctuation amplitude is less than a preset stable fluctuation threshold, it is determined that the corresponding cold aisle has intake air recirculation change characteristics caused by cabinet exhaust recirculation. S23: Extract the intake air temperature sequence of adjacent cabinets within the continuous monitoring window as intake air disturbance parameters. The intake air disturbance parameters include the rate of increase of the intake air temperature of the adjacent cabinet and the delay time relative to the rise of the exhaust air temperature of the corresponding cabinet. When the rate of increase of the intake air temperature of the adjacent cabinet exceeds a preset disturbance rate threshold and the delay time is within a preset disturbance delay interval, it is determined that the adjacent cabinet is experiencing intake air disturbance.

6. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 5, is characterized in that... S2 further includes: The characteristics of continuous exhaust temperature, changes in air inlet recirculation, and disturbances to air inlet in adjacent cabinets are jointly determined. The joint determination includes determining that if the continuous temperature rise, changes in air inlet recirculation, and disturbances to air inlet in at least one adjacent cabinet are simultaneously satisfied, the actual heat release process has broken through the local heat exchange boundary of the cabinet, and a first boundary intrusion mark is generated; otherwise, a boundary-not-broken mark is generated. After the first boundary intrusion marker is generated, the air intake temperature response time sequence of each rack arranged sequentially along the airflow direction in the same cold aisle is tracked. The response time when the air intake temperature of each rack first exceeds the preset abnormal air intake temperature threshold is recorded in order from the air intake end of the cold aisle to the far end. The response times of each rack are arranged in chronological order to form a delayed heat transfer chain in the cold aisle. When the difference in response time between adjacent racks in the delayed transfer chain is greater than the preset minimum transfer delay interval, it is determined that there is an effective heat delayed transfer path. Causal analysis is performed on the delayed transfer chain.

7. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 6, is characterized in that... The causal analysis specifically includes: Extract the exhaust temperature change history of the first responding cabinet in the delayed transmission chain. If the first responding cabinet does not show a real release of its own load heat within the preset time window before its own response time, but there is a record that the front cabinet has broken through the local heat exchange boundary, then it is determined that the abnormal intake temperature of the first responding cabinet is caused by the disturbance of hot air flow discharged from the front cabinet through the cold aisle, and it is marked as cross-cabinet induced heating phenomenon. All abnormal air intake temperatures of subsequent response cabinets in the delayed propagation chain are attributed to the same induction source, generating the original cold aisle thermal boundary intrusion result, including the induction source cabinet identifier, the induction propagation direction, and the induced cabinet set.

8. The method for dynamically adjusting the cooling power of a data center based on changes in server load, as described in claim 1, is characterized in that... S3 specifically includes: S31: Analyze the original cold aisle thermal boundary intrusion results, extract the induction source cabinet identifier, the induction propagation direction, and the set of cabinets affected by the induction; mark the cabinets corresponding to the induction source cabinet identifier and all cabinets in the set of cabinets affected by the induction as the set of cabinets to be compensated, and use the induction propagation direction as the preferred direction of directional air supply; S32: For each cabinet in the set of cabinets to be compensated, generate directional air supply compensation parameters for the corresponding cabinet based on the cabinet's position coordinates in the cold aisle and the response order in the delay transmission chain; the directional air supply compensation parameters include the deflection angle of the air supply nozzle toward the air inlet of the cabinet, the local air supply velocity increment for the cabinet, and the air supply duration. S33: The generated directional air supply compensation parameters for each cabinet are sent to the variable air supply unit actuator above the corresponding cold aisle. The actuator adjusts the nozzle direction according to the deflection angle and increases the air supply volume according to the local air supply velocity increment. During the air supply duration, the first round of directional air supply compensation is performed on the cabinet set to be compensated. S34: While performing the first round of directional air supply compensation, for cold aisle areas where cross-rack induced heating occurs, the air supply priority of the corresponding rack is dynamically adjusted according to the response order of each rack in the delay transmission chain. S35: In the first evaluation window after the first round of directional air supply compensation is completed, the exhaust air temperature data and cold aisle inlet air temperature data of each cabinet in the set of cabinets to be compensated are collected again, and the updated cold aisle thermal boundary intrusion results are regenerated; the updated cold aisle thermal boundary intrusion results are compared with the original cold aisle thermal boundary intrusion results to determine whether the thermal boundary intrusion range continues to expand. S36: If the thermal boundary intrusion range does not continue to expand or shrinks, maintain the current directional air supply compensation parameters and gradually reduce the air supply velocity increment to zero according to the preset attenuation step, ending this adjustment; if the thermal boundary intrusion range continues to expand, based on the first round of directional air supply compensation, increase the fan speed or chilled water valve opening of the corresponding refrigeration equipment according to the preset power increase step, generate a refrigeration output power increase command, and at the same time keep the air supply priority unchanged, and execute the second round of joint adjustment.

9. A method for dynamically adjusting data center cooling power based on server load changes according to claim 8, characterized in that, The airflow velocity increment of the inducing source cabinet is greater than that of any induced cabinet, and the airflow velocity increment of the cabinets further back in the induction propagation direction decreases progressively. The airflow priority includes the weighting of airflow resources allocated to each cabinet when the total airflow capacity of the variable airflow unit above the cold aisle is limited. The inducing source cabinet is assigned the highest priority weight, and the cabinets that respond earlier in the induced cabinets are assigned a higher priority weight. The airflow priority decreases progressively according to the induction propagation direction.

10. A method for dynamically adjusting data center cooling power based on server load changes according to claim 8, characterized in that, S3 further includes: Within the second evaluation window following the completion of the second round of joint regulation, the cold aisle thermal boundary intrusion results are regenerated; If the thermal boundary intrusion range stops expanding, the current cooling output power is maintained until the exhaust temperature of the cabinet assembly to be compensated returns to the normal range, after which it is gradually reduced. If the thermal boundary intrusion range continues to expand, repeat step S36 to increase the cooling output power again until the thermal boundary intrusion range stops expanding or the cooling output power reaches the preset maximum safe power threshold. Finally, the dynamic adjustment result of the computer room cooling power is generated. The dynamic adjustment result of the computer room cooling power includes the sequence of directional air supply compensation parameters executed in each round and the record of cooling output power changes.

Images