A data center dynamic loop energy consumption edge cloud cooperative optimization method

Abstract

The application discloses a data center dynamic environment energy consumption edge cloud cooperative optimization method, relates to the edge cloud cooperative data processing technical field, and comprises the following steps: collecting fast load data and slow heat field data through an edge side dynamic environment group control management platform, generating edge side standard data frames with edge side sampling time stamps and cache supplement transmission identifiers; determining a thermal response lag time based on time-aligned data of a cloud end data center operation analysis platform, calculating a causal load time corresponding to the slow heat field data, identifying a wrong phase splicing window of high load and low thermal response or low load and high thermal response, generating a cloud end optimization state containing a wrong phase type identifier, and generating an energy saving optimization instruction to issue an edge side execution according to the cloud end optimization state. The application can avoid energy saving instruction lag, reverse regulation, excessive refrigeration and frequent switching caused by wrong phase splicing of the fast load data and the slow heat field data, and improve the accuracy and stability of data center dynamic environment energy consumption optimization.

Description

Technical Field

[0001] This invention relates to the field of edge-cloud collaborative data processing technology, and in particular to a method for edge-cloud collaborative optimization of data center environmental energy consumption. Background Technology

[0002] As data centers, cloud computing centers, and intelligent computing centers continue to expand, the number of environmental monitoring devices in computer rooms, such as power supply and distribution systems, UPS systems, power distribution branches, chillers, precision air conditioners, chilled water supply and return pipelines, temperature and humidity sensors, and PUE acquisition devices, continues to increase. Cooling energy consumption and power supply and distribution losses account for a relatively high proportion of the energy consumption of data center operations. In order to improve energy utilization efficiency, existing data centers typically collect data such as IT load power, UPS output power, power distribution branch power, cabinet power, cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioner operating status through edge environmental monitoring platforms. The collected data is then uploaded to the cloud platform for energy consumption analysis, trend prediction, and energy-saving strategy generation. The edge platform then executes control actions such as adjusting the air conditioner set temperature, distributing the chiller load, or switching the precision air conditioner operating mode, thereby forming an edge-cloud collaborative optimization model for data center environmental energy consumption.

[0003] In actual operation, fast load data such as IT load power, UPS output power, power distribution branch power, and cabinet power can quickly reflect changes in the heat generation end of the data center. However, slow thermal field data such as cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status only show corresponding changes after air flow, heat exchange process, cooling equipment response, and spatial thermal inertia transfer. There is an inherent thermal response lag between the two types of data. Due to the inconsistency in the acquisition cycle, upload frequency, communication link, side-side buffering and retransmission mechanism, network transmission latency, and cloud analysis time of different environmental monitoring devices, the cloud platform is prone to incorrectly splicing fast load data and slow thermal field data from different times or different physical response stages into the same operating state at the same time, forming a pseudo operating state of high load and low thermal response or low load and high thermal response. This leads to a mismatch between the energy-saving optimization instructions generated by the cloud and the actual operating state on the side, resulting in problems such as weakening cooling before the heat peak, compensating for cooling after the heat peak subsides, overcooling, or frequent control switching, affecting the accuracy, stability, and executability of data center energy-saving optimization. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies where fast load data and slow thermal field data are spliced ​​out of phase due to differences in acquisition cycles, side-side buffering and retransmission, network transmission latency, and cloud analysis time, resulting in a mismatch between cloud-based energy-saving optimization instructions and the actual side-side operating status. Therefore, this invention proposes a data center dynamic environment energy consumption edge-cloud collaborative optimization method.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution:

[0006] A method for collaborative optimization of data center environmental energy consumption via edge-cloud integration includes:

[0007] The side-side dynamic environment group control management platform collects fast load data and slow thermal field data of the data center dynamic environment system, adds side-side sampling timestamps, and generates the original dynamic environment dataset;

[0008] The original dynamic environment dataset is cleaned, missing data is marked, and units are standardized. The original edge sampling timestamp is retained and written into the cached data transmission identifier. The edge standard data frame is generated and uploaded to the cloud data center operation and analysis platform.

[0009] The cloud data center operation analysis platform uses the edge sampling timestamp as a benchmark to perform time alignment on the edge standard data frame, extracts the fast load sequence and slow thermal field sequence from the time-aligned edge standard data frame, and determines the thermal response lag time through the cross-correlation between the fast load sequence and the slow thermal field sequence.

[0010] Within the current optimization calculation window used to generate energy-saving optimization instructions, the causal load time is determined based on the thermal response lag time and the side sampling timestamp of the slow thermal field data. The fast load state corresponding to the causal load time is compared with the fast load state of the current optimization calculation window to identify the phase misalignment window and the phase misalignment type.

[0011] Based on the current fast load status, slow heat field measured status, causal load time, and phase error type, generate a cloud-based optimization status that includes fast load status, slow heat field status, and phase error type identifiers;

[0012] Energy-saving optimization instructions are generated based on the cloud-based optimization status and sent to the side-side dynamic environmental group control management platform.

[0013] The side-side dynamic environment group control management platform executes energy-saving optimization commands and feeds back fast load data and slow thermal field data after execution to update the identification results of the phase misalignment splicing window.

[0014] Preferably, the fast load data includes at least one of IT load power, UPS output power, power distribution branch power, and cabinet power; the slow hot field data includes at least one of cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status.

[0015] Preferably, the side-side standard data frame includes a data value, a device identifier, a data type, a side-side sampling timestamp, and a cache retransmission identifier; wherein, the data value originates from fast load data or slow hot field data, the device identifier is used to determine the data source device, the data type is used to distinguish between fast load data and slow hot field data, and the cache retransmission identifier is used to distinguish between real-time uploaded data and cache retransmission data.

[0016] Preferably, the cloud data center operation analysis platform performs time alignment of the edge-side standard data frames based on the edge-side sampling timestamp, including:

[0017] Identify cached data to be re-transmitted based on the cache re-transmission identifier;

[0018] The real-time uploaded data and cached re-uploaded data are restored to the sampling time order according to the side sampling timestamp;

[0019] Data with consecutive sampling times and no missing markers are used as historical normal operation data;

[0020] Extract fast load sequences and slow thermal field sequences from historical normal operation data.

[0021] Preferably, determining the thermal response hysteresis time through the cross-correlation of the fast load sequence and the slow thermal field sequence includes:

[0022] Calculate the change in the fast load sequence between adjacent sampling times;

[0023] Calculate the change of the slow-heating field sequence under candidate lag times;

[0024] Calculate the correlation between the changes in the fast load sequence and the changes in the slow thermal field sequence at different candidate lag times;

[0025] The candidate lag time when the correlation reaches its maximum is determined as the thermal response lag time of the corresponding slow-heat field data.

[0026] The candidate lag time is determined by the actual sampling period of the corresponding slow-heat field data and the historical normal operation data.

[0027] Preferably, the determination of the causal load time includes:

[0028] Obtain the side sampling timestamps and corresponding thermal response hysteresis times of the slow-heating field data;

[0029] Subtract the corresponding thermal response hysteresis time from the side sampling timestamp of the slow-heating field data to obtain the causal load time corresponding to the slow-heating field data.

[0030] In the time-aligned fast load sequence, the fast load state corresponding to the causal load time is obtained, and this fast load state is taken as the load stage actually reflected by the slow thermal field data.

[0031] Preferably, the identification of the phase misalignment splicing window and the phase misalignment type includes:

[0032] Compare the fast load state corresponding to the causal load time with the fast load state of the current optimization calculation window;

[0033] When the fast load state of the current optimization calculation window is in an elevated state relative to the fast load state corresponding to the causal load time, the current optimization calculation window is identified as a high load low thermal response out-of-phase window.

[0034] When the fast load state of the current optimization calculation window is in a reduced state relative to the fast load state corresponding to the causal load time, the current optimization calculation window is identified as a low load high thermal response out-of-phase window.

[0035] Preferably, the generation of the cloud-based optimized state includes:

[0036] When the phase error type is a high load low thermal response phase error window, a cloud-optimized state is generated that includes the current fast load increase state, the slow thermal field lagging behind the current fast load increase state, and the high load low thermal response phase error type identifier.

[0037] When the phase error type is a low load high thermal response phase error window, a cloud-optimized state is generated that includes the current fast load reduction state, the slow thermal field lagging behind the current fast load reduction state, and the low load high thermal response phase error type identifier.

[0038] When the current optimization calculation window is not identified as a misphase splicing window, a cloud optimization status is generated that includes the current fast load status, the actual slow thermal field status, and no misphase type identifier.

[0039] Preferably, the step of generating energy-saving optimization instructions based on cloud-based optimization status includes:

[0040] When the cloud-based optimization status contains a high-load, low-thermal-response phase-out type identifier, the generation of energy-saving optimization instructions that weaken cooling capacity is restricted.

[0041] When the cloud-based optimization status contains a low-load, high-heat-response phase-out type identifier, the generation of energy-saving optimization instructions that only improve cooling capacity based on the actual measured status of the slow-heat field is restricted.

[0042] When the cloud-based optimization status includes a no-error-phase-type identifier, it generates air conditioning set temperature adjustment commands, chiller load distribution commands, precision air conditioning operation mode switching commands, or environmental protection equipment energy-saving operation commands based on the current fast load status, slow heat field measured status, and energy consumption level.

[0043] Preferably, the side-side dynamic environment group control management platform executes energy-saving optimization commands and feeds back fast load data and slow thermal field data after execution, including:

[0044] After receiving the energy-saving optimization command, the side-side dynamic environment group control management platform verifies the executable status of the corresponding dynamic environment equipment;

[0045] When the corresponding environmental equipment is in an executable state, the energy-saving optimization command is executed, and the execution time, execution result, fast load data after execution, and slow heat field data after execution are fed back to the cloud data center operation analysis platform.

[0046] When the corresponding environmental equipment is not in an executable state, the energy-saving optimization command will not be executed, and the non-execution result will be reported to the cloud data center operation analysis platform;

[0047] The cloud-based data center operation analysis platform recalculates the thermal response lag time based on the fast load data and slow thermal field data after execution, and uses the recalculated thermal response lag time to determine the next round of causal load time.

[0048] The cloud-based data center operation analysis platform updates the misphase splicing window identification results based on the fast load data and slow heat field data after execution, and adjusts the device's executable status when the next round of energy-saving optimization instructions is generated based on the unexecuted results.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1. This invention uses a side-side dynamic environment group control management platform to add side-side sampling timestamps when collecting fast load data and slow thermal field data. In the event of network interruption, side-side buffering, or data retransmission, the original side-side sampling timestamps are retained and a buffer retransmission identifier is written, enabling the cloud data center operation analysis platform to perform time alignment based on the actual sampling time rather than the upload arrival time. Furthermore, based on the time-aligned fast load sequence and slow thermal field sequence, the thermal response lag time is determined, and the causal load time corresponding to the slow thermal field data is calculated using the thermal response lag time. This can identify high-load, low-thermal-response phase misalignment windows and low-load, high-thermal-response phase misalignment windows, preventing the cloud platform from incorrectly synthesizing data from different times or different physical response stages into the same operating state, thus improving the accuracy of operating state construction during edge-cloud collaborative optimization.

[0051] 2. After identifying the phase misalignment window, this invention does not directly input the fast load data and slow thermal field data in the current optimization calculation window into the energy-saving optimization model. Instead, it generates a cloud-based optimization state that includes the fast load state, slow thermal field state, causal load time, thermal response lag time, and phase misalignment type identifier. It then constrains the candidate energy-saving optimization instruction set based on the phase misalignment type. For high-load, low-thermal-response phase misalignment windows, it restricts the generation of energy-saving optimization instructions that weaken cooling capacity, preventing premature reduction of cooling output before the heat peak arrives. For low-load, high-thermal-response phase misalignment windows, it restricts the generation of energy-saving optimization instructions that only improve cooling capacity based on the measured slow thermal field state, preventing continued compensation cooling after the heat peak subsides. This reduces problems such as energy-saving instruction lag, reverse adjustment, over-cooling, and frequent control switching, improving the stability, accuracy, and executability of edge-cloud collaborative optimization of data center environmental energy consumption. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0053] Figure 1 A flowchart illustrating a data center environmental energy consumption edge-cloud collaborative optimization method provided by the present invention;

[0054] Figure 2 This is an architecture diagram of the data center environmental energy consumption edge-cloud collaborative system provided by the present invention;

[0055] Figure 3 This is a schematic diagram of the fast load-slow thermal field phase-shifting splicing provided by the present invention. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0057] Example: This example provides a data center environmental energy consumption edge-cloud collaborative optimization method, applied to the environmental energy consumption monitoring and energy-saving optimization scenario of a data center computer room; the data center computer room includes a power supply and distribution system, a UPS system, cabinet branch lines, a cooling system, and an environmental monitoring area; the power supply and distribution system includes mains power input equipment, distribution cabinets, distribution branch lines, and smart meters; the UPS system includes a UPS host and a UPS monitoring unit; the cooling system includes chiller units, chilled water supply and return pipelines, precision air conditioners, and corresponding cooling controllers; the environmental monitoring area includes cold aisles, hot aisles, cabinet entrance areas, cabinet return air areas, and temperature and humidity monitoring points;

[0058] like Figure 2 As shown, this embodiment adopts an edge-cloud collaborative architecture, including an energy consumption data acquisition layer, an edge-side environmental control management platform, a network transmission layer, a cloud-based data center operation and analysis platform, and an AI service center. The energy consumption data acquisition layer is deployed in the data center server room to collect fast load data and slow thermal field data of the environmental system. The edge-side environmental control management platform is deployed on a local server, industrial control computer, or edge computing node in the server room. It is used to access the energy consumption data acquisition layer, perform edge-side preprocessing, write cached retransmission identifiers, and generate edge-side standard data frames for the collected data, and execute energy-saving optimization instructions issued from the cloud. The network transmission layer is used to upload the edge-side standard data frames to the cloud-based data center operation and analysis platform and to issue energy-saving optimization instructions from the cloud to the edge-side environmental control management platform. The cloud-based data center operation and analysis platform is used to perform time alignment, thermal response lag time determination, phase misalignment window identification, cloud-based optimization status generation, and energy-saving optimization instruction generation for the edge-side standard data frames. The AI ​​service center is used to assist the cloud-based data center operation and analysis platform in performing energy consumption trend analysis and energy-saving optimization calculations.

[0059] In this embodiment, fast load data refers to data that can directly or in near real-time reflect changes in the heat generation end of the data center; specifically, fast load data includes at least one of IT load power, UPS output power, power distribution branch power, and cabinet power; changes in fast load data usually occur before changes in computer room temperature, return air temperature, or chilled water return temperature, so in this embodiment, it is used as the basic data for judging the heat generation status of the data center.

[0060] In this embodiment, slow thermal field data refers to data that reflects changes in the thermal field of the data center after airflow, heat exchange process, cooling equipment response, and spatial thermal inertia transmission. Specifically, slow thermal field data includes at least one of cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status. Slow thermal field data does not necessarily correspond to the fast load status at the current upload time, but may correspond to the fast load status at an earlier time.

[0061] In this embodiment, the fast load-slow thermal field mis-phase splicing refers to the situation where the cloud data center operation analysis platform incorrectly combines fast load data and slow thermal field data with inconsistent sampling times or physical response stages into the same operating state within the same optimization calculation window. This situation will form an inverse state combination of high load and low thermal response or low load and high thermal response, causing the state on which the cloud optimization calculation is based to deviate from the actual operating state on the side. There is a physical causal relationship between fast load data and slow thermal field data: the increase or decrease of fast load data represents a change at the heat generation end, and slow thermal field data needs to go through air flow, heat exchange process, cooling equipment response and spatial thermal inertia transfer before showing the corresponding change. Therefore, slow thermal field data may not correspond to the fast load state at the side upload time, but may correspond to the fast load state at an earlier time.

[0062] In this embodiment, the data center environmental energy consumption edge-cloud collaborative optimization method includes the following steps;

[0063] S1. The side-side dynamic environment group control management platform collects the fast load data and slow thermal field data of the data center dynamic environment system, adds the side-side sampling timestamp, and generates the original dynamic environment dataset.

[0064] Energy consumption data acquisition equipment and environmental sensing equipment are deployed in the power supply and distribution system, UPS system, cabinet branch circuit, cooling system and environmental monitoring area of ​​the data center computer room, respectively; the energy consumption data acquisition equipment includes at least one of smart meters, multi-function meters, UPS monitoring units, distribution branch circuit meters and cabinet power acquisition devices; the environmental sensing equipment includes at least one of cold aisle temperature sensors, hot aisle temperature sensors, return air temperature sensors, chilled water supply and return water temperature sensors and precision air conditioning operation status acquisition units;

[0065] The side-side dynamic environment group control management platform receives data uploaded by the energy consumption data acquisition device and environmental sensing device through the dynamic environment acquisition interface; for IT load power, UPS output power, power distribution branch power and cabinet power, the side-side dynamic environment group control management platform classifies them into fast load data; for cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature and precision air conditioning supply and return air status, the side-side dynamic environment group control management platform classifies them into slow hot field data;

[0066] When collecting each piece of fast load data or slow thermal field data, the side-side dynamic environment group control management platform adds a side-side sampling timestamp, device identifier, and data type to the data. The device identifier is used to identify the device from which the data comes; the data type is used to distinguish between fast load data and slow thermal field data.

[0067] In this embodiment, the edge sampling timestamp refers to the time identifier generated by the edge environmental control management platform or its connected acquisition device when the data is actually collected; the edge sampling timestamp is used to characterize the physical sampling time corresponding to the data, and is not based on the arrival time of the data being uploaded to the cloud data center operation and analysis platform; for data that is retransmitted after a network interruption, the edge sampling timestamp remains the time identifier corresponding to the original data collection time;

[0068] The edge-side dynamic environment group control management platform aggregates the fast load data and slow thermal field data after adding edge-side sampling timestamps, device identifiers, and data types to generate the original dynamic environment dataset. In this embodiment, the edge-side sampling timestamp is used to reflect the physical sampling time corresponding to the data, rather than the arrival time of the data uploaded to the cloud. Even if caching, retransmission, or network delays occur later, the edge-side sampling timestamp will still remain the time identifier when the data was actually collected, so that the cloud data center operation and analysis platform can restore the data time sequence according to the real sampling order.

[0069] S2. Clean the original dynamic environment dataset, mark missing data and standardize units. Retain the original edge sampling timestamps and write the cached data to the cached data transmission identifier. Generate edge standard data frames and upload them to the cloud data center operation and analysis platform.

[0070] After receiving the original dynamic environment group control management platform, the edge-side dynamic environment dataset is preprocessed. The edge-side preprocessing includes data cleaning, missing data labeling, and unit standardization.

[0071] Data cleaning includes removing failed data acquisition, duplicate uploaded data, and data with incorrect format; missing data marking includes writing a missing data mark at the corresponding data position when data should have been generated at a certain sampling time but no valid data was actually generated; unit standardization includes unifying fast load data into power or energy consumption units that can be used for energy consumption calculation and load comparison, and unifying slow thermal field data into temperature or equipment status units that can be used for thermal field comparison.

[0072] The side-side dynamic environment group control management platform retains the data value, device identifier, data type and side-side sampling timestamp for each piece of preprocessed data, and writes the cache retransmission identifier according to the network transmission status;

[0073] In this embodiment, the cached retransmission identifier is a data identifier used to distinguish between real-time uploaded data and cached retransmission data. When the side-side dynamic environment group control management platform detects network interruption, link switching, side-side caching, or data retransmission events, it writes the corresponding data into the cached retransmission identifier. The cached retransmission identifier does not change the side-side sampling timestamp of the data, but is only used to indicate to the cloud data center operation analysis platform that the upload time of the data may be later than its actual sampling time.

[0074] When the side-side environmental control management platform detects that the network transmission is normal, it marks the corresponding data as real-time uploaded data; when the side-side environmental control management platform detects network interruption, link switching, side-side buffering, or data retransmission events, it marks the corresponding data as buffered retransmission data and keeps the original side-side sampling timestamp of the data unchanged.

[0075] The side-side dynamic environment group control management platform encapsulates the pre-processed data into side-side standard data frames according to a predetermined frame format;

[0076] In this embodiment, the edge standard data frame refers to the structured data unit formed by the edge dynamic environment group control management platform after cleaning, missing data marking, unit standardization, and writing of cache supplementation identifiers into the original dynamic environment dataset; the edge standard data frame includes at least data value, device identifier, data type, edge sampling timestamp, and cache supplementation identifier; the cloud data center operation analysis platform performs time alignment, thermal response lag time determination, and phase misalignment splicing window identification based on the edge standard data frame;

[0077] Among them, the data value comes from fast load data or slow hot field data; the device identifier is used to identify the data source device; the data type is used to distinguish between fast load data and slow hot field data; the cached retransmission identifier is used to distinguish between real-time uploaded data and cached retransmission data;

[0078] The edge-side dynamic environment group control management platform uploads edge-side standard data frames to the cloud data center operation analysis platform through the network transmission layer; in one implementation, the network transmission layer adopts at least one of Ethernet, 4G, 5G, and MQTT communication methods; regardless of the communication method used, the edge-side sampling timestamp in the edge-side standard data frame does not change with the upload time.

[0079] S3, the cloud data center operation analysis platform uses the edge sampling timestamp as a benchmark to perform time alignment on the edge standard data frame, extracts the fast load sequence and slow thermal field sequence from the time-aligned edge standard data frame, and determines the thermal response lag time through the cross-correlation between the fast load sequence and the slow thermal field sequence.

[0080] After receiving the edge standard data frame, the cloud data center operation analysis platform first identifies the cached data based on the cached data transmission identifier, and restores the real-time uploaded data and cached data to the sampling time order according to the edge sampling timestamp. For data with missing flags, the cloud data center operation analysis platform does not use it as valid data for determining the thermal response lag time. The cloud data center operation analysis platform uses data with continuous sampling time order and no missing flags as historical normal operation data.

[0081] In this embodiment, historical normal operation data refers to data in the side standard data frame with continuous sampling time sequence, no missing markers written, and corresponding environmental equipment not in alarm state, protection state, maintenance state, or local manual takeover state. For short-term transition data immediately after the energy-saving optimization command is executed, the cloud data center operation analysis platform does not directly use it as historical normal operation data to determine the thermal response lag time. Instead, it includes the fast load data and slow thermal field data following the short-term transition data in the historical normal operation data after they re-enter the stable sampling sequence. This limitation can avoid the influence of transient disturbances after equipment abnormality, manual maintenance, or control action on the determination result of thermal response lag time.

[0082] For fast load data and slow warm field data with different sampling periods, the cloud data center operation analysis platform uses the sampling time of the slow warm field data or the calculation time of the current optimization calculation window as the alignment benchmark. It selects the fast load data with the closest time and without missing markers from the fast load sequence as the corresponding sampling point. When there are multiple fast load data within the same alignment benchmark, the average value, median value, or end-stability value of the fast load data within the alignment benchmark is used to form the corresponding fast load state. In this way, data with different sampling periods can form a comparable data sequence according to the side sampling timestamp.

[0083] The cloud-based data center operation analysis platform extracts fast load sequences and slow heat field sequences from historical normal operation data; the fast load sequence is formed by arranging time-aligned fast load data according to the side sampling timestamps; the slow heat field sequence is formed by arranging time-aligned slow heat field data according to the side sampling timestamps.

[0084] For scenarios involving multiple types of fast load data, the cloud-based data center operation analysis platform sums or weights the valid data from the IT load power, UPS output power, power distribution branch power, and cabinet power corresponding to the same sampling time according to the edge sampling timestamp, to obtain a comprehensive fast load sequence. When only one type of fast load data is collected, the fast load data of this type is arranged according to the edge sampling timestamp to form a comprehensive fast load sequence. The weights of the weighted summation are determined by the effective sampling ratio of each type of fast load data in the historical normal operation data or the rated capacity of the corresponding branch, and remain unchanged in the same round of thermal response lag time calculation.

[0085] For scenarios involving multiple types of slow-heating field data, the cloud-based data center operation and analysis platform determines the thermal response lag time of each type of slow-heating field data relative to the fast load sequence. When the slow-heating field data is continuous temperature data such as cold aisle temperature, hot aisle temperature, return air temperature, or chilled water supply and return water temperature, the slow-heating field sequence is directly formed according to the side sampling timestamp. When the slow-heating field data is state data such as the supply and return air status of precision air conditioning, the state data is first converted into numerical state quantities corresponding to supply air temperature, return air temperature, fan frequency, valve opening degree, or operating level, and then the slow-heating field sequence is formed. Data that cannot be converted into numerical state quantities does not participate in the thermal response lag time calculation and is only used for equipment state constraints when generating energy-saving optimization instructions.

[0086] Let the fast load synthesis sequence be: Let the j-th type of slow-heating field sequence be: Where j represents the category of slow-heating field data, and t represents the side sampling time;

[0087] The cloud-based data center operations analysis platform calculates the change in the fast load composite sequence between adjacent sampling times:

[0088] ;

[0089] And calculate the change of the j-th type of slow-heating field sequence under the candidate lag time:

[0090] ;

[0091] in, Indicates the time interval between adjacent sampling times. This indicates the candidate lag time; the candidate lag time is determined by the actual sampling period of the corresponding slow-heating field data and historical normal operation data.

[0092] Specifically, the candidate lag time set consists of integer multiples of the actual sampling period of the corresponding slow-heat field data. The cloud data center operation analysis platform determines the value range of the candidate lag time set based on the time range from the start of a rapid load change in historical normal operation data to the corresponding slow-heat field data entering the stable response phase. The stable response phase refers to the stage where the change in the slow-heat field data between consecutive sampling times falls back to the natural fluctuation range of similar slow-heat field data in historical normal operation data. Thus, the candidate lag time set matches both the actual sampling period of the slow-heat field data and the thermal inertia response process of the corresponding data center room. The candidate lag time set is used for cross-correlation calculations to select the thermal response lag time.

[0093] In this embodiment, the natural fluctuation range is determined by the statistical distribution of the changes in adjacent sampling times in historical normal operation data. In one implementation, the natural fluctuation range is represented by the standard deviation range, quantile range, or absolute median range of the changes in adjacent sampling times. After selecting a statistical method in the same embodiment, it is kept consistent in subsequent determination of thermal response lag time, correlation difference judgment, and comparison tolerance determination, so as to avoid inconsistent identification results caused by using different fluctuation standards in different calculation stages.

[0094] The cloud-based data center operations analysis platform calculates the correlation between changes in the fast load sequence and changes in the slow thermal field sequence at different candidate lag times:

[0095] ;

[0096] in, The correlation calculation is indicated. For slow thermal field data that increases with increasing fast load, the cloud data center operation analysis platform uses the maximum positive correlation value to determine the thermal response lag time. For slow thermal field data that decreases or changes in the opposite direction with increasing fast load, the cloud data center operation analysis platform uses the candidate lag time with the largest absolute correlation value and whose change direction conforms to the thermal response pattern in historical normal operation data to determine the thermal response lag time.

[0097] The cloud-based data center operations analysis platform determines the candidate lag time at which the correlation reaches its maximum as the thermal response lag time corresponding to the j-th type of slow-heat field data:

[0098] ;

[0099] in, This represents the set of candidate lag times for the j-th type of slow-heating field data. This represents the thermal response lag time of the j-th type of slow thermal field data relative to the fast load data;

[0100] When multiple candidate lag times have the same correlation, or when the differences in correlation between multiple candidate lag times are within the natural fluctuation range of historical normal operation data, the cloud data center operation analysis platform does not directly use the result of a single correlation calculation. Instead, it counts the number of times these multiple candidate lag times appear in multiple consecutive historical normal operation data windows and determines the candidate lag time with the most occurrences as the thermal response lag time of the corresponding slow thermal field data. When the number of occurrences is the same, the smaller candidate lag time is selected as the thermal response lag time. In this way, the impact of a single abnormal sampling point or short-term noise on the determination result of the thermal response lag time can be reduced.

[0101] In this embodiment, thermal response lag time refers to the response delay time exhibited by slow thermal field data relative to fast load data changes; thermal response lag time is determined by the cloud data center operation analysis platform based on the cross-correlation relationship between the fast load sequence change and the slow thermal field sequence change in historical normal operation data; thermal response lag time is used to subsequently calculate the causal load time corresponding to the slow thermal field data.

[0102] Through this step, the cloud data center operation analysis platform obtains the thermal response lag time corresponding to different slow-heating field data; the thermal response lag time is used to represent the degree of lag response of the slow-heating field data relative to the fast load change, and is used in subsequent steps to determine the causal load time corresponding to the slow-heating field data.

[0103] S4. Within the current optimization calculation window used to generate energy-saving optimization instructions, determine the causal load time based on the thermal response lag time and the side sampling timestamp of the slow thermal field data, and compare the fast load state corresponding to the causal load time with the fast load state of the current optimization calculation window to identify the phase misalignment window and the phase misalignment type.

[0104] The cloud data center operation analysis platform establishes the current optimization calculation window according to the optimization cycle used to generate energy-saving optimization instructions; the current optimization calculation window refers to the set of side standard data frames used by the cloud data center operation analysis platform to form an energy-saving optimization instruction.

[0105] In this embodiment, the current optimization calculation window refers to the set of side standard data frames used by the cloud data center operation analysis platform when generating an energy-saving optimization instruction; the boundary of the current optimization calculation window is determined based on the cloud energy-saving optimization calculation cycle or the arrival batch of side standard data frames, but the data comparison within the current optimization calculation window is based on the side sampling timestamp, rather than the upload arrival time.

[0106] Within the current optimization calculation window, the cloud data center operation analysis platform obtains the edge sampling timestamps and corresponding thermal response lag times of the slow-heating field data; let the edge sampling timestamp of the j-th type of slow-heating field data be: The corresponding thermal response hysteresis time is: The causal loading time corresponding to this slow-heating field data is:

[0107] ;

[0108] in, This represents the time corresponding to the rapid load change actually reflected in the j-th type of slow heating field data;

[0109] In this embodiment, the causal load time refers to the actual time of the fast load change corresponding to a certain slow-heat field data. Because the slow-heat field data has a thermal response lag relative to the fast load change, the side-sampling timestamp of the slow-heat field data is not equivalent to its actual fast load time. The cloud data center operation analysis platform obtains the actual causal load time reflected by the slow-heat field data by subtracting the corresponding thermal response lag time from the side-sampling timestamp of the slow-heat field data. Figure 3 As shown, fast load data is used to characterize the rapid changes in heat-generating components such as IT load power, UPS output power, power distribution branch power, or cabinet power. It can quickly change from low to high when the load increases and from high to low when the load decreases. Slow thermal field data is used to characterize the thermal field response state such as cold aisle temperature, hot aisle temperature, return air temperature, or chilled water supply and return water temperature. Its changes are lagging behind the fast load data in terms of thermal response. After the load increases, if the cloud data center operation analysis platform directly splices the current high-level fast load data with the slow thermal field data that is still in the low response stage to the same operating state, a high load low thermal response misalignment window is formed. After the load decreases, if the current low-level fast load data is directly spliced ​​with the slow thermal field data that is still in the high response stage to the same operating state, a low load high thermal response misalignment window is formed. Figure 3 The thermal response lag time is used to represent the time difference between the moment of rapid load change and the moment of response of the corresponding slow thermal field, while the causal load time is used to represent the actual time of rapid load change corresponding to a certain slow thermal field observation data.

[0110] The cloud-based data center operations analytics platform obtains causal load times from the time-aligned fast load sequence. The corresponding fast load state is used as the actual load stage reflected by the slow heat field data; the cloud data center operation analysis platform also obtains the current fast load state within the current optimization calculation window.

[0111] The current fast load status is determined by the time-aligned fast load data within the current optimization calculation window, specifically represented by the average, median, or terminal stable value of the fast load data within the current optimization calculation window; the fast load status corresponding to the causal load time is determined by the fast load data of the time-aligned fast load sequence at the causal load time or at sampling points adjacent to the causal load time; to avoid misjudgment caused by acquisition noise, the cloud data center operation analysis platform determines the comparison tolerance based on the natural fluctuation range of the change in the fast load sequence at adjacent sampling times in historical normal operation data;

[0112] If the current fast load state is higher than the fast load state corresponding to the causal load time and exceeds the comparison tolerance, then the fast load state of the current optimization calculation window is determined to be in an increasing state relative to the fast load state corresponding to the causal load time; if the current fast load state is lower than the fast load state corresponding to the causal load time and exceeds the comparison tolerance, then the fast load state of the current optimization calculation window is determined to be in a decreasing state relative to the fast load state corresponding to the causal load time; if the difference between the two does not exceed the comparison tolerance, then it is determined that there is no effective change between them for identifying misaligned splicing windows.

[0113] In this embodiment, the comparison tolerance refers to the comparison range used to exclude the influence of noise or short-term jitter in fast load data acquisition; the comparison tolerance is determined by the natural fluctuation range of the change in the fast load sequence at adjacent sampling times in the historical normal operation data; if the difference between the current fast load state and the fast load state corresponding to the causal load time does not exceed the comparison tolerance, the cloud data center operation analysis platform will not use this difference as a valid change for identifying the misaligned splicing window; the comparison tolerance is adaptively determined based on the current historical normal operation data of the data center.

[0114] The cloud data center operation analysis platform compares the fast load state corresponding to the causal load time with the fast load state of the current optimized calculation window. When the fast load state of the current optimized calculation window is in an increasing state relative to the fast load state corresponding to the causal load time, it indicates that the current fast load has increased, while the current slow heat field data still corresponds to an earlier load stage. The cloud data center operation analysis platform identifies the current optimized calculation window as a high load, low heat response, misaligned window.

[0115] When the fast load state of the current optimized computing window is reduced relative to the fast load state corresponding to the causal load time, it means that the current fast load has been reduced, while the current slow heat field data still corresponds to an earlier high load stage. The cloud data center operation analysis platform identifies the current optimized computing window as a low load high heat response misphase window.

[0116] When the difference between the current fast load state and the fast load state corresponding to the causal load time does not exceed the comparison tolerance, or when the thermal response trend corresponding to the current fast load state and the measured state of the slow thermal field conforms to the thermal response lag relationship in the historical normal operation data, the cloud data center operation analysis platform will mark the current optimization calculation window as a fault-free phase window.

[0117] In this embodiment, the phase misalignment window refers to an optimization calculation window where the fast load state and the causal load stage actually corresponding to the slow thermal field data within the current optimization calculation window are inconsistent. The phase misalignment window includes a high load low thermal response phase misalignment window and a low load high thermal response phase misalignment window. The high load low thermal response phase misalignment window indicates that the current fast load state has increased, but the slow thermal field data still corresponds to the thermal response stage before or at the initial stage of the increase. The low load high thermal response phase misalignment window indicates that the current fast load state has decreased, but the slow thermal field data still corresponds to the high load thermal response stage before the decrease. The phase misalignment type is used to distinguish the above different phase misalignment windows and serves as a constraint basis for generating cloud optimization states and energy-saving optimization instructions.

[0118] Through this step, the cloud data center operation analysis platform outputs the phase misalignment splicing window and the phase misalignment type; the phase misalignment type includes high load low thermal response phase misalignment type, low load high thermal response phase misalignment type and no phase misalignment type.

[0119] S5. Based on the current fast load status, slow heat field measured status, causal load time and phase error type, generate a cloud optimization status that includes fast load status, slow heat field status and phase error type identifiers.

[0120] After identifying the phase misalignment splicing window and the phase misalignment type, the cloud data center operation analysis platform does not directly input the fast load data and slow heat field data in the current optimization calculation window as the operating status at the same time into the energy-saving optimization calculation. Instead, it generates the cloud optimization status based on the current fast load status, the actual slow heat field status, the causal load time, and the phase misalignment type.

[0121] The cloud-based optimization status is structured status data, which includes at least fast load status, slow thermal field status, and phase misalignment type identifier;

[0122] In this embodiment, the cloud optimization state refers to the structured state data formed by the cloud data center operation analysis platform before generating energy-saving optimization instructions. The cloud optimization state includes at least the following fields: fast load state field, slow heat field state field, phase error type identifier field, causal load time field, thermal response lag time field, and energy-saving optimization constraint field. The fast load state field records the heat generation state corresponding to the fast load data within the current optimization calculation window. The slow heat field state field records the thermal field response state corresponding to the slow heat field data within the current optimization calculation window. The phase error type identifier field records the phase error type of the current optimization calculation window. The causal load time field records the fast load stage actually corresponding to the slow heat field data. The thermal response lag time field records the lag degree of the slow heat field data relative to the fast load change. The energy-saving optimization constraint field indicates whether the energy optimization model allows the generation of energy-saving optimization instructions that weaken cooling capacity, or whether it allows the generation of energy-saving optimization instructions that improve cooling capacity based solely on the measured state of the slow heat field. By setting the above fields, the cloud optimization state not only records the current data value, but also records the thermal causal relationship and energy-saving optimization constraint relationship between the fast load data and the slow heat field data.

[0123] Fast load status indicates the heat generation status of the data center corresponding to the fast load data within the current optimization calculation window; slow thermal field status indicates the thermal field response status of the data center corresponding to the slow thermal field data within the current optimization calculation window; the phase mismatch type identifier is used to characterize whether there is a phase mismatch relationship between the fast load data and the slow thermal field data within the current optimization calculation window.

[0124] When the phase error type is a high load low thermal response phase error window, the cloud data center operation analysis platform generates a cloud optimization state that includes the current fast load increase state, the slow thermal field state that lags behind the current fast load increase state, and the high load low thermal response phase error type identifier. This cloud optimization state indicates that the current fast load increase has occurred, but the slow thermal field data has not yet completed the thermal response corresponding to the current fast load increase state.

[0125] When the phase error type is a low load high thermal response phase error window, the cloud data center operation analysis platform generates a cloud optimization state that includes the current fast load reduction state, the slow thermal field lagging behind the current fast load reduction state, and the low load high thermal response phase error type identifier. This cloud optimization state indicates that the current fast load reduction has occurred, but the slow thermal field data is still in a lagging thermal response state in an earlier high load stage.

[0126] When the current optimization calculation window is not identified as a misphase splicing window, the cloud data center operation analysis platform generates a cloud optimization state that includes the current fast load state, the slow heat field measured state, and no misphase type identifier. This cloud optimization state indicates that there is no misphase splicing relationship that needs to be corrected between the current fast load state and the slow heat field measured state, and can be used as a direct input state for generating energy-saving optimization instructions.

[0127] Through this step, the cloud data center operation analysis platform transforms the pseudo-operating state, which may have been formed by splicing data from different times, into a cloud-optimized state that includes a phase error type identifier. This enables subsequent energy-saving optimization instructions to identify different operating conditions, such as the heat peak not arriving, the heat peak being delayed, or the absence of a phase error.

[0128] S6. Generate energy-saving optimization instructions based on the cloud-based optimization status and send them to the side-side dynamic environmental group control management platform;

[0129] The cloud data center operation analysis platform inputs the cloud optimization status into the energy optimization model to generate energy-saving optimization instructions. The energy optimization model is either a rule-based optimization calculation program or an energy consumption optimization calculation program that combines the analysis results of the AI ​​service center. When generating energy-saving optimization instructions, the energy optimization model uses the phase error type identifier and energy-saving optimization constraint field in the cloud optimization status as constraint conditions.

[0130] In this embodiment, the candidate energy-saving optimization instruction set refers to the set of control instructions to be screened formed by the cloud data center operation analysis platform within the current optimization calculation window based on the current fast load status, slow thermal field measured status, and energy consumption level. Each candidate energy-saving optimization instruction in the candidate energy-saving optimization instruction set includes a target device identifier, control action type, and control execution time. The cloud data center operation analysis platform filters the candidate energy-saving optimization instruction set based on the phase misalignment type identifier and energy-saving optimization constraint field in the cloud optimization status to obtain the final energy-saving optimization instruction sent to the side-side dynamic environment group control management platform.

[0131] When the cloud-based optimization status includes a high-load, low-heat-response phase misalignment identifier, the cloud-based data center operation analysis platform constrains the candidate energy-saving optimization instruction set based on the energy-saving optimization constraint field. Candidate energy-saving optimization instructions that reduce the output capacity of chiller units, increase the set temperature of precision air conditioners, reduce the number of precision air conditioners in operation, or reduce the response capability of the refrigeration system will be marked as non-output instructions and excluded from the candidate energy-saving optimization instruction set in the current optimization calculation window. This prevents the current optimization calculation window from outputting energy-saving optimization instructions that weaken the cooling capacity. In this way, the cooling output is prevented from being weakened before the heat peak arrives.

[0132] When the cloud-based optimization status includes a low-load, high-heat response phase error type identifier, the cloud-based data center operation analysis platform constrains the candidate energy-saving optimization instruction set based on the energy-saving optimization constraint field. If a candidate energy-saving optimization instruction is triggered only by the measured high state of the slow-heat field and is not verified in conjunction with the current fast-load reduction state and causal load time, then the candidate energy-saving optimization instruction is marked as an instruction that cannot be directly output and is not output as an energy-saving optimization instruction in the current optimization calculation window. In this way, the lagging high state of the slow-heat field is avoided from being mistaken for the current continuous high heat load and thus increasing the cooling capacity.

[0133] When the cloud-based optimization status includes a no-error-phase-type identifier, the cloud-based data center operation analysis platform generates air conditioning set temperature adjustment instructions, chiller load allocation instructions, precision air conditioning operation mode switching instructions, or environmental monitoring equipment energy-saving operation instructions based on the current fast load status, slow heat field measured status, and energy consumption level. The environmental monitoring equipment energy-saving operation instructions include instructions to perform standby, low-power operation, operation mode switching, or operation period adjustment on environmental monitoring equipment that is allowed to operate at energy efficiency. The energy-saving optimization instructions at least include a target device identifier, a control action type, and a control execution time. The target device identifier is used to determine the environmental monitoring equipment that needs to perform the control action. The control action type is used to characterize air conditioning set temperature adjustment, chiller load allocation, precision air conditioning operation mode switching, or environmental monitoring equipment energy-saving operation. The control execution time is used to limit the time for the side-side environmental monitoring group control management platform to execute the energy-saving optimization instruction.

[0134] After generating energy-saving optimization instructions, the cloud data center operation analysis platform sends the instructions to the edge environmental control management platform through the network transmission layer.

[0135] S7, the side dynamic environmental group control management platform executes energy-saving optimization commands and feeds back fast load data and slow thermal field data after execution. The cloud data center operation analysis platform recalculates the thermal response lag time and updates the misphase splicing window identification results accordingly.

[0136] After receiving energy-saving optimization instructions from the cloud-based data center operation analysis platform, the edge-side environmental control management platform verifies the executable status of the corresponding environmental devices. The executable status includes at least one of the following: device communication status, device alarm status, device protection status, device local manual status, and device operating mode status. Device communication status indicates whether the edge-side environmental control management platform can communicate with the target device; device alarm status indicates whether the target device is in a fault or abnormal alarm state; device protection status indicates whether the target device is in a protection action state; device local manual status indicates whether the target device has been manually controlled locally; and device operating mode status indicates whether the target device's current operating mode allows the execution of the cloud-based energy-saving optimization instructions.

[0137] In this embodiment, the candidate control device set refers to the set of environmental monitoring devices that the cloud data center operation analysis platform can select as target devices when generating energy-saving optimization instructions. The candidate control device set consists of environmental monitoring devices that are reported as being in an executable state by the side-side environmental monitoring group control management platform. If a certain environmental monitoring device reports an execution result due to communication abnormality, device alarm, device protection, local manual takeover, or operating mode that does not allow execution, the cloud data center operation analysis platform will remove the environmental monitoring device from the candidate control device set when generating the next round of energy-saving optimization instructions. After the side-side environmental monitoring group control management platform reports that the environmental monitoring device has recovered to an executable state, it will be restored to the candidate control device set.

[0138] When the corresponding environmental monitoring equipment is in an executable state, the side-side environmental monitoring group control management platform executes the corresponding control action according to the target equipment identifier, control action type, and control execution time in the energy-saving optimization instruction; the control action includes adjusting the air conditioner set temperature, distributing the chiller load, switching the precision air conditioner operating mode, or energy-saving operation of the environmental monitoring equipment; after execution, the side-side environmental monitoring group control management platform feeds back the execution time, execution result, fast load data after execution, and slow heat field data after execution to the cloud data center operation analysis platform;

[0139] When the corresponding environmental control device is not in an executable state, the side-side environmental control group management platform will not execute the energy-saving optimization command and will report the non-execution result to the cloud data center operation analysis platform; the non-execution result includes at least the target device identifier and the reason for non-execution; the reason for non-execution includes one of the following: communication abnormality, device alarm, device protection, local manual takeover, or the operating mode does not allow execution;

[0140] After receiving the fast load data and slow thermal field data after execution, the cloud data center operation analysis platform uses them as one of the historical operation data sources for determining the thermal response lag time in the next round, and recalculates the thermal response lag time according to step S3; the recalculated thermal response lag time is used to determine the causal load time in the next round.

[0141] The cloud-based data center operation analysis platform also updates the phase misalignment window identification results based on the fast load data, slow thermal field data, side sampling timestamps, and recalculated thermal response lag time after execution. Specifically, the cloud-based data center operation analysis platform determines the causal load time corresponding to the slow thermal field data after execution based on the recalculated thermal response lag time, and compares the fast load state corresponding to this causal load time with the fast load state of the current optimization calculation window in the next round to re-identify high load low thermal response phase misalignment windows, low load high thermal response phase misalignment windows, or no phase misalignment windows. Thus, the update of the phase misalignment window identification results comes from the fast load data and slow thermal field data after execution, rather than from the unexecuted results.

[0142] When the cloud-based data center operations analysis platform receives an unexecuted result, it uses this result to adjust the device's executable status when generating the next round of energy-saving optimization instructions. Specifically, if the reason for the unexecuted result is communication abnormality, device alarm, device protection, local manual takeover, or the operating mode does not allow execution, the cloud-based data center operations analysis platform will remove the corresponding target device from the candidate control device set when generating the next round of energy-saving optimization instructions, and will no longer generate energy-saving optimization instructions that need to be executed immediately for that target device. After the side-side dynamic environmental group control management platform reports that the target device has recovered to an executable state, the cloud-based data center operations analysis platform will restore the target device to the candidate control device set. In this way, the unexecuted result will not be mistakenly used to correct the thermal response lag time, but will instead be used to constrain the range of executable devices for the next round of energy-saving optimization instructions.

[0143] Through the above steps, this embodiment can identify the phase misalignment state formed by the differences in acquisition cycle, side-side buffering and retransmission, network transmission latency, and cloud analysis time during the edge-cloud collaborative optimization of data center dynamic environment energy consumption. It also forms a cloud optimization state containing a phase misalignment type identifier before generating energy-saving optimization instructions, thereby preventing the cloud data center operation analysis platform from generating energy-saving optimization instructions based on pseudo-operational states. For high-load, low-thermal-response phase misalignment windows, this embodiment can avoid prematurely reducing cooling output before the heat peak arrives. For low-load, high-thermal-response phase misalignment windows, this embodiment can avoid increasing cooling capacity solely based on the lagging high-level state of the slow thermal field. For windows without phase misalignment, this embodiment can generate normal energy-saving optimization instructions based on the current fast-load state, the measured state of the slow thermal field, and the energy consumption level.

[0144] In a specific operational example, the data center is equipped with smart meters, UPS monitoring units, power distribution branch meters, cabinet power acquisition devices, cold aisle temperature sensors, hot aisle temperature sensors, return air temperature sensors, chilled water supply and return water temperature sensors, and precision air conditioning operation status acquisition units; the side-side dynamic environment group control management platform is deployed on the local industrial control computer in the data center, and the cloud data center operation analysis platform is deployed on the cloud server;

[0145] The side-side dynamic environment group control management platform collects IT load power, UPS output power, power distribution branch power, and cabinet power as fast load data; it also collects cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status as slow heat field data; the side-side dynamic environment group control management platform adds a side-side sampling timestamp, equipment identifier, and data type to each data point to form the original dynamic environment dataset;

[0146] When the network is normal, the edge-side environmental control management platform uploads the edge-side standard data frames to the cloud data center operation and analysis platform; when the network is interrupted, the edge-side environmental control management platform caches the collected data; after the network is restored, the edge-side environmental control management platform retransmits the cached data to the cloud data center operation and analysis platform, writes a cache retransmission identifier to the retransmitted data, and retains the original edge-side sampling timestamp of the retransmitted data.

[0147] After receiving the standard data frames from the edge, the cloud-based data center operation analysis platform restores the sampling time sequence according to the edge sampling timestamps, extracts the fast load sequence and slow thermal field sequence from the historical normal operation data, and calculates the thermal response lag time of cold aisle temperature, hot aisle temperature, return air temperature, and chilled water supply and return water temperature relative to the fast load change through cross-correlation calculation. For the supply and return air status of precision air conditioning, the cloud-based data center operation analysis platform first converts it into numerical state variables corresponding to supply air temperature, return air temperature, fan frequency, valve opening degree, or operating level. Supply and return air status that cannot be converted into numerical state variables does not participate in the thermal response lag time calculation, but is only used for equipment state constraints when generating subsequent energy-saving optimization instructions.

[0148] Within a certain current optimization calculation window, if the cloud data center operation analysis platform finds that the current fast load state is in an elevated state relative to the fast load state corresponding to the causal load time, then the current optimization calculation window is identified as a high load low thermal response misphase window. The cloud data center operation analysis platform generates a cloud optimization state that includes the current fast load elevated state, the slow thermal field state that lags behind the current fast load elevated state, and the high load low thermal response misphase type identifier, and restricts the generation of energy-saving optimization instructions that weaken cooling capacity.

[0149] Within another current optimization calculation window, if the cloud data center operation analysis platform finds that the current fast load state is in a reduced state relative to the fast load state corresponding to the causal load time, then it identifies the current optimization calculation window as a low load high thermal response misphase window. The cloud data center operation analysis platform generates a cloud optimization state that includes the current fast load reduced state, the slow thermal field state that lags behind the current fast load reduced state, and the low load high thermal response misphase type identifier, and restricts the generation of energy-saving optimization instructions that only improve cooling capacity based on the measured state of the slow thermal field.

[0150] When the difference between the current fast load state and the fast load state corresponding to the causal load time does not exceed the comparison tolerance, or when the thermal response trend corresponding to the current fast load state and the measured state of the slow thermal field conforms to the thermal response lag relationship in the historical normal operation data, the cloud data center operation analysis platform will identify the current optimization calculation window as a fault-free phase window. For fault-free phase windows, the cloud data center operation analysis platform will generate a cloud optimization state that includes the current fast load state, the measured state of the slow thermal field, and the fault-free phase type identifier. Based on the current fast load state, the measured state of the slow thermal field, and the energy consumption level, the platform will generate air conditioning set temperature adjustment instructions, chiller load distribution instructions, precision air conditioning operation mode switching instructions, or environmental protection equipment energy-saving operation instructions.

[0151] After receiving an energy-saving optimization command, the side-side environmental control group management platform first verifies whether the corresponding environmental device is in an executable state. If the corresponding environmental device is in an executable state, the energy-saving optimization command is executed, and the execution time, execution result, post-execution fast load data, and post-execution slow thermal field data are fed back to the cloud data center operation analysis platform. If the corresponding environmental device is not in an executable state, the energy-saving optimization command is not executed, and the non-execution result is fed back to the cloud data center operation analysis platform. The cloud data center operation analysis platform recalculates the thermal response lag time based on the post-execution fast load data and post-execution slow thermal field data and updates the phase misalignment splicing window identification result. At the same time, it adjusts the device's executable state when the next round of energy-saving optimization commands is generated based on the non-execution result.

[0152] In other implementations, fast load data can be selected from one or more of the following based on the actual deployment of the data center: IT load power, UPS output power, power distribution branch power, and cabinet power, without requiring all of them to be collected simultaneously; slow hot field data can be selected from one or more of the following based on the type of cooling system: cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status.

[0153] In other implementations, the edge-side environmental control group management platform is deployed on a local server, industrial control computer, edge gateway, or edge computing node in the computer room; the network transmission layer adopts Ethernet, 4G, 5G, leased line, MQTT, HTTP, or other communication methods suitable for data transmission in the environmental system;

[0154] In other implementations, the thermal response lag time is determined by cross-correlation calculation or by a time delay estimation algorithm that can characterize the time delay relationship between rapid load changes and slow thermal field response; regardless of the time delay estimation method used, time alignment is performed based on the side sampling timestamp and the calculation is based on historical normal operation data;

[0155] In other implementations, the energy optimization model is an optimization calculation program based on rule constraints, or an optimization calculation program that combines the output results of the AI ​​service center. When generating energy-saving optimization instructions, the energy optimization model uses the phase misalignment type identifier in the cloud optimization status as a constraint condition to avoid the incorrect direction of the energy-saving optimization instructions caused by the phase misalignment splicing of fast load data and slow thermal field data.

[0156] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for collaborative optimization of energy consumption in a data center's environmental conditions and cloud environments, characterized in that... Includes the following steps: The side-side dynamic environment group control management platform collects fast load data and slow thermal field data of the data center dynamic environment system, adds side-side sampling timestamps, and generates the original dynamic environment dataset; The original dynamic environment dataset is cleaned, missing data is marked, and units are standardized. The original edge sampling timestamp is retained and written into the cached data transmission identifier. The edge standard data frame is generated and uploaded to the cloud data center operation and analysis platform. The cloud data center operation analysis platform uses the edge sampling timestamp as a benchmark to perform time alignment on the edge standard data frame, extracts the fast load sequence and slow thermal field sequence from the time-aligned edge standard data frame, and determines the thermal response lag time through the cross-correlation between the fast load sequence and the slow thermal field sequence. Within the current optimization calculation window used to generate energy-saving optimization instructions, the causal load time is determined based on the thermal response lag time and the side sampling timestamp of the slow thermal field data. The fast load state corresponding to the causal load time is compared with the fast load state of the current optimization calculation window to identify the phase misalignment window and the phase misalignment type. Based on the current fast load status, slow heat field measured status, causal load time, and phase error type, generate a cloud-based optimization status that includes fast load status, slow heat field status, and phase error type identifiers; Energy-saving optimization instructions are generated based on the cloud-based optimization status and sent to the side-side dynamic environmental group control management platform. The side-side dynamic environment group control management platform executes energy-saving optimization commands and feeds back fast load data and slow thermal field data after execution to update the identification results of the phase misalignment splicing window.

2. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 1, characterized in that, The fast load data includes at least one of IT load power, UPS output power, power distribution branch power, and cabinet power; the slow hot field data includes at least one of cold aisle temperature, hot aisle temperature, return air temperature, chilled water supply and return water temperature, and precision air conditioning supply and return air status.

3. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 1, characterized in that, The edge standard data frame includes a data value, a device identifier, a data type, an edge sampling timestamp, and a cache retransmission identifier. The data value originates from fast load data or slow hot field data, the device identifier is used to determine the data source device, the data type is used to distinguish between fast load data and slow hot field data, and the cache retransmission identifier is used to distinguish between real-time uploaded data and cache retransmission data.

4. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 1, characterized in that, The cloud-based data center operation and analysis platform performs time alignment of the edge-side standard data frames based on the edge-side sampling timestamp, including: Identify cached data to be re-transmitted based on the cache re-transmission identifier; The real-time uploaded data and cached re-uploaded data are restored to the sampling time order according to the side sampling timestamp; Data with consecutive sampling times and no missing markers are used as historical normal operation data; Extract fast load sequences and slow thermal field sequences from historical normal operation data.

5. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 4, characterized in that, The determination of thermal response hysteresis time through the cross-correlation of the fast load sequence and the slow thermal field sequence includes: Calculate the change in the fast load sequence between adjacent sampling times; Calculate the change of the slow-heating field sequence under candidate lag times; Calculate the correlation between the changes in the fast load sequence and the changes in the slow thermal field sequence at different candidate lag times; The candidate lag time when the correlation reaches its maximum is determined as the thermal response lag time of the corresponding slow-heat field data. The candidate lag time is determined by the actual sampling period of the corresponding slow-heat field data and the historical normal operation data.

6. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 1, characterized in that, The determination of the causal load time includes: Obtain the side sampling timestamps and corresponding thermal response hysteresis times of the slow-heating field data; Subtract the corresponding thermal response hysteresis time from the side sampling timestamp of the slow-heating field data to obtain the causal load time corresponding to the slow-heating field data. In the time-aligned fast load sequence, the fast load state corresponding to the causal load time is obtained, and this fast load state is taken as the load stage actually reflected by the slow thermal field data.

7. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 6, characterized in that, The identification of the phase misalignment splicing window and the phase misalignment type includes: Compare the fast load state corresponding to the causal load time with the fast load state of the current optimization calculation window; When the fast load state of the current optimization calculation window is in an elevated state relative to the fast load state corresponding to the causal load time, the current optimization calculation window is identified as a high load low thermal response out-of-phase window. When the fast load state of the current optimization calculation window is in a reduced state relative to the fast load state corresponding to the causal load time, the current optimization calculation window is identified as a low load high thermal response out-of-phase window.

8. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 7, characterized in that, The generation of the cloud-based optimization status includes: When the phase error type is a high load low thermal response phase error window, a cloud-optimized state is generated that includes the current fast load increase state, the slow thermal field lagging behind the current fast load increase state, and the high load low thermal response phase error type identifier. When the phase error type is a low load high thermal response phase error window, a cloud-optimized state is generated that includes the current fast load reduction state, the slow thermal field lagging behind the current fast load reduction state, and the low load high thermal response phase error type identifier. When the current optimization calculation window is not identified as a misphase splicing window, a cloud optimization status is generated that includes the current fast load status, the actual slow thermal field status, and no misphase type identifier.

9. A data center environmental energy consumption edge-cloud collaborative optimization method according to claim 8, characterized in that, The process of generating energy-saving optimization instructions based on cloud-based optimization status includes: When the cloud-based optimization status contains a high-load, low-thermal-response phase-out type identifier, the generation of energy-saving optimization instructions that weaken cooling capacity is restricted. When the cloud-based optimization status contains a low-load, high-heat-response phase-out type identifier, the generation of energy-saving optimization instructions that only improve cooling capacity based on the actual measured status of the slow-heat field is restricted. When the cloud-based optimization status includes a no-error-phase-type identifier, it generates air conditioning set temperature adjustment commands, chiller load distribution commands, precision air conditioning operation mode switching commands, or environmental protection equipment energy-saving operation commands based on the current fast load status, slow heat field measured status, and energy consumption level.

10. The data center environmental energy consumption edge-cloud collaborative optimization method according to claim 1, characterized in that, The side-side dynamic environment group control management platform executes energy-saving optimization commands and feeds back fast load data and slow thermal field data after execution, including: After receiving the energy-saving optimization command, the side-side dynamic environment group control management platform verifies the executable status of the corresponding dynamic environment equipment; When the corresponding environmental equipment is in an executable state, the energy-saving optimization command is executed, and the execution time, execution result, fast load data after execution, and slow heat field data after execution are fed back to the cloud data center operation analysis platform. When the corresponding environmental equipment is not in an executable state, the energy-saving optimization command will not be executed, and the non-execution result will be reported to the cloud data center operation analysis platform; The cloud-based data center operation analysis platform recalculates the thermal response lag time based on the fast load data and slow thermal field data after execution, and uses the recalculated thermal response lag time to determine the next round of causal load time. The cloud-based data center operation analysis platform updates the misphase splicing window identification results based on the fast load data and slow heat field data after execution, and adjusts the device's executable status when the next round of energy-saving optimization instructions is generated based on the unexecuted results.

Images