Method, Apparatus, and Program for Targeted Air Conditioning Control of a Data Center Based on Prediction of 3D Thermal Volume

By integrating 3D thermal volume data and power load data, the method predicts heat island zones and optimizes air conditioning control in data centers, addressing inefficiencies and turbulence in conventional HVAC systems.

KR103044703B1Active Publication Date: 2026-11-11MPOLE CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
KR1020260072070
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-04-21
Publication Date
2026-11-11
Estimated Expiration
2046-04-21

AI Technical Summary

Technical Problem

Conventional HVAC control methods in data centers fail to accurately reflect heat distribution and power load changes over time, leading to over- or delayed air conditioning operations, inefficient energy use, and turbulence from simultaneous system activations, without predicting localized heat concentration or urban heat islands.

Method used

A method that combines three-dimensional thermal volume data and server power load data to predict heat island occurrence zones, calculate dynamic thresholds, and perform pre-activation and asymmetric cooperative control of air conditioning equipment.

Benefits of technology

Enables precise prediction and proactive control of heat distribution, reducing energy inefficiencies and airflow turbulence by dynamically adjusting air conditioning operations based on 3D thermal data and power load fluctuations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026048794026-PAT00009_ABST
    Figure 112026048794026-PAT00009_ABST
Patent Text Reader

Abstract

A data center targeting air conditioning control method based on three-dimensional thermal volume prediction according to various embodiments of the present invention is disclosed. The method may include the steps of: collecting three-dimensional thermal volume data and server power load data to construct a time-series multidimensional data tensor; inputting the data tensor into an artificial intelligence model to predict temperature changes, calculate dynamic thresholds, and identify areas where a heat island is expected to occur; calculating a local cooling load based on the difference between the predicted temperature and the target temperature for the areas where a heat island is expected to occur; identifying air conditioning equipment based on the coordinates of the areas where a heat island is expected to occur and calculating a pre-operation time; and performing cooperative control to drive the air conditioning equipment according to the pre-operation time and asymmetrically control the airflow or airflow direction.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a data center targeting air conditioning control method, apparatus, and program based on three-dimensional thermal volume prediction. More specifically, it relates to a method, apparatus, and program that predicts heat island occurrence zones by combining three-dimensional thermal volume data and server power load data, calculates dynamic thresholds and local cooling loads, and performs pre-operation of air conditioning equipment and asymmetric cooperative control. Background Technology

[0003] Recently, Building Management Systems (BMS) have been evolving toward the integrated management of various facilities and sensors for efficient energy management and facility maintenance. In particular, HVAC systems are recognized as a key component accounting for a significant portion of energy consumption in facility spaces. Accordingly, technologies that control HVAC systems based on environmental data, such as temperature, humidity, and airflow within the facility space, are being widely utilized.

[0004] Meanwhile, with the recent introduction of high-performance computing environments leading to increased heat density and power load variability in server equipment within data centers, the importance of technology capable of precisely identifying heat distribution within a space and actively controlling HVAC systems based on this is becoming increasingly prominent. However, conventional HVAC control methods operate primarily based on temperature sensors at specific locations or thermal imaging data from a single point in time, and thus have limitations in that they fail to adequately reflect the heat distribution across the entire space and the characteristics of changes over time.

[0005] In particular, conventional technology primarily applies fixed threshold-based control that activates air conditioning systems when a certain reference temperature is exceeded. Consequently, this fails to reflect the dynamic heat generation characteristics that change according to variations in the power load of server equipment, which can lead to over- or delayed operation of the air conditioning system. Furthermore, there is a problem in that it fails to predict localized heat concentration or the urban heat island effect in specific areas in advance, and relies on passive control methods that respond only after the temperature has risen.

[0006] Furthermore, in conventional technology, air conditioning systems within a space often operate collectively across the entire area, failing to selectively respond to localized areas that actually require cooling; consequently, this can lead to a decrease in energy efficiency. In particular, when multiple air conditioning systems operate simultaneously, turbulence caused by airflow collisions also occurs, resulting in reduced cooling efficiency.

[0007] Furthermore, conventional technology fails to comprehensively consider temperature data within the space and IT power load data collected from server equipment, and has limitations in predictive control that reflects the correlation between temperature and load changes over time. Consequently, it is difficult to determine the potential for urban heat islands in advance and to preemptively control air conditioning systems.

[0008] Therefore, there is a need for a new type of air conditioning control technology that can comprehensively analyze heat distribution and power load characteristics based on three-dimensional thermal volume data in facility spaces such as data centers, predict heat island areas in advance, make judgments based on dynamic thresholds, and perform pre-activation and asymmetric cooperative control of local air conditioning equipment. Prior art literature

[0010] Republic of Korea Published Patent Application No. 10-2020-0072765 The problem to be solved

[0011] The problem that the present invention aims to solve is to provide a method, apparatus, and program that combines three-dimensional thermal volume data and server power load data to predict heat island occurrence zones, calculate dynamic thresholds and local cooling loads, and perform pre-operation of air conditioning equipment and asymmetric cooperative control.

[0012] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below. means of solving the problem

[0014] A data center targeting air conditioning control method based on three-dimensional volume thermal prediction according to an embodiment of the present invention is disclosed to solve the aforementioned problem. The method comprises the steps of: configuring a time-series multidimensional data tensor by collecting three-dimensional volume thermal data generated by dividing a three-dimensional space into a plurality of three-dimensional cube (voxel) units and IT power load data collected from server equipment corresponding to the three-dimensional cubes; inputting the time-series multidimensional data tensor into an artificial intelligence model to predict future temperature changes for each three-dimensional cube area, calculating a dynamic threshold that is variably set by reflecting the hardware limit temperature and power load fluctuation rate of the server equipment, and identifying an area predicted to exceed the dynamic threshold as a heat island occurrence zone; and for the heat island occurrence zone, calculating the difference between the predicted temperature and the target maintenance temperature, and calculating a local cooling load by reflecting the time-series temperature change characteristics of the three-dimensional volume thermal data. The method may include: a step of identifying a plurality of adjacent air conditioning facilities based on the three-dimensional spatial coordinates of the heat island-prone area, and calculating a pre-operation time (T-minus) by reflecting the mechanical operation delay time of each air conditioning facility and the thermal inertia coefficient of the space; and a step of pre-operating the selected air conditioning facility according to the pre-operation time, and performing cooperative control to suppress the generation of turbulence caused by collision between airflows by asymmetrically controlling the air volume or direction of each air conditioning facility when a plurality of air conditioning facilities operate simultaneously.

[0015] In an alternative embodiment, the step of constructing the time-series multidimensional data tensor may include: collecting temperature data for each 3D cube; and collecting IT power load data collected from server equipment corresponding to the spatial coordinates of the 3D cube and combining them into a multidimensional array arranged in chronological order.

[0016] In an alternative embodiment, the step of identifying the heat island-prone zone may include: setting a reference threshold temperature using the maximum allowable temperature and current temperature of the server equipment corresponding to each three-dimensional cube; and variably adjusting the reference threshold temperature according to the rate of change in the power load of the server equipment.

[0017] In an alternative embodiment, the step of identifying the heat island-prone zone may include: a step of predicting temperature changes over time using an artificial intelligence model that takes as input temperature data and power load data of a 3D cube collected over multiple time points; and a step of identifying a high-temperature area that is continuously formed in a specific zone based on the connectivity of the temperature distribution between adjacent 3D cubes.

[0018] In an alternative embodiment, the step of calculating the local cooling load may include: a step of calculating the difference between the predicted temperature and the target maintenance temperature; and a step of calculating the cooling requirement for each cube unit using the volume of the three-dimensional cube and the time-series temperature change value corresponding to the difference.

[0019] In an alternative embodiment, the step of back-calculating the prior operation time may include: a step of identifying air conditioning equipment within a certain distance based on the three-dimensional coordinates of the heat island-prone area; and a step of calculating the operation start time using time values ​​corresponding to the mechanical operation delay time and the temperature change rate of each air conditioning equipment.

[0020] In an alternative embodiment, the step of performing the coordinate control may include: a step of reducing the airflow of some of the air conditioning facilities or changing the airflow direction; and a step of maintaining or increasing the airflow of other air conditioning facilities to disperse the airflow so that the air conditioning airflows do not collide at the same point.

[0021] In an alternative embodiment, the step of performing the cooperative control may include: setting an airflow or cooling water control amount to a value corresponding to a cooling load for each of the plurality of air conditioning facilities; and adjusting the control value so that the airflow movement directions between the air conditioning facilities are formed in different directions.

[0022] According to another embodiment of the present invention, an apparatus for performing a data center targeting air conditioning control method based on three-dimensional thermal volume prediction is disclosed. The apparatus includes a memory for storing one or more instructions; and a processor for executing the one or more instructions stored in the memory, and the processor can perform the aforementioned three-dimensional thermal volume prediction-based data center targeting air conditioning control method by executing the one or more instructions.

[0023] According to another embodiment of the present invention, a computer program stored on a computer-readable recording medium is disclosed. The computer program is combined with a computer, which is hardware, to perform a data center targeting air conditioning control method based on three-dimensional thermal volume prediction.

[0024] Other specific details of the present invention are included in the detailed description and drawings. Effects of the invention

[0026] According to one embodiment of the present invention, by combining three-dimensional thermal volume data and IT power load data collected from server equipment to form a time-series multidimensional data tensor, it is possible to perform an analysis that simultaneously reflects the temperature distribution and equipment load status within the space.

[0027] In addition, by using an artificial intelligence model based on the above time series data to predict future temperature changes for each 3D cube zone, it is possible to reflect the trend of thermal change over time rather than simple temperature measurement.

[0028] In addition, by setting dynamically changing thresholds based on the maximum allowable temperature and power load change rate of the server equipment, risk zones can be determined according to environmental conditions rather than fixed criteria.

[0029] In addition, by calculating the local cooling load in three-dimensional cube units using the difference between the predicted temperature and the target maintenance temperature, the cooling requirement for a specific area within the space can be distinguished and calculated.

[0030] In addition, by identifying adjacent air conditioning equipment based on the 3D coordinates of the area where the heat island is expected to occur, and calculating a pre-operation time by reflecting the delay time of operation of the equipment and the thermal characteristics of the space, the air conditioning equipment can be controlled at a time before the temperature rises.

[0031] In addition, when multiple air conditioning systems operate simultaneously, it is possible to configure a control system that can reduce the generation of turbulence caused by collisions between airflows by performing cooperative control that asymmetrically adjusts the airflow volume or direction.

[0032] In addition, by reflecting the above air conditioning equipment control results back into the time series data, a structure can be formed in which the prediction and control processes are performed repeatedly.

[0033] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing

[0035] FIG. 1 is a drawing illustrating a system according to one embodiment of the present invention. FIG. 2 is a hardware configuration diagram of a server according to another embodiment of the present invention. FIG. 3 is a flowchart illustrating the entire process of a method for generating a three-dimensional volume based on physical-data fusion interpolation according to an embodiment of the present invention. FIG. 4 is a block diagram illustrating the overall configuration of a device for performing a three-dimensional thermal volume generation method based on physical-data fusion interpolation according to an embodiment of the present invention. FIG. 5 is a diagram illustrating the concept of cross-arrangement and three-dimensional spatial mapping of a multi-angle thermal imaging camera according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating the process of mapping three-dimensional solid cubes based on raycast-based two-dimensional temperature data according to one embodiment of the present invention. FIG. 7 is a flowchart illustrating the process of estimating the temperature of a blind spot using a physical-data fusion model according to one embodiment of the present invention. FIG. 8 is a diagram illustrating the concept of visualization of three-dimensional thermal volume data and air flow analysis according to one embodiment of the present invention. FIG. 9 is a flowchart illustrating the entire process of a data center targeting air conditioning control method based on three-dimensional thermal volume prediction according to an embodiment of the present invention. FIG. 10 is a block diagram illustrating the overall configuration of a device for performing an air conditioning control method using a three-dimensional thermal volume according to an embodiment of the present invention. FIG. 11 is a flowchart illustrating the process of combining equipment data with three-dimensional thermal volume data according to an embodiment of the present invention to form time-series data. FIG. 12 is a flowchart illustrating the process of calculating a dynamic threshold based on three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 13 is a flowchart illustrating the process of calculating a pre-operation time based on three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 14 is a diagram illustrating a concept of analyzing air flow distribution based on three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 15 is a flowchart illustrating the entire process of analyzing heat distribution based on three-dimensional thermal volume data according to one embodiment of the present invention. FIG. 16 is a diagram illustrating the concept of visualizing three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 17 is a flowchart illustrating the process of generating and processing three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 18 is a drawing illustrating an example of a visual representation of three-dimensional thermal volume data according to an embodiment of the present invention. FIG. 19 is a drawing illustrating an example of a user interface for selecting areas and facilities within a facility space based on three-dimensional thermal volume data according to an embodiment of the present invention. Specific details for implementing the invention

[0036] Various embodiments are now described with reference to the drawings. In this specification, various descriptions are provided to facilitate an understanding of the invention. However, it is evident that these embodiments can be practiced without such specific descriptions.

[0037] As used herein, terms such as “component,” “module,” “system,” etc. refer to computer-related entities, hardware, firmware, software, combinations of software and hardware, or executions of software. For example, a component may be a process, processor, object, execution thread, program, and / or computer running on a processor, but is not limited thereto. For example, both an application running on a computing device and the computing device itself may be a component. One or more components may reside within a processor and / or execution thread. A component may be localized within a single computer. A component may be distributed among two or more computers. Additionally, these components may be executed from various computer-readable media having various data structures stored therein. Components may communicate through local and / or remote processes, for example, according to signals having one or more data packets (e.g., data from a component interacting with another component in a local system or distributed system, and / or data transmitted through signals to other systems and networks such as the Internet).

[0038] Furthermore, the term "or" is intended to mean an implicit "or" rather than an exclusive "or." That is, unless otherwise specified or evident from the context, "X uses A or B" is intended to mean one of the natural implicit substitutions. In other words, if X uses A; if X uses B; or if X uses both A and B, "X uses A or B" may apply to any of these cases. Additionally, the term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the enumerated related items.

[0039] Additionally, the terms “comprising” and / or “comprising” should be understood to mean that such features and / or components are present. However, the terms “comprising” and / or “comprising” should be understood not to exclude the presence or addition of one or more other features, components and / or groups thereof. Furthermore, unless otherwise specified or clearly evident from the context to indicate a singular form, the singular in this specification and claims should generally be interpreted to mean “one or more.”

[0040] Those skilled in the art should recognize that the various exemplary logical blocks, configurations, modules, circuits, means, logics, and algorithmic steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or a combination of both. To clearly exemplify the interchangeability of hardware and software, various exemplary components, blocks, configurations, means, logics, modules, circuits, and steps have been generally described above in terms of their functionality. Whether such functionality is implemented in hardware or software depends on the specific application and design constraints imposed on the overall system. Skilled technicians may implement the described functionality in various ways for each specific application. However, such decisions regarding implementation should not be construed as moving out of the scope of the invention.

[0041] The description of the presented embodiments is provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art. The general principles defined herein may be applied to other embodiments without departing from the scope of the present invention. Thus, the present invention is not limited to the embodiments presented herein. The present invention should be interpreted in the broadest possible scope consistent with the principles and novel features presented herein.

[0042] In this specification, the term "computer" refers to any type of hardware device comprising at least one processor, and may be understood to include software configurations operating on said hardware device according to the embodiments. For example, the term "computer" may be understood to include smartphones, tablet PCs, desktops, laptops, and user clients and applications running on each of these devices, but is not limited thereto.

[0043] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0044] Each step described in this specification is described as being performed by a computer, but the subject of each step is not limited thereto, and depending on the embodiment, at least some of each step may be performed on different devices.

[0046] FIG. 1 is a drawing illustrating a system according to one embodiment of the present invention.

[0047] Referring to FIG. 1, a system according to one embodiment of the present invention may include a server (100), a user terminal (200), and an external server (300).

[0048] Here, the system illustrated in FIG. 1 is according to one embodiment, and its components are not limited to the embodiment illustrated in FIG. 1 and may be added, changed, or deleted as needed.

[0049] In one embodiment, the server (100) can perform the method according to the disclosed embodiment.

[0050] For example, the server (100) can be connected to a user terminal (200) through a network (400) and can provide a UI for the use of the method according to the disclosed embodiment to the user terminal (200), and can receive commands through the UI and output an analysis result therefrom.

[0051] In various embodiments, the server (100) may provide Web or Application-based services. However, it is not limited thereto.

[0052] Here, the user terminal (200) includes a display on at least a part of the user terminal (200) and may include an operating system for running an application or extension-based service provided by the server (100). For example, the user terminal (200) may be a smartphone, but is not limited thereto, and the user terminal (200) may include all kinds of handheld-based wireless communication devices such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartpad, tablet PC, etc., as a wireless communication device that ensures portability and mobility.

[0053] Additionally, the network (400) may refer to a connection structure capable of exchanging information between each node, such as multiple terminals and servers. For example, the network (400) includes a Local Area Network (LAN), a Wide Area Network (WAN), the World Wide Web (WWW), a wired and wireless data network, a telephone network, a wired and wireless television network, etc.

[0054] In addition, wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), 5GPP (5th Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), RF (Radio Frequency), Bluetooth network, NFC (Near-Field Communication) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, etc.

[0055] In one embodiment, an external server (300) may be connected to a server (100) through a network (400), and may store and manage various information / data necessary for the server (100) to perform analysis, and may store and manage various information / data generated as the server (100) provides services.

[0056] In various embodiments, the external server (300) may be a storage server (e.g., a database server) separately provided outside the server (100), but is not limited thereto. Instead of separately providing the external server (300) outside the server (100) for storing various information and data, a database may be provided in a separate space within the server (100) to store and manage data in that space. Hereinafter, the hardware configuration of the server (100) will be described with reference to FIG. 2.

[0058] FIG. 2 is a hardware configuration diagram of a server according to another embodiment of the present invention.

[0059] Referring to FIG. 2, a server (100) according to another embodiment of the present invention may include one or more processors (110), a memory (120) for loading a computer program (151) executed by the processor (110), a bus (130), a communication interface (140), and a storage (150) for storing the computer program (151). Here, FIG. 2 illustrates only the components related to the embodiment of the present invention. Therefore, a person skilled in the art to which the present invention pertains will understand that other general-purpose components may be included in addition to the components illustrated in FIG. 2.

[0060] The processor (110) controls the overall operation of each component of the server (100). The processor (110) may be configured to include a CPU (Central Processing Unit), an MPU (Micro Processor Unit), an MCU (Micro Controller Unit), a GPU (Graphic Processing Unit), or any form of processor well known in the art of the present invention.

[0061] Additionally, the processor (110) may perform operations for at least one application or program for executing the method according to embodiments of the present invention, and the server (100) may have one or more processors.

[0062] In various embodiments, the processor (110) may further include RAM (Random Access Memory, not shown) and ROM (Read-Only Memory, not shown) for temporarily and / or permanently storing signals (or data) processed within the processor (110). Additionally, the processor (110) may be implemented in the form of a system-on-chip (SoC) comprising at least one of a graphics processing unit, RAM, and ROM.

[0063] Memory (120) stores various data, instructions and / or information. Memory (120) may load a computer program (151) from storage (150) to execute a method / operation according to various embodiments of the present invention. When the computer program (151) is loaded into memory (120), the processor (110) may perform the method / operation by executing one or more instructions constituting the computer program (151). Memory (120) may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

[0064] The bus (130) provides communication functions between components of the server (100). The bus (130) can be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

[0065] The communication interface (140) supports wired and wireless internet communication of the server (100). Additionally, the communication interface (140) may support various communication methods other than internet communication. To this end, the communication interface (140) may be configured to include a communication module well known in the art of the present invention. In some embodiments, the communication interface (140) may be omitted.

[0066] Storage (150) can store a computer program (151) non-temporarily. When performing a process according to an embodiment disclosed through a server (100), storage (150) can store various information necessary to perform an operation according to the disclosed embodiment.

[0067] The storage (150) may be configured to include non-volatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0068] A computer program (151) may include one or more instructions that cause a processor (110) to perform a method / operation according to various embodiments of the present invention when loaded into memory (120). That is, the processor (110) may perform the method / operation according to various embodiments of the present invention by executing the one or more instructions.

[0069] In one embodiment, the computer program (151) may include one or more instructions to perform the method according to the disclosed embodiment.

[0070] The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. The software module may reside in RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, a hard disk, a removable disk, a CD-ROM, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0071] The components of the present invention may be implemented as a program (or application) and stored on a medium to be executed in combination with a computer, which is hardware. The components of the present invention may be implemented as software programming or software elements, and similarly, embodiments may be implemented in programming or scripting languages ​​such as C, C++, Java, assembler, etc., including various algorithms implemented as combinations of data structures, processes, routines, or other programming configurations. Functional aspects may be implemented as algorithms executed on one or more processors.

[0072] Hereinafter, various embodiments according to the present invention will be described.

[0074] FIG. 3 is a flowchart illustrating the entire process of a method for generating a three-dimensional volume based on physical-data fusion interpolation according to an embodiment of the present invention.

[0075] Referring to FIG. 3, a method for generating a three-dimensional volumetric thermal volume based on physical-data fusion interpolation according to one embodiment of the present invention may be performed by a step of acquiring a two-dimensional temperature image (S301), a step of generating grid-based temperature data (S302), a step of performing virtual camera calibration (S303), a step of raycast-based three-dimensional mapping (S304), a step of physical-data fusion spatial interpolation (S305), and a step of generating a three-dimensional volumetric thermal volume (S306).

[0076] First, in the 2D temperature image acquisition step (S301), a process of acquiring 2D temperature images having different shooting directions and angles of view from a plurality of thermal imaging cameras installed within the facility space may be performed.

[0077] More specifically, the step of acquiring a two-dimensional temperature image (S301) may be performed by including the step of acquiring a first two-dimensional temperature image including a surface heat distribution of the floor surface from a first thermal imaging camera facing a passage area of ​​the facility space, and the step of acquiring a second two-dimensional temperature image including a surface heat distribution of the side portion of the facility from a second thermal imaging camera facing an equipment side area of ​​the facility space.

[0078] The first thermal imaging camera may be installed in the ceiling area of ​​the facility space and positioned to face the floor of the passageway area, for example, arranged at regular intervals above the centerline of the passageway. In this case, the field of view of each first thermal imaging camera may be set to cover the entire width of the passageway, and the gaps in temperature data may be minimized by positioning the cameras so that their shooting areas partially overlap. For example, in the case of a long passageway, multiple cameras may be installed in a linear arrangement at regular intervals, and each camera can generate continuous floor surface temperature images while maintaining the same shooting direction.

[0079] Meanwhile, the second thermal imaging camera may be positioned at an angle toward the side of the rack, and, for example, may be installed in an intersecting direction along each rack row. In this case, the direction of the second thermal imaging camera may be set to face the area where heat exhaust is concentrated between the front and rear of the rack, and the shooting range may be configured to include the section where heat exhausted from a specific rack spreads to an adjacent passage. In particular, multiple second thermal imaging cameras may be positioned to have different angles, and, for example, by installing cameras with different angles of inclination at the same location, temperature information from multiple directions can be obtained for the same equipment.

[0080] In addition, the camera installation location can be set differently depending on the structure of the facility space. For example, in a space with a high ceiling, the installation height of the first thermal imaging camera can be adjusted to maintain shooting resolution, and in a case where the rack height is large, the second thermal imaging camera can be installed in multiple layers to obtain the heat distribution of the upper and lower parts respectively. For example, the upper camera can be installed to face the heat exhaust port at the top of the rack, and the lower camera can be configured to photograph the air intake area at the bottom of the rack.

[0081] Additionally, each thermal imaging camera can internally generate image data containing temperature values ​​at the pixel level, and this data can be managed on a frame-by-frame basis. As a more specific embodiment, a single frame may be structured to include camera identification information, shooting time information, resolution information, and an array of temperature values ​​by pixel coordinate. For example, a camera ID, frame index, shooting time, resolution information, and temperature array data may be stored as a single record in a database, and the temperature array may be stored in the form of a two-dimensional matrix.

[0082] In addition, data acquired from multiple cameras can be synchronized based on the same time. For example, each camera can generate data including a timestamp synchronized with a common time server, and data from the same point in time at regular intervals can be grouped into a single data set. This allows temperature images captured simultaneously by different cameras to be managed as a single group and can be used to construct spatial information from the same point in time in a subsequent step.

[0083] In addition, in some embodiments, noise removal or correction processing may be performed on the output data of the camera. For example, if an abnormally high temperature value is detected at a specific pixel, it can be determined as an outlier and corrected by comparing the difference with surrounding pixels, and data stability can be ensured by applying an average value to values ​​that occur repeatedly at the same location over a certain period.

[0084] In addition, corrections considering the influence of external factors can be performed depending on the camera's installation environment. For example, since values ​​lower or higher than the actual temperature may be measured on metal surfaces with high reflectivity, a correction factor may be applied to the corresponding area, and different correction values ​​may be set for each camera. These correction values ​​can be calculated by comparing with a previously measured reference temperature and can be stored as camera setting information.

[0085] In addition, sensors for recognizing spatial structures may be added alongside thermal imaging cameras. For example, spatial distance information can be collected using LiDAR sensors, in which case depth information aligned with each camera data point can be stored together. This depth information can be configured to correspond to each pixel in the camera image and can be applied as information indicating the location of a specific pixel in the actual space.

[0086] In addition, when using RGB-D sensors, color and depth information can be collected simultaneously and utilized to identify the location or structure of equipment. For example, data can be configured by recognizing equipment with a specific color pattern and linking it to temperature data at that location.

[0088] Next, in the grid-based temperature data generation step (S302), a process may be performed to generate a plurality of grid unit temperature data by dividing the first two-dimensional temperature image and the second two-dimensional temperature image obtained in step S301 according to a predetermined resolution system.

[0089] More specifically, in step S302, a process may be performed to divide the horizontal and vertical directions of the image into multiple cells in order to treat the image frames collected from each thermal imaging camera at the same level of spatial resolution, wherein each cell may be defined as a single grid unit data. For example, if a first two-dimensional temperature image includes the entire floor surface of a passageway, the image may be divided into a grid structure of N rows and M columns, and each grid cell may store a representative temperature value corresponding to the area, grid position information, camera identification information, and shooting time information together. A second two-dimensional temperature image may also be divided in the same manner, and the surface heat distribution of the side of the equipment may be reconstructed into a cell unit temperature array.

[0090] At this time, the grid division resolution may vary depending on the scale of the facility space, the spacing between equipment, the camera installation height, the camera field of view, and the required spatial representation density, and different division densities may be applied to the passageway area and the equipment side area. For example, the floor surface of the passageway may be divided into cells of relatively uniform size to represent the temperature distribution over a wide area, and since the equipment side area may include exhaust vents or localized heat sections, more fine division may be applied to specific height areas. In addition, since the first thermal camera image and the second thermal camera image have different field of view and distance conditions, independent grid division parameters are set for each image, and then a separate alignment structure may be created so that they can correspond to the same spatial unit in a subsequent step.

[0091] Furthermore, during the grid generation process, a process may be performed to convert pixel-unit temperature values ​​into grid-unit representative values; in this process, the average, median, maximum, or a value reflecting specific weights may be selected. For example, if locally high temperatures occur on the side of the equipment, the heat generation characteristics may be diluted if the average value alone is used; therefore, the maximum value or the ratio of pixels exceeding a certain threshold may be stored together. Additionally, to represent the temperature distribution characteristics of a set of pixels contained within a single cell, information such as the minimum value, deviation value, number of high-temperature pixels, and temperature histogram may be additionally generated. For instance, even for cells with the same average temperature, information on variance or the temperature distribution ratio by interval may be recorded to distinguish between cases where the temperature distribution is uniform and cases where it is concentrated in certain areas.

[0092] In addition, temperature data for each grid unit can be organized in the form of a structured data record containing multiple attributes. For example, a single cell may include the cell's row and column indices, image coordinate position, cell center coordinate, representative temperature value, maximum temperature value, minimum temperature value, temperature deviation, camera identifier, frame identifier, and capture time information. Furthermore, additional information such as cell boundary coordinates, cell size information, the number of valid pixels contained in the cell, and the number of outlier pixels may be stored together. This data can be grouped by frame and stored in the form of an array structure in memory or a database table, and each frame may consist of a set of multiple cell records.

[0093] Additionally, grid data acquired from multiple thermal imaging cameras can be sorted according to time information, and frames collected within a certain time interval can be organized into a single synchronized data set. In this case, each cell may include an auxiliary value indicating the difference between the capture times, and a flag may be added to determine whether correction based on the time difference is required. For example, if a group of frames considered to be from the same point in time is formed, a common index may be assigned to that group, and they can be processed as data based on the same time in a subsequent step.

[0094] In addition, mask information can be generated to select valid cells for the outer edges of the image or areas where significant lens distortion occurs, and low-reliability cell data can be distinguished through this mask. For example, if a sudden temperature change occurs at the edge of the image, a separate status value can be recorded for cells identified as abnormal by comparing the temperature difference with adjacent cells. Furthermore, camera-specific correction coefficients can be applied to the cell data to compensate for measurement errors caused by equipment surfaces with high reflectivity.

[0095] In addition, for images including the side of the equipment, since a single cell may simultaneously contain both the equipment surface and the air area, information regarding the proportion of pixels within the cell occupied by the equipment area can be stored together. For example, if a certain percentage or more of the pixels within a cell correspond to the equipment surface, the cell may be classified as an equipment cell; otherwise, it may be classified as an air area cell. Furthermore, in the case of cells located in boundary areas, since they may span multiple pieces of equipment, multiple equipment identifiers may be associated with each cell, and the inclusion ratio for each piece of equipment may be recorded together.

[0096] Furthermore, since multiple pieces of equipment may be included in a single image depending on the arrangement of equipment within the facility space, identification information indicating which equipment the corresponding area belongs to may be additionally recorded in each cell. More specifically, each cell may store equipment ID, equipment type, equipment location coordinates, and equipment height range information together, thereby enabling equipment differentiation at the cell level even within the same image. For example, a specific cell may be identified as a rack unit, while an adjacent cell may be identified as a cooling device.

[0097] Additionally, in an embodiment including multiple sensors, depth information acquired from a LiDAR or RGB-D sensor in addition to a thermal imaging camera may be stored corresponding to each cell, and distance values ​​or spatial location information regarding the cell center position may also be included. In this case, each cell data may be configured as an extended structure that includes not only temperature values ​​but also depth values, sensor type, and sensor accuracy information.

[0098] Additionally, each cell can be configured to contain information on temperature changes over time, and temperature values ​​from multiple points in time can be stored in an array format for a cell at the same location. For example, temperature values ​​acquired from a recent certain number of frames can be stored sequentially, allowing for the identification of temperature rise or fall patterns. Furthermore, if the amount of temperature change exceeds a certain level, an event flag can be set for the corresponding cell.

[0099] In addition, grid data generated by the first thermal imaging camera and grid data generated by the second thermal imaging camera can be generated independently and then combined into a single integrated data set based on the same point in time, and each data set may include information on camera type, installation location, shooting direction, and field of view. Furthermore, if cell data generated by multiple cameras exists for the same space, a mapping table indicating the correspondence between the cells may be generated together.

[0100] In addition, the center position information of each cell can be used as input coordinates for subsequent expansion into a 3D space, and cell center coordinates, cell size, and cell boundary range information can be stored together. This allows the position of each cell to be accurately reflected during ray projection operations performed in subsequent steps.

[0101] In addition, the grid generation order may vary depending on the type of camera and the target being captured. Passage area images may be processed by dividing the entire area evenly, and equipment side images may be subdivided around specific heat-generating sections. Additionally, in certain embodiments, an adaptive segmentation method may be applied in which an area where heat is concentrated is detected first, a high-resolution grid is applied to that area, and a low-resolution grid is applied to the remaining area.

[0103] Next, in the virtual camera calibration execution step (S303), a process can be performed to precisely set the alignment relationship between the 2D image coordinates and 3D spatial coordinates obtained from a plurality of thermal imaging cameras, and thereby correct the virtual camera viewpoint to reflect the shooting conditions of the actual camera.

[0104] More specifically, the step of performing calibration (S303) may be performed by including the step of calculating a correspondence relationship between an actual shooting coordinate system and a virtual camera coordinate system using lens distortion coefficients and installation location information for each of the plurality of thermal imaging cameras, and the step of correcting a virtual camera projection relationship based on the correspondence relationship so that the virtual camera viewpoint reflects the shooting time of each of the plurality of thermal imaging cameras.

[0105] First, lens distortion characteristics can be considered for each thermal imaging camera, and in this process, radial and tangential distortions caused by the optical lens structure can be corrected simultaneously. More specifically, to correct the phenomenon where pixel positions between the center and periphery of the image are non-linearly deformed, camera-specific distortion coefficients can be measured and stored in advance, and these coefficients can be used to recalculate the corrected position for each pixel coordinate within the image.

[0106] For example, if a linear structure appears as a curve in an image, the corresponding pixel coordinates can be restored to coordinates corresponding to the actual linear structure by inversely transforming them using distortion coefficients, and the position of each pixel can be corrected down to the subpixel level. Additionally, pixel-level temperature values ​​can be repositioned in synchronization with coordinate correction so that distortion correction can be performed together with temperature information, and adjacent pixel-based interpolation can be performed for empty areas that may occur during the correction process. Through this processing, coordinate errors can be controlled so that they do not accumulate when the cell center coordinates are expanded into 3D space in a subsequent step.

[0107] In addition, the installation location and orientation information of each thermal imaging camera can be defined within a three-dimensional spatial coordinate system, and this information can be used as a key element in constructing the transformation relationship between the camera coordinate system and the facility spatial coordinate system. More specifically, each camera can be installed in a fixed state on the ceiling or structure of the facility, the position of the camera can be expressed as a three-dimensional coordinate value, and the shooting direction can be defined as rotation angle information.

[0108] For example, the first thermal imaging camera may be positioned vertically so as to face the floor, and the second thermal imaging camera may be positioned tilted at a certain angle so as to face the side of the facility. In this case, the camera's rotation information can be expressed as angle values ​​for multiple axes, through which the direction vector of the camera's view can be calculated. Additionally, when the same facility area is photographed from different angles, the coordinate transformation relationship can be precisely established to point to the same point within space, taking into account the difference in viewpoints of each camera.

[0109] In addition, the correspondence between the actual camera coordinate system and the virtual camera coordinate system can be configured to simultaneously reflect the camera's internal characteristics and external placement information, and in this process, a transformation structure combining the camera's internal and external parameters can be used. More specifically, the internal parameters may include focal length, image center position, and pixel spacing information, while the external parameters may include the camera's 3D position and rotation state. Using these parameters, the cell center position on the 2D image can be converted into camera-based orientation information, and subsequently, that orientation can be converted back into facility space-based coordinates.

[0110] For example, if a cell center point is located at a specific coordinate in the image, that coordinate can be converted into a direction vector based on the camera center, and this direction vector can be realigned to the facility space reference direction by reflecting the camera's rotation information.

[0111] In addition, since data generated from multiple cameras has different coordinate system references, alignment processing can be performed to ensure they are consistently represented on the same spatial coordinate system. More specifically, the cell center coordinates generated by each camera can be expressed in the camera's unique coordinate system, and a coordinate transformation operation can be performed to convert them to a common coordinate system based on the facility's spatial reference. In this process, the position and rotation information of each camera is reflected so that the same spatial location is expressed at the same coordinates. For example, if the same facility is captured by different cameras, the cell center coordinates extracted from each image can be corrected to align to the same 3D location, thereby controlling the process to prevent duplication or errors in subsequent stages.

[0112] In addition, the virtual camera can be configured to reflect the shooting conditions of the actual camera, and in this process, position, direction, and angle of view information can be reflected simultaneously. More specifically, the virtual camera can be positioned to have the same location coordinates as the actual camera, and the shooting direction can also be configured to have the same rotation information. Furthermore, by reflecting the angle of view information, the field of view of the virtual camera can be configured to be identical to that of the actual camera, thereby allowing cell positions in a 2D image to be extended in the same direction in 3D space. For example, a virtual camera corresponding to a camera shooting the floor surface of a passageway can be configured to have a vertical viewpoint facing the floor surface, and a virtual camera corresponding to a camera shooting the side of the facility can be configured to have an inclined viewpoint.

[0113] In addition, multiple virtual cameras can be operated simultaneously within a single visualization environment, and in this process, each virtual camera can be defined independently and then placed on the same spatial coordinate system. More specifically, on the visualization engine, each virtual camera can be placed at the same coordinates as the actual camera's location, and cameras with different viewpoints can be configured to look at the same spatial object. In this case, each virtual camera has an independent projection path, and rays can be extended in different directions within the same space. For example, if a virtual camera looking at a passageway area and a virtual camera looking at the side of an facility exist simultaneously, the system can be configured so that information from different directions regarding the same spatial location can be combined.

[0114] Additionally, the virtual camera projection relationship can be used to generate orientation information starting from the cell center point, and this orientation information can be directly linked to subsequent position search operations in 3D space. More specifically, each cell center point can be defined as the starting point of a ray having a specific direction relative to the camera center, and that direction can be determined by the camera's field of view and rotation information. For example, a cell located in the center of the image has a direction that aligns with the camera's center direction, while an edge cell may have a tilted direction depending on the field of view range. This orientation information can be used in subsequent steps for intersection operations with a specific location in 3D space.

[0115] In addition, when a camera configuration with multiple viewing angles is applied, data with different projection paths for the same space may be generated, so an additional alignment relationship between them may be established. More specifically, when a single spatial area is captured from different cameras at different angles, the direction vectors generated from each cell center point may also appear differently, and these differences can be taken into account to correct them so that they correspond to the same 3D location. For example, in the boundary area where the passageway floor and the side of the equipment meet, the difference in viewpoint between the two cameras may be significant, so the direction information may be fine-tuned to increase the alignment accuracy for that area.

[0116] In addition, to reduce positional errors caused by viewpoint differences between multiple cameras, correction using a common structure may be performed; in this process, a fixed physical structure within the facility space may be used as a reference point. More specifically, elements such as floor tile boundaries, rack corners, and straight boundaries of structures may be set as reference positions, and coordinates may be adjusted so that the corresponding structure appears at the same location in each camera image. For example, if the location of a specific structure is displayed with a slight error in different cameras, the camera position or rotation information may be readjusted based on that location.

[0117] In addition, in a multi-sensor environment, depth information obtained from LiDAR or RGB-D sensors in addition to the thermal imaging camera may be used, and coordinate matching between these sensors may be additionally performed. More specifically, a correspondence relationship between distance values ​​obtained from the depth sensor and image coordinates may be established, thereby connecting the cell center point to its position in actual space. For example, if a depth value corresponding to a specific cell exists, the 3D position of that cell can be calculated more accurately, and this can improve the accuracy of the initial direction setting during ray projection calculations.

[0118] Additionally, virtual camera settings can be managed in the form of a data structure, and setting information for each camera can be stored as a separate record. More specifically, a single camera setting record may include position coordinates, rotation information, field of view, resolution, distortion correction coefficients, and intrinsic parameter information, and this information can be managed in conjunction with each cell data. For example, specific cell data may include identification information indicating which camera generated the cell, through which the camera settings to be applied to the cell can be determined.

[0119] In addition, virtual camera settings can be updated according to environmental changes, and during this process, additional corrections may be performed to maintain consistency with existing data. More specifically, if a camera position is changed or a new camera is added, the entire coordinate transformation relationship may be recalculated and adjusted so that the existing data and the new data are consistently represented on the same coordinate system. For example, if there is a positional difference between an existing camera and a new camera, the relative position between the two cameras may be corrected based on a common structure.

[0120] Furthermore, the field of view of the virtual camera may be limited to account for computational efficiency, and projection operations may be omitted for data outside specific areas. More specifically, cell data located outside the virtual camera's angle of view may be excluded from ray projection, thereby reducing unnecessary computations. Additionally, the processing priority of some cells may be adjusted based on their distance from the camera or the direction of view, allowing for the efficient allocation of computational resources.

[0121] Furthermore, in the visualization engine environment, operations can be controlled so that they are not performed on spatial objects not included in the virtual camera's field of view; in this process, spatial occlusion and distance-based segmentation can be applied. More specifically, objects not included in the camera's field of view can be excluded from the calculation, and the level of detail representation of the objects can be adjusted according to their distance from the camera. For example, high-resolution spatial representation can be applied to nearby areas, while simplified representation can be applied to distant areas.

[0123] Next, in the raycast-based 3D mapping step (S304), a process can be performed to precisely map temperature data corresponding to 2D image coordinates to a 3D cube object in 3D space using grid unit temperature data generated in step S302 and virtual camera coordinate transformation information set in step S303.

[0124] More specifically, in step S304, a ray is generated from the center point of each grid cell toward three-dimensional space by reflecting the position and field of view information of the virtual camera, and by searching for the point where the ray intersects with a plurality of three-dimensional cube objects arranged in three-dimensional space, the temperature value corresponding to each cell can be made to correspond to an accurate spatial location.

[0125] First, multiple three-dimensional cube objects configured in the same proportion as the structure of the facility space may be arranged in a three-dimensional space, and these three-dimensional cube objects may be defined as unit spatial elements having a fixed size. More specifically, each three-dimensional cube object may be composed of a spatial block having a fixed range in a three-dimensional coordinate system, and each cube may be defined as a structure including center coordinates, boundary ranges, and spatial size information. For example, a single cube may be stored in a form including lengths in the horizontal, vertical, and height directions along with (x, y, z) center coordinates, and such a cube may be configured as a structure that divides the entire space into a grid. Additionally, in certain embodiments, the cube size may be set differently for each area to reflect the difference in spatial representation between the facility-dense area and the non-facility area.

[0126] Additionally, each three-dimensional cube object may include a structure for data storage; for example, a cube ID, coordinate information, a list of mapped temperature values, a set of input camera identifiers, data update time information, and a measurement status value may be configured as a single record. More specifically, a single cube may have a list structure capable of accommodating multiple temperature values, and each temperature value may be associated with the camera ID that generated the data and time information. Furthermore, to represent the distribution characteristics among temperature values ​​within the cube, maximum, minimum, average, and deviation values ​​may be calculated and stored separately.

[0127] Additionally, rays generated from the center point of each grid cell can be set to a direction toward the cell center position, using the virtual camera's location as the starting point. More specifically, the coordinates of each cell center can be converted into a direction vector based on the virtual camera coordinate system, and this direction vector can be realigned based on the facility space coordinate system to set the direction of the ray's progression. For example, a ray can be generated toward the camera center from a cell located in the center of the image, while a ray can be generated toward an inclined direction from cells on the outer edges of the image based on the field of view. Furthermore, each ray can be managed by a data structure containing information such as the starting point, direction vector, maximum search distance, and current travel distance.

[0128] Additionally, the generated ray can be processed by checking for intersections with three-dimensional cube objects while moving at regular intervals within three-dimensional space. More specifically, the ray's path can be composed of sample points divided at regular intervals, and operations to determine which cube each sample point belongs to can be performed repeatedly. For example, the cube index can be calculated for each position as the ray moves in units of a certain distance, and if the corresponding cube exists, it can be determined as an intersection. Furthermore, the total amount of search computation can be reduced by limiting the cubes to be searched using a spatial partitioning structure.

[0129] Additionally, a collider structure for collision detection can be configured for each three-dimensional cube object, thereby enabling rapid determination of whether a ray intersects with a cube. More specifically, each cube has a spatial region representing its boundary range, and an operation can be performed to calculate whether a ray passes through that region. For example, a method to determine whether the ray's path intersects with the cube's boundary face can be applied, thereby determining the exact intersection point. Furthermore, the collision detection operation can be performed using a parallel processing structure, allowing inspection operations for multiple cubes to be performed simultaneously.

[0130] In addition, when a single ray intersects multiple cubes, conditions for selecting the final mapping target cube may be set. More specifically, a method of selecting the cube closest to the camera may be applied, or a method of removing noise by selecting a cube at a certain distance may be applied. For example, if an equipment surface and an air layer exist continuously, the cube corresponding to the equipment surface may be set to be selected preferentially. Additionally, when temperature values ​​for multiple cubes are recorded along the ray path, the cube data may be configured to be stored while maintaining the multiple values.

[0131] Furthermore, if rays generated from multiple cameras intersect the same cube, multiple temperature data can be accumulated in that cube. More specifically, each cube has a structure capable of accommodating multiple input values, and these values ​​can be combined according to certain rules. For example, an average value or a weighted average value can be calculated, and weights can be set differently depending on the position, distance, and field of view information of a specific camera. Additionally, each temperature value stored in the cube is maintained as an individual record and can be selectively used in a subsequent step.

[0132] Additionally, areas not directly observed by the camera during the mapping process may remain empty, and these cubes can be identified separately. More specifically, cubes where light rays do not reach may be marked as having no temperature value, and a flag indicating an unmeasured state may be stored in such cubes. Furthermore, these cubes may contain spatial connection information with surrounding cubes and can be utilized as interpolation target areas in a subsequent step.

[0133] In addition, mapping result data can be stored in a structured form, and data corresponding to each three-dimensional cube object can be managed in record units having a specific format. More specifically, each cube data can be composed of a structure that includes spatial information representing the cube's center coordinates and boundary range, a list structure for storing mapped temperature values, a camera identifier where the temperature value was generated, time information where the temperature value was input, and a status flag indicating whether the cube is in the directly measured area. For example, a single cube can be defined in a form that includes a cube ID, center coordinates, boundary size, a list of temperature values, a list of camera IDs, a timestamp list, and a status flag. Furthermore, values ​​from multiple time points can be accumulated and stored in the list of temperature values, and each value can be managed as time-series data along with the input time.

[0134] In addition, such cube data may be stored in a form where a 3D array structure and a coordinate-based hash structure are applied in parallel, considering spatial search and update efficiency. More specifically, the entire space may be configured to be composed of a 3D index-based array to enable basic location access, and at the same time, a fast lookup of a specific cube may be performed using a hash structure in which coordinate values ​​are converted into keys. Furthermore, in certain embodiments, a structure that hierarchically divides the space may be applied to improve the access speed to a specific area.

[0135] In addition, statistical characteristics can be stored in the cube data along with temperature values; for example, the maximum, minimum, average, and deviation values ​​of the temperature values ​​entered for the same cube can be updated. More specifically, the system can be configured so that existing statistical values ​​are updated whenever a new temperature value is entered, thereby allowing for the immediate identification of temperature distribution characteristics at the cube level. Furthermore, a structure can be added to mark a corresponding cube in a separate state if the deviation between temperature values ​​exceeds a certain level.

[0136] Furthermore, mapping operations can be performed using a parallel processing structure, allowing a large volume of ray operations to be executed simultaneously. More specifically, sets of rays generated from each camera can be divided into independent processing units, and each processing unit can be executed concurrently on different computation threads or GPU computation blocks. For example, thousands or more rays generated from a single frame can be configured to be processed in parallel, and the results of the intersection operations for each ray can be stored in a separate buffer and then reflected in the cube data. Additionally, a cube-level access control structure can be applied to account for cases where simultaneous access to the same cube occurs.

[0137] Furthermore, in a multi-sensor environment, information acquired from depth sensors or distance measuring devices can be reflected in the ray generation and intersection determination processes. More specifically, if a distance value corresponding to the center position of each grid cell exists, the maximum search distance of the ray can be limited to that value, thereby reducing unnecessary spatial exploration. Additionally, in areas where depth information exists, the initial intersection position of the ray can be set more accurately, and consequently, the accuracy of cube mapping can be improved.

[0138] In addition, depth information and temperature data can be stored in a cube in a combined form, and each data point can be managed separately according to the sensor type and measurement conditions. More specifically, the cube data may include the sensor type that generated the value and distance information along with the temperature value, and a structure for distinguishing data reliability can be applied based on this.

[0139] Furthermore, mapping results can be continuously updated over time, and new temperature data can be reflected in the cube for each frame. More specifically, temperature values ​​can be accumulated and stored in chronological order within the same cube, and each value can be maintained as time-series data along with the input time. For example, a certain number of recent data can be maintained in a cyclical structure, while older data can be automatically removed. Additionally, the amount of temperature change over time can be calculated and stored separately, allowing for the analysis of temperature change patterns within the space.

[0141] Next, in the physical-data fusion spatial interpolation step (S305), for areas among the three-dimensional cube objects generated in step S304 for which direct measurement values ​​do not exist, a process of estimating the temperature by reflecting the temperature data of surrounding cubes, spatial characteristics, and temporal changes together can be performed.

[0142] More specifically, in step S305, measured cubes and unmeasured cubes are distinguished, and interpolated input data can be constructed using temperature data collected from adjacent cubes and spatial positional relationships for the unmeasured cubes, thereby enabling the formation of a continuous temperature distribution throughout the space.

[0143] First, three-dimensional cube objects can be classified into measured cubes and unmeasured cubes based on their measurement status, and unmeasured cubes can be classified into a separate set of interpolation targets. More specifically, a cube with a temperature value mapped by one or more rays can be defined as a measured cube, while a cube with no temperature value because no rays reach it can be defined as an unmeasured cube. For example, cubes located in the air gap area between racks or in the space behind the equipment can be classified as unmeasured cubes, and these cubes can be set as input targets for subsequent interpolation operations.

[0144] Additionally, interpolated input data can be configured for each unmeasured cube, and this may consist of temperature data extracted from multiple adjacent cubes and spatial location information. More specifically, adjacent cubes located in the up, down, left, right, and front, back directions for a single unmeasured cube may be searched, and temperature data may be collected from cubes among these that have temperature values. For example, a set of adjacent cubes in 6 or 26 surrounding directions of a specific cube may be configured, and among these, cubes having valid temperature values ​​may be selected and used as input data. Furthermore, distance information and relative location information with adjacent cubes may be included to reflect spatial relationships.

[0145] Furthermore, the characteristics of heat distribution within the space can be reflected by utilizing the temperature difference between adjacent cubes. More specifically, the direction and magnitude of temperature change can be extracted using the difference in temperature values ​​between adjacent cubes, and this information can be used to estimate the temperature of an unmeasured cube. For example, if the temperature continuously increases in a specific direction, it can be determined that heat transfer is occurring in that direction, and the temperature value of the unmeasured cube can be set accordingly. Additionally, temperature differences from multiple directions can be considered simultaneously to reflect the temperature change pattern within the space.

[0146] In addition, information on temperature changes over time can be included in the interpolation input data. More specifically, time-series temperature data stored in each cube can be utilized to extract trends in temperature changes over time, and this change information can be reflected in the temperature estimation of unmeasured cubes. For example, if the temperature of a specific cube continuously rises over a certain period, a similar upward trend may be reflected in adjacent unmeasured cubes. Furthermore, in the event of a sudden temperature change, the spread of such change to surrounding cubes can be considered.

[0147] Furthermore, parameters reflecting changes in air density and natural convection characteristics within the space can be generated. More specifically, changes in air density due to temperature differences can be estimated, and consequently, rising or descending airflow can be considered. For example, a tendency for air to rise in high-temperature regions can be reflected, while a tendency for air to descend in low-temperature regions can be reflected. Additionally, this direction of air movement can be used as a factor in determining the temperature transfer path between adjacent cubes.

[0148] In addition, interpolation operations can be performed using an artificial intelligence model, and the model can be structured to process spatial and temporal data simultaneously. More specifically, the input data can be structured in the form of multidimensional data including the temperature values ​​of adjacent cubes, location information, time information, and air movement characteristic information, and the temperature value of an unmeasured cube can be estimated using this. For example, a neural network structure capable of reflecting a three-dimensional spatial structure can be applied, and data from surrounding cubes can be processed simultaneously around each cube.

[0149] Furthermore, the input data for the artificial intelligence model can be normalized to have a specific structure. More specifically, it can be configured so that an equal number of adjacent cube data points are input for each cube, and if insufficient, they can be supplemented with default values ​​or data from a previous time point. For example, if some surrounding cubes are unmeasured, the temperature value or average value from a previous time point can be input for those locations. Additionally, the input data is normalized to a specific range to maintain the computational stability of the model.

[0150] Additionally, the interpolation result can be stored as a new temperature value for each unmeasured cube, and this value can be managed by adding it to the existing cube data structure. More specifically, the interpolated temperature value can be stored not as a single value, but as a record that includes the generation time, characteristics of the input data used, and the operation execution status; this allows direct measurement values ​​and interpolated values ​​to be clearly distinguished within the same cube.

[0151] For example, a single cube may store a list of measured values ​​and a list of interpolated values ​​separately, and each list may be structured to include information such as temperature value, time of creation, data type (measurement or interpolation), and the number of input data. Additionally, in certain embodiments, a value indicating the confidence level for the interpolated values ​​may be additionally stored, which may be set based on the distribution state of adjacent cube data or the stability of time change.

[0152] In addition, during the interpolation operation, the cube data structure can be extended to include configuration information of the input data. More specifically, each cube may store a list of identifiers of adjacent cubes used to generate the corresponding interpolation value, the temperature values ​​of each adjacent cube, relative position information, and time information at the time of operation execution. For example, if the interpolation value of a specific cube is calculated using the temperature values ​​of the upper, lower, left, and right cubes, the IDs and temperature values ​​of those four cubes may be included in the interpolation record. Additionally, a timestamp at the time of interpolation execution may be stored together, so that the data state at a specific point in time can be reproduced or analyzed thereafter.

[0153] Furthermore, the interpolation operation can be configured as a parallel processing structure to be performed simultaneously on multiple unmeasured cubes. More specifically, each unmeasured cube can be defined as an independent operation unit, and the entire set of cubes can be divided into work units of a certain size and distributed to multiple operation threads or GPU operation blocks. For example, it can be configured so that thousands or more unmeasured cubes perform interpolation operations simultaneously in a single operation cycle, and each cube can independently calculate temperature values ​​by referencing only the data of cubes adjacent to it. Additionally, the operation results are written to memory areas corresponding to each cube, and a cube-level access control structure can be applied to prevent concurrent access conflicts.

[0154] Additionally, the interpolation operation can be configured to be performed repeatedly over time, and the existing interpolation result can be updated whenever new measurement data is input. More specifically, each cube can be set to have an update cycle at regular time intervals, and the interpolation operation can be re-executed by reflecting the latest temperature value of an adjacent cube according to that cycle. For example, in a real-time processing environment, the interpolation operation can be performed at the same interval as the camera frame update cycle, and the previous interpolation value can be replaced with a new result or stored as separate historical data. Furthermore, in certain embodiments, a method of omitting the interpolation operation for cubes that have not changed for a certain period of time may be applied.

[0155] Furthermore, the interpolation results can be accumulated and stored as time-series data, and each cube can be maintained in a structure containing changes in temperature values ​​over time. More specifically, each cube can store a certain number of recent interpolated values ​​in a cyclic structure and be managed in such a way that the oldest value is removed when a new value is input. For example, each cube can be configured to retain the most recent N interpolated values, thereby allowing the pattern of temperature change over time to be continuously recorded. Additionally, this time-series data can be used in subsequent stages to analyze temperature change characteristics or determine abnormal conditions.

[0156] Additionally, the interpolation result data can be arranged into a consistent structure so that it can be used in the step of generating a three-dimensional thermal volume. More specifically, all cubes can be maintained in a state containing at least one temperature value, either a measured value or an interpolated value, and the temperature value of each cube can be normalized to the same data format. For example, the temperature value per cube can be scaled to a certain range or converted into the same unit and stored, thereby allowing the temperature distribution over the entire space to be configured in a uniform form. Furthermore, in certain embodiments, additional data indicating the difference between the interpolated value and the measured value may be generated and included in the cube data.

[0157] In addition, the interpolated result data can be configured to form a continuous set of temperature data for the entire space. More specifically, the interpolation operation can be performed such that every cube contains a temperature value, thereby generating a complete three-dimensional temperature distribution that includes even unmeasured air gap areas. For example, temperature values ​​can be assigned to areas such as air gaps between racks, the space between the ceiling and the floor, and the space behind the equipment, allowing the entire space to be represented as a single continuous thermal volume.

[0159] Finally, in the step of generating a three-dimensional thermal volume (S306), a process may be performed to generate a three-dimensional thermal volume having a continuous three-dimensional temperature distribution over the entire facility space by integrating the temperature data mapped in step S304 and the temperature data interpolated in step S305. More specifically, in this step, a data set constituting the entire space may be completed using the temperature values ​​stored in each three-dimensional cube object, and based on this, a continuous heat distribution may be expressed in the three-dimensional space.

[0160] First, three-dimensional cube objects can be arranged to include measured values ​​or interpolated values, and configured so that valid temperature values ​​exist for all cubes. More specifically, each cube has a data structure containing a temperature value field; if a measured value exists, that value is set preferentially, and if no measured value exists, an interpolated value may be assigned. For example, cube data may be structured in a form such as [Cube ID, coordinate information, final temperature value, data type, time series data], and the final temperature value may be selected and stored as a single value. Additionally, in certain embodiments, the measured value and the interpolated value may be maintained simultaneously and used selectively depending on the display purpose.

[0161] Furthermore, the temperature values ​​of each cube can be rearranged to ensure spatial continuity. More specifically, post-processing operations can be performed to mitigate abrupt temperature differences between adjacent cubes, thereby ensuring a natural connection in the temperature distribution within the space. For example, if the temperature difference between a specific cube and an adjacent cube exceeds a certain level, the temperature value can be adjusted by reflecting the values ​​of surrounding cubes. Additionally, this correction operation can be performed repeatedly over the entire space to form a continuous thermal volume.

[0162] In addition, three-dimensional volumetric thermal data may be composed of a data set having a specific structure. More specifically, the entire space may be configured in the form of a three-dimensional array, where each array element represents a cube and may include a temperature value at that location. For example, an array element with an index (i, j, k) may represent the temperature value of a cube corresponding to a spatial coordinate. Additionally, in certain embodiments, the cube data may be stored in a separate list structure and configured to be accessible using a coordinate index.

[0163] In addition, the generated three-dimensional thermal volume data can be managed in a time-series format. More specifically, three-dimensional temperature distribution data generated at each point in time can be stored as a separate data set, and changes over time can be accumulated and recorded. For example, thermal volume data generated at specific time intervals can be stored sequentially, thereby allowing temperature changes over time to be tracked in three-dimensional space. Additionally, in certain embodiments, a cyclic structure that maintains only a certain number of recent thermal volume data can be applied.

[0164] In addition, three-dimensional thermal volume data can be converted into color information for visualization processing. More specifically, the temperature value of each cube can be mapped to a color value according to a preset temperature range, thereby allowing the temperature distribution in three-dimensional space to be intuitively represented. For example, low-temperature regions can be displayed in shades of blue, high-temperature regions in shades of red, and intermediate temperature regions in between. Additionally, in certain embodiments, the color can be set to change according to the rate of temperature change.

[0165] In addition, during the visualization process, various optimization techniques may be applied to reduce the amount of computation, taking into account the burden of rendering operations for large-scale 3D cube objects. More specifically, cubes that are not included in the user's viewpoint may be set to be excluded from the rendering target, and to this end, an operation to determine whether a cube is included based on the field of view of a virtual camera may be performed.

[0166] For example, cubes not included within the virtual camera's field of view may be removed from the list of rendering targets, and cubes located at the boundary of the field of view may be configured to be displayed only partially. Additionally, a method of gradually adjusting the rendering resolution based on the distance value between the camera and the cube may be applied, whereby cubes in the near area are represented in finer units, and cubes in the far area are represented in merged units. Furthermore, in certain embodiments, a method of omitting rendering update operations for cubes that have not recently changed may be applied using a cube update status flag.

[0167] In addition, three-dimensional thermal volume data can be converted into a data structure that facilitates transmission and storage, taking into account integration with external systems. More specifically, each cube data can be serialized into a structure including coordinate information, temperature values, data types, and time-series information, and this serialized data can be configured in the form of a message capable of network transmission. For example, each cube consists of unit data including a coordinate index and a temperature value, and multiple cube data can be bundled into a single packet for transmission. Furthermore, the transmitted data includes time information, allowing the thermal volume state at a specific point in time to be reconstructed at the receiving end, and the receiving end can restore the data into a three-dimensional array structure to perform visualization or analysis processing.

[0168] Furthermore, the generated three-dimensional thermal volume data can be configured to enable selective analysis of specific areas. More specifically, the user terminal may provide input for designating a specific coordinate range or facility area, thereby allowing cube data included in that area to be filtered and extracted. For example, if a coordinate range corresponding to the area surrounding a specific rack is entered, temperature values ​​and time-series data of the cubes within that range can be collected separately. Additionally, partial cross-sectional analysis can be performed by selecting only specific height sections or specific directional areas, thereby enabling detailed verification of the temperature distribution characteristics of specific areas within the space.

[0169] In addition, three-dimensional thermal volume data can be used as analysis data for determining abnormal conditions. More specifically, the temperature value of each cube can be compared with adjacent cubes to calculate the relative temperature difference, and if it exceeds a certain range, the corresponding cube can be identified as a candidate for an abnormal condition. For example, a cube with a temperature higher than a certain level compared to the average value of surrounding cubes can be marked as an abnormal area, and such a cube can be configured to have a separate status flag. Furthermore, a cube can be classified as an abnormal condition if the rate of temperature change over time increases beyond a certain level, and this information can be provided to the user terminal in the form of a warning.

[0170] Furthermore, three-dimensional thermal volume data can be combined with various sensor data to expand into multidimensional environmental information. More specifically, each cube can be expanded into a structure that includes various environmental data, such as airflow velocity, humidity, and pressure, in addition to temperature values, and each data set can be managed integrally within the same coordinate system. For example, a specific cube may include the airflow velocity and humidity values ​​at a given location along with the temperature value, and this composite data can be used to represent the environmental conditions within the space from multiple perspectives. Additionally, data by sensor type can be stored in separate fields, allowing for the selective analysis of only specific sensor data.

[0171] In addition, three-dimensional thermal volume data can be structured so that it can be used as input data for artificial intelligence models. More specifically, an input structure combining temperature data in the form of a three-dimensional array and time-series data can be constructed, and this can be used to predict temperature changes or analyze abnormal conditions. For example, a method can be applied to predict future temperature distributions by inputting the three-dimensional temperature distribution at a specific point in time along with data from previous points in time, or a method can be applied to detect similar patterns in advance by learning temperature change patterns occurring in a specific area. Furthermore, spatial and temporal characteristics can be reflected simultaneously by using the positional information and time information of each cube together as input.

[0172] In addition, three-dimensional thermal volume data can be managed in various ways considering storage and management efficiency. More specifically, three-dimensional array data at each point in time can be stored as individual data sets and indexed according to time information to be retrieved quickly. For example, a time-based index can be set to selectively retrieve only data corresponding to a specific time range, and each data set can be stored in a compressed form to reduce storage space. Furthermore, in certain embodiments, cube-unit data can be stored as individual records to enable coordinate-based retrieval, thereby improving the speed of data access to specific locations.

[0174] Below, we will examine in more detail the apparatus for generating a three-dimensional thermal volume based on such physical-data fusion interpolation.

[0175] FIG. 4 is a block diagram illustrating the overall configuration of a device for performing a three-dimensional thermal volume generation method based on physical-data fusion interpolation according to an embodiment of the present invention.

[0176] Referring to FIG. 4, an apparatus (500) for performing a three-dimensional thermal volume generation method based on physical-data fusion interpolation according to one embodiment of the present invention may include first and second thermal imaging cameras (510, 520), a screen splitting unit (530), a three-dimensional space mapping unit (540), a space interpolation module (550), and a visualization unit (560).

[0177] The first and second thermal imaging cameras (510, 520) can perform the role of generating three-dimensional temperature data by acquiring the temperature distribution within the facility space from different viewing directions.

[0178] The first thermal imaging camera (510) may be positioned vertically on the ceiling so as to face the floor surface of the passageway area, and the second thermal imaging camera (520) may be positioned at an angle so as to face the side of the equipment or the rack structure. With this arrangement, the temperature distribution of the central area of ​​the passageway and the temperature distribution of the side of the equipment can be acquired simultaneously at different points in time, and the occluded area that may occur when shooting in a single direction can be reduced. For example, when the first thermal imaging camera (510) shooting towards the floor surface from the top of the passageway and the second thermal imaging camera (520) shooting obliquely toward the side of the rack are combined, the cold air distribution on the floor surface and the heat emission distribution on the side of the rack can be collected simultaneously.

[0179] More specifically, the first thermal imaging camera (510) may be installed in the center of the ceiling of the facility space or above the passageway, and the shooting direction may be set to a vertical direction toward the floor surface. For example, in a facility space where the ceiling height is in the range of about 2.8 m to 4.5 m, the first thermal imaging camera (510) may be installed at a height of about 2.5 m to 4.0 m from the floor surface, and the shooting angle may be set to a range of about 70 degrees to 110 degrees so that the entire width of the passageway is included in a single image. In addition, if the width of the passageway is about 1.0 m to 1.5 m and the length of the passageway is about 5 m to 20 m, 2 to 6 first thermal imaging cameras (510) may be distributed along the length of the passageway at regular intervals, for example, at intervals of about 2.0 m to 4.0 m. In this case, the field of view overlap between adjacent cameras can be set to be approximately 10% to 25% of the length of the passageway, and data gaps can be prevented at the boundary. For example, in a passageway where cold air supply floor tiles are laid, a low-temperature area formed directly above the floor tiles may appear in the image as a blue or cyan distribution, and a relatively high-temperature area along the center of the rack may appear as a yellow or red distribution.

[0180] Additionally, a correction value for correcting the difference in emissivity according to the floor surface material may be included in the setting value of the first thermal imaging camera (510). For example, an epoxy floor may be set to an emissivity of about 0.90 to 0.95, vinyl tiles to about 0.92 to 0.96, rubber mats to about 0.93 to 0.97, and metal floor panels or perforated plates to about 0.55 to 0.80. At this time, the first thermal imaging camera (510) may store a material map for each floor section within the facility space and apply different emissivity correction coefficients depending on the shooting location.

[0181] For example, even if the actual temperature is approximately 24°C, the metal panel section may be measured as low as approximately 21°C to 22°C without correction, so a temperature correction value of approximately +2°C to +3°C may be applied to that section. In addition, a zone-specific correction table may be formed based on the difference in contamination or reflection state between floor tiles, and differentially applied, for example, a correction factor of 1.00 for the first zone, 1.03 for the second zone, and 0.98 for the third zone.

[0182] Additionally, the second thermal imaging camera (520) may be positioned with an angle of inclination toward the side of the rack or equipment, and the angle may be set in the range of 30 to 60 degrees. For example, if the heat exhaust is concentrated at the top or rear upper part of the equipment, the angle may be set to about 35 to 45 degrees, and if the goal is to widely capture the temperature distribution across the entire side, an angle in the range of about 45 to 60 degrees may be applied.

[0183] Additionally, the second thermal imaging camera (520) may be installed at a position approximately 1.5 m to 3.0 m away from the front of the rack or at a position approximately 0.5 m to 1.5 m away from the upper corner of the rack, and the shooting height may be set to approximately 2.0 m to 3.5 m from the floor surface. For example, if the rack height is approximately 2.0 m, the second thermal imaging camera (520) may be installed at a height of approximately 2.4 m to include the exhaust port at the top of the rack, and the local high temperature distribution formed in an area of ​​approximately 0.2 m to 0.5 m around the exhaust port may be recorded as an image. Additionally, in certain embodiments, the second thermal imaging camera (520) for the top and the second thermal imaging camera (520) for the middle may be placed simultaneously so that the upper exhaust heat and the lateral heat diffusion at the middle height can be observed together.

[0184] Additionally, the first and second thermal imaging cameras (510, 520) may be configured to have different viewing angles and resolutions. For example, the first thermal imaging camera (510) may have a resolution of 320×240, 640×480, or 1024×768, and a wide viewing angle of approximately 90 degrees may be applied to capture a wide area. On the other hand, the second thermal imaging camera (520) may have a resolution of 640×480 or 1280×720 to finely identify localized heat sections on the side of the rack, and the viewing angle may be set relatively narrow to approximately 35 to 60 degrees.

[0185] For example, when the second thermal imaging camera (520) captures at a resolution of 640×480, the rack height section of approximately 2.0 m is positioned to correspond to the entire image height, thereby ensuring a spatial resolution of approximately 3 mm to 8 mm per pixel. Additionally, in certain embodiments, a digital magnification function may be applied so that after capturing the entire side of the rack, only a specific server slot area is magnified to approximately 2 to 4 times to re-capture the heat distribution.

[0186] Additionally, each thermal imaging camera (510, 520) may operate with consideration for synchronization of the shooting time. For example, each camera may acquire images at a level of 5 frames, 10 frames, or 30 frames per second and may be connected to a common time server to receive timestamps in milliseconds. At this time, synchronization may be maintained so that the frame acquisition time difference between the first thermal imaging camera (510) and the second thermal imaging camera (520) is within approximately 10 ms to 100 ms, and in certain embodiments, if the allowable time difference exceeds approximately 50 ms, the corresponding frame set may be excluded or reordered. For example, in situations where the temperature distribution changes within approximately 1 second due to rapid server load fluctuations, the spatial temperature state at the same time can be configured more accurately by reducing the synchronization error.

[0187] Additionally, each thermal imaging camera (510, 520) can perform periodic correction processing to maintain temperature measurement accuracy. For example, the response characteristics of the camera sensor can be corrected by photographing reference panels at a certain temperature level of 25 ℃, 35 ℃, and 45 ℃, and if the measured value differs from the actual value by a range of about ±0.5 ℃ to ±2.0 ℃, a corresponding correction coefficient can be calculated.

[0188] In addition, if sensor sensitivity deteriorates due to prolonged operation, automatic recalibration may be performed at intervals of approximately 12, 24, or 72 hours. For example, if a 35°C reference panel is measured at 33.8°C in a specific camera, a correction value of approximately +1.2°C may be stored and applied to subsequent shooting data. Additionally, if the ambient temperature changes rapidly, a separate temperature drift correction table may be applied using the value from the camera's internal temperature sensor.

[0189] Additionally, the first and second thermal imaging cameras (510, 520) can be expanded in combination with various sensors. For example, each camera can be equipped with an RGB camera, a depth sensor, and a LiDAR sensor, and shape information or distance information can be acquired simultaneously with temperature data. In this case, a depth value corresponding to a specific pixel can be provided in the range of approximately 0.5 m to 5.0 m, thereby allowing for more accurate identification of which location in the actual space corresponds to the same temperature data. Additionally, when RGB images are collected together, visual elements such as rack numbers, equipment labels, and structural boundaries are recorded together, which can increase the accuracy of correspondence between temperature data and location information.

[0190] Additionally, each thermal imaging camera (510, 520) may apply a region-of-interest-based shooting method depending on the shooting environment and the purpose of analysis. More specifically, when a specific rack row or a specific exhaust section is designated as a region of interest, the frame rate, shooting resolution, and digital magnification rate may be selectively adjusted for that region. For example, a second thermal imaging camera (520), which normally shoots the entire area at 5 frames per second with a resolution of 640×480, can increase to 15 to 30 frames per second when the upper exhaust section of a specific rack is set as a region of interest, and can generate high-resolution subframes magnified 2 to 4 times only for the region of interest.

[0191] Additionally, the computational burden can be reduced by lowering the resolution or the number of frames for sections outside the area of ​​interest. For example, the entire frame can be maintained at 640×480, while the area of ​​interest can be stored as a separate 320×320 or 512×512 detailed frame. Additionally, the area of ​​interest may be specified by the user, or it may be automatically specified if a section is detected where the temperature change amount in the previous frame has increased by a certain level, for example, about 2°C or more, or is about 5°C or more higher than the surrounding average. At this time, the result of specifying the area of ​​interest can be stored as a setting record in the form of [area ID, start coordinate, end coordinate, magnification ratio, frame rate] and can be transmitted together to the screen splitting unit (530) or the 3D space mapping unit (540).

[0192] Additionally, the first and second thermal imaging cameras (510, 520) may include various interfaces for data transmission and storage. For example, each camera may transmit image data to the screen splitting unit (530) via wired Ethernet, an optical communication module, or a wireless communication module, and a compressed format or a streaming format may be applied for real-time processing. Additionally, if an internal buffer memory is included, frames from the last few seconds to tens of seconds may be stored even if a temporary communication delay occurs. For example, the buffer may be set to temporarily store data at 10 frames per second with a resolution of 640×480 per frame for about 5 seconds, and missing frames may be transmitted sequentially upon communication recovery.

[0193] Additionally, the first and second thermal imaging cameras (510, 520) can be configured in various arrangements depending on the facility structure and requirements. For example, in addition to installing a single first thermal imaging camera (510) in the center of the passageway, multiple first thermal imaging cameras (510) can be placed at the top of both sides of the passageway to capture images toward the floor from both sides. Furthermore, regarding the side of the equipment, multiple second thermal imaging cameras (520) can be placed at upper, middle, and lower positions to simultaneously collect temperature distributions by height. For example, in a high-density rack environment, two or more second thermal imaging cameras (520) can be installed at different heights for each rack row, so that the heat concentration section of the upper exhaust port and the heat diffusion section of the middle equipment surface can be recorded separately.

[0194] In addition, in certain embodiments, a mobile thermal imaging camera may be additionally applied. For example, a thermal imaging camera moving along a ceiling rail or a thermal imaging camera mounted on an autonomous vehicle may move through the facility space and collect temperature data in areas not covered by a fixed camera. In this case, the mobile camera may approach the location where an abnormal temperature pattern is detected in the image collected by the fixed camera to additionally acquire detailed images. For example, if a localized high temperature area of ​​about 40°C or higher is detected on the rear of a specific rack, the mobile camera may move to that location and perform magnified shooting from a distance of about 0.5 m to 1.0 m.

[0195] Additionally, the first and second thermal imaging cameras (510, 520) may be configured to provide complementary field of view information. More specifically, the first thermal imaging camera (510) provides a temperature distribution of the center of the passageway and floor, and the second thermal imaging camera (520) provides a temperature distribution of the side of the equipment and the heat exhaust area, so that the data from the two cameras can be combined to obtain temperature data for the entire space. For example, if a low-temperature distribution in the center of the passageway is confirmed by the first thermal imaging camera (510) and a high-temperature distribution on the upper side of a specific rack is confirmed by the second thermal imaging camera (520), the heat distribution that may be formed in the air layer of that area can be estimated more precisely. Furthermore, by supplementing areas not captured by one camera with another camera, the density of temperature data for the entire space can be increased.

[0197] The screen splitting unit (530) can perform the role of dividing a two-dimensional temperature image obtained from the first and second thermal imaging cameras (510, 520) into a grid structure of a certain resolution and converting it into unit temperature data that can be used in the three-dimensional space mapping unit (540). More specifically, the screen splitting unit (530) can apply a variable grid splitting structure according to the resolution of the input image, the characteristics of the target area, and the purpose of analysis, and generate representative temperature values ​​and location information corresponding to each grid unit.

[0198] More specifically, the screen splitting unit (530) can divide the input 2D temperature image into multiple cells in the horizontal and vertical directions, and each cell can be defined as a single grid unit data. For example, when an image with a resolution of 640×480 is input, it can be divided into a grid structure such as 32×24, 64×48, or 80×60, and each cell can include an area of ​​approximately 10×10 pixels to 20×20 pixels in size. In addition, in the case of an image with a resolution of 1024×768, it can be divided to a level of 64×48 or 128×96 to ensure finer spatial resolution. At this time, the average temperature value, median value, or specific weighted average value of the pixels within the corresponding area can be assigned to each cell as a representative temperature value.

[0199] In addition, each grid cell can be created with a structure that includes not only simple temperature values ​​but also various additional information. More specifically, each cell data may include cell center coordinates, cell size, cell boundary information, input camera identifier, time of capture, temperature distribution statistics, etc. For example, a single cell may be stored in a structure such as [Cell ID, center coordinates (u, v), average temperature value, maximum temperature value, minimum temperature value, standard deviation, camera ID, timestamp]. Furthermore, in certain embodiments, an indicator representing the uniformity of the temperature distribution within the cell may be additionally included, and if the temperature deviation exceeds a certain level, the cell may be marked as a non-uniform area.

[0200] In addition, the grid division resolution can be variably set according to the characteristics of the target area. For example, a relatively coarse grid structure of 32×24 or 48×36 may be applied to areas with a relatively uniform temperature distribution, such as a passageway floor, while a fine grid structure of 64×48 or 128×96 may be applied to areas with significant localized thermal changes, such as the side of equipment. Furthermore, different division resolutions may be applied to different regions within a single image; for example, a method in which the central area is divided into 16×16 pixel units and the edge areas into 32×32 pixel units may be applied.

[0201] Additionally, the screen splitting unit (530) may apply a variable splitting method based on the region of interest. More specifically, when a specific area is designated as the region of interest, a smaller cell size may be applied to that area to generate grid data of higher resolution. For example, while a general area may be divided into units of 20×20 pixels, the region of interest may be subdivided into units of 5×5 pixels or 8×8 pixels. Additionally, the region of interest may be automatically designated as an area where the amount of temperature change compared to the previous frame has changed by a certain level or more, for example, about 2°C or more, and fine grid splitting may be performed on that area. At this time, the region of interest information is stored in the form of [region coordinates, applied cell size, applied frame range] so that the same splitting structure can be maintained across consecutive frames.

[0202] Additionally, the method for calculating the representative temperature value of each cell can be configured in various ways. For example, a simple average of all pixels within the cell may be used, or in certain embodiments, a weighted average method that assigns higher weight to the center pixel may be applied. Furthermore, the maximum value or the top percentile value within the cell (e.g., the top 10% average) may be selected as the representative value, which may be a setting intended to more sensitively reflect localized high-temperature areas. For example, in the server exhaust area, the highest temperature value within the cell may be selected as the representative value to highlight the heat concentration area.

[0203] Additionally, the screen splitting unit (530) can generate coordinate information to maintain a correspondence between the image coordinate system and the actual space coordinate system. More specifically, the center coordinates (u, v) of each cell can be stored so that they can be used as a starting point for ray projection in the 3D space mapping unit (540) later, and can be recorded in the form of normalized coordinates or pixel coordinates according to the image resolution. For example, in a 640×480 image, if the center of a specific cell corresponds to the position (320, 240), the coordinates can be stored as they are or converted into coordinates (0.5, 0.5) normalized to the range of 0 to 1 and stored.

[0204] Additionally, the screen splitting unit (530) can perform integrated processing for multiple camera inputs. More specifically, images input from the first thermal imaging camera (510) and the second thermal imaging camera (520) can each be independently grid-divided and subsequently aligned and stored in the same data structure. For example, if two camera images with different resolutions are divided into 64×48 and 80×60 grids, respectively, the two grid data can be converted into a data set having the same field structure and transmitted to a subsequent stage. Additionally, camera ID information can be included in each cell data so that source identification is possible in a subsequent stage.

[0205] Additionally, the screen splitting unit (530) may include a data alignment function along the time axis. More specifically, each cell data may be stored with a frame-unit timestamp, and data at the same point in time may be organized into a single data set. For example, cell data generated by splitting images acquired from the first and second thermal cameras at a specific point in time t may be grouped into the same time group and processed simultaneously in a subsequent step. Additionally, cell data may be accumulated and stored in chronological order to enable time-series analysis.

[0206] Additionally, the screen splitting unit (530) may include a data structure for efficiently storing splitting result data. More specifically, each cell data may be stored in the form of a two-dimensional array, and the array index may be configured to indicate the location of the cell. For example, an array element having an index (i, j) may represent a cell at a specific location within the image. Additionally, in a specific embodiment, the cell data may be stored in a hash structure so that it can be quickly accessed using cell coordinates as a key. For example, specific cell data of a specific camera can be immediately retrieved using a key such as "cam1_x32_y24".

[0207] Additionally, the screen splitting unit (530) can perform splitting operations using a parallel processing structure. More specifically, the image can be divided into blocks of a certain size and processed simultaneously by multiple computation threads or GPU computation units, thereby enabling real-time splitting processing even in high-resolution images. For example, a 1280×720 image can be divided into 16 blocks and processed independently, and each block can perform cell splitting operations simultaneously in a separate computation unit.

[0208] Additionally, the screen splitting unit (530) may include a preprocessing function to correct noise or abnormal values ​​within the image. More specifically, if a specific pixel value has a large difference from surrounding pixels, the value may be corrected, and this may be a process to reduce errors caused by sensor noise or reflection effects. For example, a pixel that has a difference of about ±5° or more from the average value of surrounding pixels may be determined as an abnormal value and replaced with the average value of surrounding pixels.

[0210] The 3D space mapping unit (540) can perform the role of precisely assigning grid unit temperature data transmitted from the screen splitting unit (530) to a 3D solid cube structure corresponding to the actual facility space. More specifically, the 3D space mapping unit (540) can calculate the correspondence between the shooting coordinate system of the thermal imaging camera (510, 520) and the 3D coordinate system in virtual space, and map the temperature data to the 3D cube position using a ray generated from the center point of each grid cell.

[0211] More specifically, the 3D space mapping unit (540) may include a calibration module (541) and a raycast operation module (542), and the calibration module (541) may calculate a transformation relationship between an image coordinate system and a real space coordinate system using the lens distortion characteristics and installation location information of each thermal imaging camera (510, 520). For example, the distortion coefficients of the camera lens may be set in the form of radial distortion coefficients (k1, k2, k3) and tangential distortion coefficients (p1, p2), and for a general thermal imaging camera, the k1 value may be set in the range of approximately -0.2 to -0.4, and the k2 value may be set in the range of approximately 0.05 to 0.15. Additionally, the installation position of the camera can be defined by (X, Y, Z) coordinates, for example, the first thermal imaging camera (510) can be set to coordinates such as (0 m, 0 m, 3.0 m) and the second thermal imaging camera (520) can be set to coordinates such as (2.0 m, 1.5 m, 2.8 m). Along with this, the rotation information of the camera can be defined by pitch, yaw, and roll values, for example, the second thermal imaging camera (520) can be set to pitch approximately -45 degrees, yaw 0 degrees, and roll 0 degrees.

[0212] Additionally, the calibration module (541) can generate a transformation matrix for coordinate transformation between real space and virtual space. For example, the intrinsic parameter matrix can be configured to include focal lengths fx and fy and principal points (cx, cy), and the fx and fy values ​​can be set to a range of about 500 to 1200 pixels. The extrinsic parameters consist of a rotation matrix and a translation vector, so that the camera position and orientation can be accurately reproduced within virtual space. Due to these transformation relationships, specific pixel coordinates of an image can be transformed into specific direction vectors in three-dimensional space.

[0213] Additionally, the raycast operation module (542) can generate a ray into three-dimensional space based on the center coordinates of each grid cell and determine whether the ray intersects with the three-dimensional cube. More specifically, each cell center coordinate (u, v) can be converted into normalized coordinates, then converted into a direction vector in the camera coordinate system, and subsequently converted into a virtual space coordinate system to generate a ray. For example, if a 64×48 grid division is performed in one frame, a total of 3072 rays can be generated, and each ray can be projected toward space in different directions.

[0214] Additionally, the three-dimensional space can be configured in the form of solid cubes, and each cube can be defined to have a fixed size. For example, if a data center space has a size of 10 m × 6 m × 3 m, dividing it into cubes of 0.1 m units can generate approximately 100 × 60 × 30 cubes, and a total of approximately 180,000 cubes can be formed. Additionally, in certain embodiments, the cube size may be set to units of 0.05 m or 0.2 m to balance the resolution and the amount of computation.

[0215] Additionally, each ray can be connected to a specific cube through an intersection operation with the cube. For example, sampling is performed at regular intervals, such as 0.05 m to 0.2 m, in the direction the ray travels, and the cube containing each sample location can be searched. In this process, the cube that the ray first collides with or the cube that satisfies specific conditions can be selected. For example, if a physical structure such as a rack surface is present, the ray can terminate at that location, and in an air layer area, the ray can continue to travel up to a certain distance.

[0216] Additionally, multiple rays may be mapped to a single cube, in which case the final temperature value of the cube may be determined in various ways. For example, if three or more rays reach the same cube, an average value may be applied, and in certain embodiments, a maximum value or a weighted average value may be applied. For example, weights may be set differently depending on the angle of incidence of the rays, so that a higher weight is given to rays arriving from the front direction.

[0217] Additionally, the raycast operation module (542) can perform raycast operations in a parallel processing structure to process a large amount of ray operations. More specifically, the raycast operation module (542) can separate the rays generated from each grid cell center point into independent operation units and assign them to a GPU-based parallel operation environment, and can perform cube intersection determination operations for each ray through a separate thread or operation block.

[0218] For example, the raycast operation module (542) can process approximately 3,072 rays generated from a 64×48 grid division structure simultaneously, or process approximately 12,288 rays generated from a 128×96 grid structure in parallel by distributing them among multiple GPU threads. At this time, the raycast operation module (542) is configured to perform sampling at intervals of approximately 0.05 m for each ray, thereby performing approximately 50 to 150 cube inclusion determination operations for a single ray, and can control the execution of 256 to 1,024 rays grouped together in units of thread blocks within the GPU. Additionally, the raycast operation module (542) can be configured to use a CUDA-based kernel operation structure to transmit the starting coordinates, direction vector, and maximum travel distance information of each ray in an array form, and for each thread to independently repeat cube intersection operations using the information.

[0219] Additionally, the raycast calculation module (542) can incorporate depth information obtained in a multi-sensor environment into the ray calculation process. More specifically, the raycast calculation module (542) can be configured to receive a depth value corresponding to the center point of each grid cell and limit the maximum travel distance of the corresponding ray, thereby allowing the intersection position with the actual object surface to be calculated more precisely.

[0220] For example, the raycast operation module (542) can control the ray so that it is no longer sampled in the section after 1.2 m when a depth value corresponding to a specific cell is input as approximately 1.2 m. Additionally, if no depth value exists, the raycast operation module (542) can control the ray's propagation by applying a basic maximum distance value, for example, a range of approximately 3.0 m to 5.0 m.

[0221] In a specific embodiment, the raycast calculation module (542) may be configured to determine the reliability of depth data using a signal strength value provided along with the depth value, and to invalidate the depth value and apply a basic distance value if the signal strength is below a certain level. Additionally, when multiple depth values ​​are input from multiple sensors, the raycast calculation module (542) may determine the final ray length by calculating an average value or a reliability-based weighted average value.

[0222] Additionally, the raycast operation module (542) can store the mapping results in the form of structured data for each cube unit in a three-dimensional space. More specifically, the raycast operation module (542) can assign a unique identifier to each cube and create a data structure for the cube that includes coordinate information (x, y, z), a list of multiple temperature values, a list of input camera identifiers, data generation time information, the number of mappings, and the number of ray intersections.

[0223] For example, the raycast operation module (542) can be configured to store a specific cube in a structure such as [Cube_ID="C_120_45_10", X=1.2 m, Y=0.45 m, Z=1.0 m, Temp_List={28.5, 29.1, 28.9}, Camera_List={510, 520}, Timestamp_List={t1, t2, t3}, Hit_Count=3]. Additionally, the raycast operation module (542) can be configured to additionally store the incident direction vector, incident angle, and sampling depth information of the ray from which the value was derived, along with the temperature value, so that it can be used for spatial reliability evaluation thereafter. The storage structure can be configured in the form of a three-dimensional array, and spatial coordinates can be directly represented through the array index. Additionally, the raycast operation module (542) can be configured to enable fast data access by applying a hash table structure with cube coordinates as keys, for example, converting coordinate information such as "(x=12, y=45, z=10)" into a single key.

[0224] Additionally, the raycast operation module (542) can continuously update the mapping operation results along the time axis. More specifically, the raycast operation module (542) can be configured to update the temperature value of each cube at intervals of approximately 0.1 seconds when video data of approximately 10 frames per second is input, and can store the last 10 to 50 temperature data for each cube in the form of a circular buffer. For example, the raycast operation module (542) can be configured to store the last approximately 5 seconds of temperature data for a specific cube to maintain a total of 50 data points in chronological order.

[0225] Additionally, the raycast operation module (542) can calculate and store the amount of temperature change per time interval for each cube, and, for example, can be configured to set a change state flag on the cube if the amount of temperature change compared to the previous time is about +0.5 ℃ or more. Along with this, if the temperature rise pattern persists for a certain period of time, the cube can be classified into a specific state and stored as separate state information.

[0226] Additionally, the raycast operation module (542) can perform data conversion and transmission functions for linkage with the visualization engine. More specifically, the raycast operation module (542) can convert the temperature value of each cube into a color value and transmit it to the visualization engine, and, for example, when the temperature range is set to 20 ℃ to 40 ℃, a color conversion table can be applied in which 20 ℃ is mapped to blue, 30 ℃ to green, and 40 ℃ to red.

[0227] Additionally, the raycast operation module (542) can be configured to generate object data including coordinate information and color information of each cube and convert it into a cube object on the visualization engine. In a specific embodiment, the raycast operation module (542) can be configured to set the virtual camera to be identical to the position and field of view of the actual thermal imaging camera (510, 520) so that the correspondence between the 2D image and the 3D space can be visually verified. Additionally, the raycast operation module (542) can be configured to transmit time-series temperature data and mapping history information of the cube to a separate interface when a specific cube is selected according to user input.

[0229] The spatial interpolation module (550) can distinguish between a three-dimensional cube object to which a temperature value is directly assigned and a three-dimensional cube object to which a temperature value is not directly assigned through the three-dimensional spatial mapping unit (540), and perform an interpolation operation that reflects physical parameters and data characteristics together for a three-dimensional cube object corresponding to an area that is not directly measured.

[0230] More specifically, the spatial interpolation module (550) can identify among a plurality of three-dimensional cube objects a cube in which a measured temperature value is stored because a ray intersection history exists, and a cube in which a temperature value is empty because a ray intersection history does not exist. After classifying the empty cube as an air layer three-dimensional cube object or a blind spot three-dimensional cube object, the temperature value of the cube can be estimated by considering the temperature state, positional relationship, time change information, air movement characteristics, and heat transfer characteristics of surrounding cubes together. For example, spaces that are difficult to measure directly, such as the air layer above the passage between racks, the hidden space behind the side of the rack, the stagnant area near the ceiling, and the air layer around the rear exhaust of the equipment, can be identified as interpolation target areas.

[0231] More specifically, the spatial interpolation module (550) can first divide the three-dimensional cube objects within the entire three-dimensional space into a set of measured cubes and a set of unmeasured cubes. At this time, the set of measured cubes can be defined as cubes in which one or more temperature values ​​are recorded by the raycast operation module (542), and the set of unmeasured cubes can be defined as cubes in which no actual temperature values ​​are recorded. Additionally, the spatial interpolation module (550) can further divide the unmeasured cubes into cubes representing the interior of the facility, cubes representing the walls of the structure, and cubes representing the air layer.

[0232] For example, using facility outer shape data and depth information, cubes corresponding to solid structures may be excluded from interpolation targets, and only cubes corresponding to empty spaces between racks, upper passages, and unoccupied spaces around facilities may be selected as interpolation targets. Additionally, in certain embodiments, attribute fields such as [cube ID, coordinate information, structure type, measurement status, interpolation target status, existence of recent temperature value] may be stored for each cube, and the spatial interpolation module (550) may select target cubes using these fields.

[0233] Additionally, the spatial interpolation module (550) can construct interpolation input data by searching multiple cubes adjacent to each unmeasured cube. More specifically, the spatial interpolation module (550) can first search for six adjacent cubes located above, below, left, right, front, and rear of a specific unmeasured cube, and, if necessary, can additionally construct a set of 18 or 26 adjacent cubes including diagonal directions. For example, among the 26 cubes surrounding the center cube, only cubes with actual temperature values ​​can be selected and constructed as an input set, and if a cube in a specific direction does not exist or is blocked by a structure, the value for that direction can be left empty.

[0234] Additionally, the spatial interpolation module (550) can record the relative coordinates, distance, and direction vector between each adjacent cube and the center cube together so that the spatial arrangement relationship is reflected in the interpolation input. For example, the upper direction cube can be stored with direction codes such as (0,0,+1), the left cube (-1,0,0), and the front cube (0,+1,0).

[0235] Additionally, the spatial interpolation module (550) can calculate thermal gradient information using the temperature difference between adjacent cubes. More specifically, by comparing the temperature value differences of each direction cube surrounding the center cube, spatial change characteristics indicating which direction the temperature increases or decreases can be extracted. For example, if the temperature value of the lower cube is stored as 24.0 ℃ and the temperature value of the upper cube is stored as 28.5 ℃, it can be determined that there is a characteristic of the temperature rising in the vertical direction. Also, if the left cube is 31.0 ℃ and the right cube is 27.0 ℃, it can be determined that there is a characteristic of heat spreading from left to right.

[0236] In a specific embodiment, the temperature difference for each direction can be stored as a separate field and configured as a data vector in the form of [△x+, △x-, △y+, △y-, △z+, △z-], and the spatial interpolation module (550) can use this as input to the interpolation model. In addition, a value representing local heat concentration in space can be additionally calculated using a combination of multiple directions as well as a single-direction difference.

[0237] Additionally, the spatial interpolation module (550) can generate physical parameters representing changes in air density and natural convection characteristics. More specifically, the spatial interpolation module (550) can extract a directionality in which the rising component is greater near the high-temperature cube and the falling component is greater near the low-temperature cube by reflecting the characteristic that the difference in air density can increase as the temperature difference between cubes increases.

[0238] For example, if the lower cube is 23°C and the upper cube is 29°C, it can be determined that there is an upward heat flow, and conversely, if the upper is 24°C and the lower is 28°C, it can be classified as an abnormal downward temperature pattern and a separate flag can be assigned. In addition, the spatial interpolation module (550) can reflect the natural movement component formed by the heat difference, the forced movement component formed by the equipment, and the local heat generation characteristics formed by the heat source together when there is an air flow direction sensor, air conditioning equipment location information, and heat amount information of the server equipment.

[0239] More specifically, when air flow direction information obtained from an air flow direction sensor and discharge direction and location information of the air conditioning equipment are input, the air movement trend formed in a specific direction can be configured in the form of a parameter. For example, if the discharge direction of the air conditioning equipment is directed toward the center of the passage and the cube temperature in that direction is maintained at a relatively low level, a movement component reflecting the directionality of cold air inflow can be generated.

[0240] In addition, heat output information corresponding to each location can be input based on IT power load data collected from server equipment, and if heat output increases in a specific cube or area, the tendency for heat to diffuse from that location can be additionally reflected. For example, if the server load increases in a specific rack zone, it is reflected that the heat output in the cube corresponding to that location has increased, which can be expressed as a strengthening of the heat diffusion component to surrounding cubes.

[0241] In this way, the spatial interpolation module (550) can generate a plurality of physical parameters that simultaneously include a movement component formed by natural convection, a forced movement component formed by air conditioning equipment, and a heat generation component formed by server heat generation. In a specific embodiment, for each cube, a physical parameter field may be configured in the form of [rising component, falling component, left-right diffusion component, front-back movement component] along with a heat generation component corresponding to the heat intensity.

[0242] Additionally, the spatial interpolation module (550) may include temperature change information over time in the interpolation input. More specifically, it may extract the time change pattern of the same location by reading the most recent N time series temperature values ​​stored in each measurement cube or already interpolated cube.

[0243] For example, if the last 10 temperature values ​​in a specific cube are stored as [27.1, 27.4, 27.9, 28.3, 28.6, 28.9, 29.1, 29.4, 29.7, 30.0], it can be determined that a continuous rising pattern exists. Also, if the last 10 values ​​are maintained as [26.5, 26.6, 26.4, 26.5, 26.6, 26.5, 26.4, 26.5, 26.5, 26.6], it can be classified as a stable state. The spatial interpolation module (550) can extract such time change characteristics for adjacent cubes and reflect them when estimating the temperature of an unmeasured cube. In certain embodiments, time windows of different lengths, such as the last 5 frames, 10 frames, and 30 frames, may be used simultaneously, and short-term changes and long-term changes may be distinguished and configured as separate input channels.

[0244] Additionally, the spatial interpolation module (550) can arrange the input data of the interpolation model into a multidimensional structure and provide it to a physical-data fusion type neural network. More specifically, for a single unmeasured cube, a set of adjacent cube temperature values, a direction code, a relative distance, a time change value, an air movement component, structural type information, and distance information from the facility location can be combined into a single input record.

[0245] For example, the input record may be configured in a form such as [current temperature value of 26 surrounding cubes, average value of the last 5 time points of 26 surrounding cubes, direction vector with respect to the center cube, type of adjacent cube, air movement component, distance from the air conditioning equipment, distance from the rack]. Additionally, in certain embodiments, this input may be rearranged into a 3D tensor form so that temperature, time change, and physical parameters are stored separately for each channel. For example, Channel 1 may be the current temperature, Channel 2 the time change value, Channel 3 the density difference component, and Channel 4 the convection direction component.

[0246] Additionally, the spatial interpolation module (550) can perform neural network operations that reflect physical constraints. More specifically, the spatial interpolation module (550) can extract spatial characteristics from a set of surrounding cubes using a model having a three-dimensional convolution operation structure, and at the same time, can also reflect characteristics of change over time by configuring an input structure that includes time axis data.

[0247] For example, when a 3D CNN structure is applied, the input size can be set to a cube block of size 3×3×3 or 5×5×5, and each cube block can be input to the neural network in the form of multiple channels containing temperature data, thermal gradient information, and airflow-related information. In addition, in certain embodiments, a time-series processing layer may be additionally combined with the 3D CNN structure so that operations reflecting both spatial and temporal information can be performed.

[0248] Additionally, the artificial intelligence model applied to the spatial interpolation module (550) may not be limited to a 3D CNN structure and may be implemented in various forms depending on the structure and processing method of the input data. For example, a Transformer-based model that converts the relationship between cubes into a token form and performs multi-head attention operations may be applied, and a time series prediction model centered on predicting temperature changes over time may also be applied.

[0249] In this way, the spatial interpolation module (550) can perform temperature estimation of an unmeasured area using an artificial intelligence model that can reflect spatial adjacency and temporal change characteristics together, regardless of the applied model structure.

[0250] Additionally, the spatial interpolation module (550) can perform correction operations on results that do not satisfy physical constraints. More specifically, if a value that is excessively high or excessively low compared to adjacent cubes is predicted, the result can be adjusted using the difference from the surrounding cube distribution. For example, if the temperature range of the six cubes surrounding the center cube is 25°C to 28°C and the predicted value is calculated as 34°C, it can be classified as an unrealistic value and re-estimated or corrected.

[0251] Additionally, correction may be performed even when the temperature of the lower cube is consistently lower than that of the upper cube, but only the center cube is predicted to have an abnormally high value. In certain embodiments, automatic correction may be applied when the predicted value exceeds ±3 ℃ or ±5 ℃ relative to the average value of the ambient temperature range, and the value after correction may be stored with a separate status flag.

[0252] Additionally, the spatial interpolation module (550) can store the interpolation results in a cube-unit data structure. More specifically, for each unmeasured cube, the final interpolated temperature value, the time of interpolation execution, the number of surrounding cubes used, the model type used, and the prediction confidence value may be recorded together. For example, for a specific cube, a record such as [Estimated temperature = 27.8 ℃, Estimated time = t15, Number of cubes used = 18, Model = 3D CNN, Confidence = 0.91] may be stored. Additionally, if interpolation values ​​from a previous time point exist for the same cube, a circular buffer structure that maintains only recent values ​​may be applied, and the last 20 or 50 interpolation results may be stored. In a specific embodiment, the direct measurement value and the interpolation value may be maintained in separate lists so that the two values ​​can be distinguished and used in a subsequent analysis step.

[0253] Additionally, the spatial interpolation module (550) may store source information of the input data used during the interpolation process. More specifically, each interpolation result may record which surrounding cubes were used as inputs, what the temperature values ​​of those cubes were, and what directional relationships they had. For example, the interpolation result of a specific cube may be stored as a log record in the form of [Input Cube={C101, C102, C103, C111, C112, C113}, Input Temperature={26.1, 26.4, 26.8, 27.0, 27.2, 27.3}, Direction={Up, Left, Right, Forward, Backward, Down}]. In addition, the record may also include the time window length, physical parameter values, and boundary condition setting values ​​used during the interpolation process. Through this, it is possible to retrospectively track which data determined the predicted value of a specific cube.

[0254] Additionally, the spatial interpolation module (550) can parallelize interpolation operations for a number of unmeasured cubes. More specifically, each unmeasured cube can be separated into independent operation units and distributed to multiple CPU threads or GPU operation blocks for simultaneous processing. For example, if there are approximately 5,000 or 20,000 unmeasured cubes in a single frame, these cubes can be divided into blocks to perform parallel interpolation. Additionally, the spatial interpolation module (550) can improve memory access efficiency by processing cubes located within the same spatial block as a single operation group. In a specific embodiment, adjacent cube data blocks are rearranged so that they are stored contiguously within GPU memory, thereby improving the speed of interpolation operations.

[0255] Additionally, the spatial interpolation module (550) can repeatedly update the interpolation results according to newly input actual measurement data. More specifically, when a new frame is input from the first thermal imaging camera (510) and the second thermal imaging camera (520), the spatial interpolation module (550) can re-check the existing measurement cube and unmeasured cube states and re-perform the interpolation operation by reflecting the changed ambient temperature distribution. For example, if data is updated at a rate of 5 frames or 10 frames per second in a real-time environment, the interpolation operation can also be repeated in the same cycle. In addition, in certain embodiments, a method may be applied in which areas with little change compared to the previous frame are not recalculated, and only areas with concentrated changes are selectively recalculated.

[0256] Additionally, the spatial interpolation module (550) can arrange the interpolation results into a data structure suitable for transmitting to the entire three-dimensional thermal volume generation step. More specifically, all cubes can be maintained in a state containing at least one temperature value among the measured value or the interpolated value, and each cube can be organized into an integrated record structure having the same field configuration.

[0257] For example, the entire set of cubes can be arranged in the format [cube ID, coordinates, final temperature value, data type, recent time series value, confidence, status flag]. Additionally, in certain embodiments, the entire space may be reconfigured into a three-dimensional array to form a continuous data set in which valid temperature values ​​exist for all locations. The result configured in this way can then be directly used for three-dimensional thermal volume rendering or visualization processing.

[0258] In addition, the spatial interpolation module (550) according to one embodiment of the present invention can calculate the estimated temperature by using a physical-data fusion type 3D CNN model in which physical-based constraints, including temperature differences between adjacent cubes, changes in air density, and natural convection characteristics, are reflected in the loss function when estimating the temperature of a three-dimensional cube region that is not directly measured by a thermal imaging camera.

[0259] More specifically, the spatial interpolation module (550) can be configured with temperature data for multiple adjacent three-dimensional cubes centered on an unmeasured cube as input, and the temperature data can be included in the form of time-series data collected over multiple time points. Accordingly, not only the temperature distribution of a specific area but also the characteristics of change over time can be reflected.

[0260] Additionally, the spatial interpolation module (550) can generate thermal gradient information based on the temperature difference between adjacent cubes, and the thermal gradient information can be configured to include the magnitude and direction of the temperature change for each direction. At this time, a constraint condition to prevent the estimated temperature value from forming a sharp difference with adjacent cubes can be included in the loss function.

[0261] Additionally, the spatial interpolation module (550) can generate air density information that changes according to the temperature distribution, and a relative density state corresponding to the temperature of each cube can be calculated. Along with this, natural convection characteristics formed by the difference in air density can be additionally reflected, and constraints can be configured so that the tendency of air movement formed in a specific direction is consistent with the temperature distribution.

[0262] In this case, thermal gradient information, air density change information, and natural convection characteristics are not applied individually but are combined into a single set of physical constraints and can be simultaneously reflected in the learning and estimation processes of the artificial intelligence model. For example, if temperature changes between adjacent cubes appear excessively discontinuous, or if a temperature distribution conflicting with the direction of air movement is formed, the loss function can be configured to increase the error for the corresponding result.

[0263] Additionally, the spatial interpolation module (550) can input temperature data, thermal gradient information and air flow-related information into a 3D CNN model in the form of multi-channel inputs, thereby enabling temperature estimation that reflects spatial proximity and characteristics of change over time.

[0265] The visualization unit (560) receives a set of temperature data in three-dimensional cube units generated through the spatial interpolation module (550) as input, converts the data into a visually representable form, and can perform the role of displaying a three-dimensional thermal volume on a user terminal or management system screen. More specifically, the visualization unit (560) can be configured to convert the temperature value corresponding to each three-dimensional cube object into color information and render the cube object in the three-dimensional space based on the converted color information so that the heat distribution of the entire space can be intuitively checked.

[0266] More specifically, the visualization unit (560) can convert the temperature value of each three-dimensional cube object into a color value according to a predefined color map. For example, temperatures below 20°C can be mapped to a blue color scheme, a range from 20°C to 26°C to a green color scheme, a range from 26°C to 32°C to a yellow color scheme, and temperatures above 32°C to a red color scheme, and the color transition for each temperature range can be set in the form of linear interpolation or a non-linear curve. Additionally, in certain embodiments, the color map can be dynamically adjusted according to environmental conditions, and, for example, if the average temperature inside the data center rises above 28°C, the color reference value can be automatically adjusted upward to emphasize the relative temperature difference.

[0267] Additionally, the visualization unit (560) can generate and update three-dimensional cube objects in real time on a three-dimensional graphics engine. More specifically, each cube object may be composed of a box-shaped three-dimensional mesh having a fixed size, and a color corresponding to the temperature value of the cube may be applied to each mesh. For example, if the size of one cube is set to 0.2 m × 0.2 m × 0.2 m, thousands or more cube objects may be arranged in the entire data center space, and each cube may independently have a color value while simultaneously constituting the entire space. Additionally, in certain embodiments, the boundary lines between cubes may be displayed semi-transparently, or a highlight may be applied to specific areas.

[0268] Additionally, the visualization unit (560) may apply a selective rendering technique considering the efficiency of rendering operations. More specifically, for cube objects that are not visible from the user's viewpoint, rendering operations may not be performed, and cubes located outside the camera's field of view may be excluded from the rendering target. For example, when a user looks at the space from a specific direction, cubes located behind the opposite wall may be processed so that they are not displayed on the screen. Also, in a specific embodiment, cubes with a distance value from the camera of 5 m or more may be replaced with a low-resolution mesh, and if the distance is 10 m or more, they may be reduced and represented as simple color blocks.

[0269] Additionally, the visualization unit (560) can adjust the cube representation resolution according to distance or importance. For example, cubes within a 2 m range around a specific rack set by the user as an area of ​​interest can be rendered with a high-resolution mesh and precise color representation, while other areas can be displayed as simplified color blocks. Additionally, in certain embodiments, the resolution can be set to automatically increase only for areas with rapid temperature changes, for example, areas where the temperature difference between adjacent cubes is 3 ℃ or more can be automatically enlarged.

[0270] Additionally, the visualization unit (560) can display three-dimensional thermal volume data in various ways by combining it with a user interface. More specifically, in addition to a basic screen that displays the entire three-dimensional space, a slice screen that displays a specific layer in a cross-sectional form, a layer screen that selectively displays only a specific height area, and a zoom screen that displays a specific coordinate range in an enlarged manner may be provided. For example, if the user selects a point at a height of 1.5 m, the cross-sectional temperature distribution at that height may be displayed in a two-dimensional planar form, and if the user selects a specific rack number, only the area around that rack may be displayed in an enlarged manner.

[0271] Additionally, the visualization unit (560) can display the temperature change state over time in the form of an animation. More specifically, by continuously playing three-dimensional thermal volume data at each point in time, one can check how the temperature change in the space changes over time. For example, by playing data from the last 10 minutes at 1-second intervals, one can visually check the pattern of a gradual temperature rise occurring in a specific area. In addition, in certain embodiments, the playback speed can be adjusted to 1x speed, 2x speed, 5x speed, etc., and a search function to move to a specific point in time may be included.

[0272] Additionally, the visualization unit (560) can display detailed information about a specific cube or area. More specifically, when a user clicks or selects a specific location, the coordinates of the cube, the current temperature value, recent time series data, whether interpolation is performed, the confidence value, etc., may be displayed in a separate information window. For example, when a specific cube is selected, information such as "Coordinates (x=12, y=8, z=3), Temperature 29.4 ℃, Recent change +0.8 ℃, Data type: Interpolated value, Confidence: 0.93" may be displayed.

[0273] Additionally, the visualization unit (560) can visually highlight and display abnormal conditions. More specifically, if the temperature value of a specific cube exceeds a certain range relative to the average of surrounding cubes, the cube may be displayed with a separate color or icon. For example, if it is more than +4°C higher than the surroundings, a red flashing indicator may be applied, and if it persists for a certain period of time, a warning message may be displayed along with it. Additionally, in certain embodiments, the abnormal area may be displayed by surrounding the entire area with a boundary line.

[0274] Additionally, the visualization unit (560) may include a data output function capable of interacting with an external system, and the three-dimensional thermal volume data may be converted into a structure suitable for network transmission and transmitted to a management server, a control system, or a user terminal. More specifically, the visualization unit (560) may serialize data corresponding to each three-dimensional cube object and configure it in units of transmission packets, and each cube data may be generated in a structure including [cube identifier, coordinates (x, y, z), temperature value array, time information, data type flag (actual measurement / interpolation), number of updates]. For example, for one cube, the temperature values ​​of the last 20 frames may be included in the form of a time-series array, and the data may be encoded and transmitted in the form of a JSON string or a binary packet structure for high-speed transmission.

[0275] In addition, in certain embodiments, to improve data transmission efficiency, the temperature value may be compressed and stored as a 2-byte integer when limited to a certain range (e.g., 20°C to 40°C), and the coordinate information may also be compressed into 16-bit units using a relative coordinate system. Upon receiving the serialized data, the receiving side can restore it into a 3D array structure through a decoding process, and the restored data can be immediately reflected in visualization on the same spatial coordinate system. Furthermore, data transmission may be performed using a WebSocket-based real-time streaming method or a batch transmission method at a fixed interval (e.g., 100 ms, 500 ms), and in the event of network delay or packet loss, a correction display may be performed by maintaining the previous frame data.

[0276] Additionally, the visualization unit (560) can provide a composite visualization function combined with multiple sensor data, and can simultaneously display not only temperature information but also air flow, humidity, and equipment status information within the same three-dimensional space. More specifically, each cube object can store air flow vectors, humidity values, pressure information, etc., in addition to temperature values, and the visualization unit (560) can display the information by separating it into different visual elements.

[0277] For example, air flow information can be represented in the form of a three-dimensional vector arrow starting from the center of the cube, the length of the arrow can be set in proportion to the wind speed value (e.g., in the range of 0.5 m / s to 3.0 m / s), and the direction can reflect the flow direction collected from the sensor. Additionally, humidity values ​​can be represented by the transparency of the cube color or a separate color channel, for example, with low transparency at relative humidity of 40% or less and high transparency at 60% or more.

[0278] Additionally, in certain embodiments, the operating status of the air conditioning system may be displayed as icons around the cube; for example, a blue icon may be displayed when the cooling unit is operating, and a gray icon when it is stopped. This composite visualization can be configured to allow simultaneous assessment of the spatial environment status on a single screen and may include a filter function to selectively display or hide specific sensor data.

[0279] Additionally, the visualization unit (560) can support various display modes, and the visualization method can be dynamically changed according to user requirements. More specifically, in the full space mode, the three-dimensional thermal volume of the entire data center can be displayed, and in the specific zone mode, only cubes corresponding to a selected coordinate range (e.g., x=10~20, y=5~15, z=0~5) can be displayed. Also, in the rack center mode, only cubes within a certain radius around a specific rack (e.g., a range of 1 m to 2 m) can be highlighted, and in the aisle center mode, only floor and air layer temperature distributions in the aisle area can be selectively displayed.

[0280] Additionally, in certain embodiments, a first-person view or a free-movement view may be provided on the user interface, and the user can move the camera view and zoom in or rotate to view specific locations using mouse or touch input. For example, if the user selects the front of a specific rack, the camera view moves to that location, and the cube resolution of that area automatically increases so that a detailed temperature distribution can be viewed. Additionally, in certain embodiments, a function may be provided to save the movement path of the virtual camera and repeatedly play back the same path.

[0281] Additionally, the visualization unit (560) may include a function for data storage and reuse, and three-dimensional thermal volume data at a specific point in time may be stored in the form of a file and reused for subsequent analysis. More specifically, three-dimensional cube array data at each point in time may be stored in file units, and the file may include time information, spatial resolution information, and sensor configuration information.

[0282] For example, a file such as "2026-04-10_14-30-00.vol" may be created, and this file may contain an array of temperature values ​​and time-series data for the entire cube. Additionally, the database system may be configured to apply an index structure using time information as a key to enable rapid retrieval of data within a specific time range. For instance, methods such as selecting and loading only the data from the last hour or comparing data from the same time period on a specific date may be applied.

[0283] In addition, in certain embodiments, the temperature difference between two points in time can be calculated and the difference value can be expressed in color; for example, an area that has risen by +2 ℃ or more compared to the previous point in time can be displayed in red, and an area that has fallen by -2 ℃ or less can be displayed in blue. Through this comparison function, changes in heat distribution over time can be intuitively confirmed, and it can be configured to enable tracking of abnormal changes in specific areas.

[0285] FIG. 5 is a diagram illustrating the concept of cross-arrangement and three-dimensional spatial mapping of a multi-angle thermal imaging camera according to an embodiment of the present invention.

[0286] Referring to FIG. 5, a ceiling may be formed at the top of the facility space, and a plurality of server racks may be arranged facing each other at the bottom of the ceiling. A passageway may be formed between each server rack, and this passageway may be set as an area where air flow is concentrated and where significant temperature changes occur. Additionally, a floor surface may be located at the bottom of the passageway, and this floor surface may be set as a primary observation area for determining the state of cold air inflow or heat accumulation.

[0287] The first thermal imaging camera (510) can be installed at a ceiling position above the center of the passageway and can capture an area including the floor surface of the passageway through a viewing angle set in a downward direction. More specifically, the first thermal imaging camera (510) can be configured to have a vertical viewing angle of approximately 90 degrees, and accordingly, the temperature distribution of the floor area at the center of the passageway and the lower air layer formed along that area can be measured. For example, if the distribution of cold air supplied through the floor tiles is uneven in a specific section, the temperature deviation can be confirmed through the first thermal imaging camera (510).

[0288] The second thermal imaging camera (520) may be installed in the upper corner area of ​​each server rack and may be positioned in a tilted direction toward the front or rear of the rack. More specifically, the second thermal imaging camera (520) may be set to have a tilt in the range of about 30 to 60 degrees, and accordingly, the temperature distribution of the exhaust area where heat emitted from equipment inside the rack is concentrated and the center of the rack can be measured. For example, if high-heat equipment is concentrated in a specific rack, the pattern of heat generated in that area spreading into the surrounding air layer can be confirmed through the second thermal imaging camera (520).

[0289] In this way, by arranging the first thermal imaging camera (510) and the second thermal imaging camera (520) to have different viewing angles, temperature data for the floor surface of the passageway and the side area of ​​the rack can be obtained simultaneously. At this time, since the field of view of each camera is limited by the installation position and the angle of view, some areas of the air layer between the racks may remain as areas where direct measurement data does not exist.

[0290] These areas are spaces composed of air layers and may correspond to sections where heat emitted from the rack interacts with cold air entering from the floor. More specifically, they may be regions where a temperature gradient is formed as high-temperature air rising from the top of the rack mixes with low-temperature air supplied from the bottom of the passageway; such regions may be characterized by concentrated heat transfer and convection phenomena within the space. Furthermore, since these regions are not included within the field of view of a thermal imaging camera, direct temperature measurement may not be performed.

[0291] Such a non-measured air layer region (A) may be difficult to resolve with a single-directional sensing structure, and even when multiple cameras are deployed, some space may remain as a measurement gap. Accordingly, in one embodiment of the present invention, a process for estimating the temperature of the non-measured air layer region (A) can be performed using temperature data measured at an adjacent location and heat transfer characteristics within the space.

[0292] For example, if the temperature measured at the front of the rack is approximately 31°C and the temperature measured at the floor of the aisle is approximately 24°C, it can be determined that a distribution in which the temperature changes continuously depending on the location is formed in the air layer between the two areas. In addition, if the temperature distribution changes over time, time-series temperature data can be reflected together so that the pattern of temperature change within the air layer can be continuously updated.

[0293] In addition, multiple camera fields of view may overlap in some areas, and in this case, multiple temperature data may be acquired for the same location. In such cases, processing to correct sensor errors by comparing the deviations between each data point may be performed. For example, if values ​​of 29.8 ℃ and 30.6 ℃ are measured at the same location, an average value or a confidence-based weighting value may be applied to determine the final temperature value.

[0295] FIG. 6 is a flowchart illustrating the process of mapping three-dimensional solid cubes based on raycast-based two-dimensional temperature data according to one embodiment of the present invention.

[0296] Referring to FIG. 6, the process of mapping three-dimensional solid cubes based on raycast-based two-dimensional temperature data according to one embodiment of the present invention can be performed by a two-dimensional grid division step (S601), a virtual camera calibration step (S602), a virtual ray projection step (S603), a three-dimensional intersection coordinate calculation step (S604), and a three-dimensional temperature mapping step (S605).

[0297] First, in the two-dimensional grid segmentation step (S601), a two-dimensional temperature image acquired from a thermal imaging camera can be segmented into a grid structure according to a set resolution. More specifically, the input image can be segmented into multiple cells in the horizontal and vertical directions to generate temperature data for each cell, and each cell can store a representative temperature value, cell center coordinates, location information within the image, and shooting time information together. For example, if an image with a resolution of 640×480 is segmented into a 64×48 grid, each cell can correspond to an area of ​​approximately 10×10 pixels, and the center position of each cell can be set as a reference coordinate for subsequent three-dimensional spatial mapping.

[0298] Next, in the virtual camera calibration step (S602), a correspondence relationship between the actual shooting coordinate system and the virtual camera coordinate system can be calculated based on the lens distortion coefficient and installation position information of each thermal imaging camera. More specifically, the image coordinates can be corrected by applying the radial distortion coefficient and the tangential distortion coefficient, and the projection matrix of the virtual camera can be generated by reflecting the camera's position coordinates and field of view information. For example, coordinate deviations occurring at the outer edge of the image due to lens distortion can be corrected in units of a few pixels, and if the camera installation height is 2.5 m and the field of view is 90 degrees, the corresponding conditions can be reflected in the virtual camera.

[0299] Next, in the virtual ray projection step (S603), a ray can be projected in a three-dimensional spatial direction from the center point of each grid cell based on the corrected virtual camera viewpoint. More specifically, the center coordinates of each cell can be set as a starting point for generating a ray in three-dimensional space, and the ray can be set to travel a certain distance along the camera's field of view. For example, if there are thousands of cell center points in a single frame, the same number of rays can be generated, and each ray can independently perform exploration in three-dimensional space.

[0300] Next, in the 3D intersection coordinate calculation step (S604), the position coordinates where the projected ray intersects the 3D solid cube structure can be calculated. More specifically, a plurality of solid cubes divided into a certain size may be arranged in the 3D space, and each ray may sequentially determine whether it collides with the corresponding cubes. For example, if a single ray passes through multiple cubes, the first cube that intersects or a cube satisfying a specific condition may be selected, and the spatial coordinates of the cube may be determined as the mapping target location. Additionally, in a specific embodiment, the operation may be set to terminate when the ray travel distance is greater than a certain distance.

[0301] Finally, in the 3D temperature mapping step (S605), temperature data corresponding to each grid cell can be mapped to the calculated 3D coordinates. More specifically, temperature values ​​generated from a 2D grid can be assigned to corresponding cube positions, and if multiple temperature values ​​are input for the same cube, an average value or a weighted value can be applied to determine the final temperature value. For example, if temperature values ​​are input from different viewpoints or different cameras for a single cube, an integrated temperature value can be calculated by considering the reliability of each value or the input time.

[0303] FIG. 7 is a flowchart illustrating the process of estimating the temperature of a blind spot using a physical-data fusion model according to one embodiment of the present invention.

[0304] Referring to FIG. 7, the blind spot temperature estimation process according to one embodiment of the present invention

[0305] It can be performed in the step of inputting adjacent cube actual temperature data and physical constraints (S701), the step of processing a physical-data fusion type artificial intelligence model (S702), and the step of estimating the temperature of the blind spot air layer (S703).

[0306] First, in the step of inputting actual temperature data and physical constraints of adjacent cubes (S701), physical constraints corresponding to thermodynamic characteristics within the space may be input along with temperature data of three-dimensional cubes whose temperatures have been directly measured. More specifically, temperature values, coordinate information, and time-series data for multiple adjacent cubes centered on an unmeasured air layer cube may be included, and at the same time, thermal gradient information based on the temperature difference between cubes and flow characteristics based on natural convection and changes in air density may be configured together. For example, if a temperature difference occurs in a specific direction, the heat transfer trend in that direction may be reflected, and upward or downward airflow characteristics based on changes in air density according to temperature may be input together.

[0307] Next, in the physical-data fusion type artificial intelligence model processing step (S702), the input temperature data and physical constraints can be input into the artificial intelligence model and processed in a combined form. More specifically, adjacent cube temperature values, thermal gradient information, and air flow characteristics can be combined in a multidimensional array form and configured as model input, and the data can be configured to include a time-series structure. For example, temperature data collected over multiple time points can be input in an accumulated form so that temperature change characteristics over time are reflected together, thereby enabling estimation that reflects physical characteristics rather than simple spatial interpolation.

[0308] Finally, in the blind spot air layer temperature estimation step (S703), the temperature value of an air layer cube whose temperature is not directly measured can be calculated based on the processing result of the artificial intelligence model. More specifically, for each unmeasured cube, the estimated temperature value can be determined in a way that maintains temperature continuity with surrounding cubes, and a result reflecting thermal gradients and air flow characteristics can be generated. For example, if there is a tendency for the temperature to gradually increase in a specific area, the temperature value of the corresponding unmeasured area can be set to follow a similar trend, and in the case of abrupt temperature changes, an unrealistic distribution can be suppressed by physical constraints.

[0310] FIG. 8 is a diagram illustrating the concept of visualization of three-dimensional thermal volume data and air flow analysis according to one embodiment of the present invention.

[0311] Referring to FIG. 8, FIG. 8(a) first illustrates the basic structure of a facility space in which server racks are installed. More specifically, a plurality of server racks may be arranged in parallel at regular intervals, and a passageway may be formed between each rack. Additionally, air conditioning equipment supplying cold air may be installed along the ceiling or wall of the space, and a cold air inflow path and a heat exhaust path may be formed together. For example, a structure may be formed in which cold air is supplied through the floor surface and hot air is exhausted through the rear or top of the rack. In such a structure, the temperature distribution within the space may vary by location depending on the heat generation characteristics of the equipment inside the rack, and different heat distribution characteristics may appear in the center of the passageway, the side of the rack, and the area around the equipment.

[0312] Next, FIG. 8(b) can represent a visualized state of temperature distribution for each location based on the structure of FIG. 8(a). More specifically, spatial data composed of three-dimensional cube units can be represented by color values, and temperature values ​​can be individually displayed for each rack area and aisle area. For example, if a temperature of approximately 30°C or higher is consistently observed in a specific rack section, that location can be represented as a relatively high heat accumulation area, while a low-temperature area of ​​approximately 24°C can be formed in the lower part of the aisle. In addition, temperature distributions for various equipment sections, such as power equipment areas, battery areas, and areas adjacent to air conditioning equipment, can be displayed together, and the heat volume distribution for the entire space can be continuously represented. Through such a visualization structure, locations where temperature deviations occur within the space and heat concentration areas can be intuitively identified.

[0313] Next, FIG. 8(c-1) illustrates the airflow and heat distribution state when air conditioning equipment is operated with uniform output and direction in the same spatial structure as FIG. 8(a) and FIG. 8(b). More specifically, when cold air is supplied with the same airflow rate from multiple air conditioning units, a phenomenon may occur where airflows facing each other collide at the center of the passage. In this airflow collision zone, an area where the airflow is not dispersed and becomes stagnant may be formed, and heat accumulation may occur in that area. For example, as high-temperature air rising from the center of the passage between racks mixes with cold air flowing in from both sides, irregular turbulence may be formed, and consequently, a localized high-temperature area of ​​approximately 32°C or higher may be formed at a specific location. Furthermore, if the airflow does not have a constant directionality and repeatedly collides and diffuses, a state may occur where the cold air does not sufficiently reach some areas.

[0314] On the other hand, FIG. 8(c-2) can show the air flow and heat distribution state when the air volume and spray direction of the air conditioning equipment are adjusted in the same spatial structure as FIG. 8(c-1). More specifically, some air conditioning equipment may operate with reduced air volume, and other air conditioning equipment may be set so that the spray direction is tilted in a specific direction.

[0315] Under this control, a structure can be formed in which the airflow moves along a constant circulation path, and cold air can be continuously distributed throughout the entire space. For example, a flow can be formed in which cold air supplied from one air conditioning unit moves along a passage and then circulates to the opposite side, and in this process, heat around the rack can be continuously removed. In addition, as airflow collisions are reduced, the generation of turbulence is suppressed, and the airflow can be maintained in a state with a constant directionality. Accordingly, the localized high-temperature area shown in Fig. 8(c-1) can be alleviated, and a state in which a uniform temperature distribution of approximately 26°C is formed throughout the entire space can be realized.

[0317] FIG. 9 is a flowchart illustrating the entire process of a data center targeting air conditioning control method based on three-dimensional thermal volume prediction according to an embodiment of the present invention.

[0318] Referring to FIG. 9, a data center targeting air conditioning control method based on three-dimensional thermal volume prediction according to one embodiment of the present invention can be performed by a three-dimensional thermal volume data collection step (S901), an AI-based heat island prediction step (S902), a local cooling load calculation step (S903), a step of inverse calculation of the pre-operation time of the air conditioning equipment (S904), and a step of performing asymmetric cooperative air conditioning control (S905).

[0319] First, in the 3D thermal volume data collection step (S901), a time-series multidimensional data tensor can be constructed by combining the 3D thermal volume data generated through a physical-data fusion interpolation-based 3D thermal volume generation method with IT power load data collected from server equipment corresponding to each 3D cube.

[0320] More specifically, in the 3D thermal volume data collection step (S901), 3D thermal volume data including temperature values ​​corresponding to each 3D cube can be collected while the facility space is divided into a plurality of 3D cubes by a physical-data fusion interpolation-based 3D thermal volume generation method.

[0321] Here, 'temperature data' may refer to temperature values ​​assigned to individual three-dimensional cubes and may consist of unit information representing the thermal state at each cube location. Additionally, 'thermal volume data' may refer to a three-dimensional array structure in which temperature data corresponding to each of a plurality of three-dimensional cubes is spatially combined and may consist of a data set representing a continuous temperature distribution over the entire facility space.

[0322] Thermal volume data can be generated as a result of performing raycast-based 3D mapping and spatial interpolation on 2D temperature images acquired from multiple thermal imaging cameras, and can be composed of a continuous 3D temperature distribution that includes an air layer region not included in the camera's field of view.

[0323] For example, different temperature data can be assigned to each three-dimensional cube for the floor surface of the passageway area between server racks, the intermediate air layer, and the upper exhaust area, respectively, and time-series thermal volume data can be formed by accumulating these temperature data in chronological order.

[0324] Additionally, in the 3D thermal volume data collection step (S901), IT power load data can be collected from server equipment corresponding to the spatial coordinates of each 3D cube. More specifically, data including power consumption values, computational resource utilization rates, or power supply status can be collected for each server rack or individual server equipment, and the data can be mapped to a 3D cube corresponding to the physical location where the equipment is installed. For example, if a specific rack is located within a certain coordinate range, the same power load value may be assigned to multiple 3D cubes included in that range, or a weighted distributed value may be assigned according to the equipment distribution within the rack.

[0325] In addition, in the 3D thermal volume data collection step (S901), temperature data and power load data can be aligned based on the same time to form time series data. More specifically, if the collection cycles of temperature data and power data are different, resampling can be performed at regular time intervals, thereby generating data of the same structure for each time step. For example, if temperature data is collected 10 times per second and power data is collected once per second, the same value may be repeatedly applied to the power data, or an intermediate value may be generated through interpolation between adjacent time point data.

[0326] Additionally, in the 3D thermal volume data collection step (S901), the collected data can be combined into a multidimensional array form including a time axis and a 3D spatial axis to form a time series multidimensional data tensor. More specifically, the data tensor can be formed into a structure including time, spatial coordinates (x, y, z), and data channels, and each channel may include a temperature value, a power load value, a temperature change amount, and a power change amount. For example, data corresponding to a specific coordinate at a specific time point can be structured to include a temperature value, a power load value, a temperature change rate, and a power change rate, and such data can be accumulated along the time axis.

[0327] In addition, in the 3D thermal volume data collection step (S901), input data for dynamic threshold calculation and heat island prediction performed in subsequent steps can be generated through the configuration of a data tensor. More specifically, as a structure is formed in which temperature values ​​and power load values ​​exist simultaneously for each 3D cube, heat generation patterns due to load fluctuations of server equipment can be spatially reflected. For example, even if the temperature in a specific area is below a reference temperature, if the power load corresponding to that area increases rapidly, the dynamic threshold can be set to a low value in subsequent steps, and accordingly, the area can be identified early as a heat island-prone area.

[0328] In addition, in the 3D thermal volume data collection step (S901), data types can be distinguished and stored according to the data generation method for each 3D cube. More specifically, temperature values ​​directly measured by a thermal imaging camera and temperature values ​​estimated by spatial interpolation can be distinguished and stored, and each data may include status information indicating whether measurement has been taken. For example, the area included in the camera's field of view may be displayed as a measured value, and the area of ​​the air layer blind spot may be displayed as an interpolated value.

[0329] Additionally, in the 3D thermal volume data collection step (S901), a data storage structure may be configured. More specifically, data for each 3D solid cube may be stored in a structure including coordinates, a list of temperature values, a list of power load values, and time information, and such data may be managed using a spatial coordinate-based index and a time-based index. For example, a coordinate-based index may be formed to quickly retrieve data at a specific location, and a time index may be configured together to extract data for a specific time interval.

[0330] Additionally, in the 3D thermal volume data collection step (S901), correction processing for outliers and missing data may be performed during the data collection process. More specifically, if data is not collected in a specific time interval, the value may be interpolated using data from a previous time point, and if an abnormally large change occurs, the value may be corrected through filtering. For example, if a rapid temperature change occurs within a short period, the value may be replaced with the average value of an adjacent time point.

[0331] In addition, according to various embodiments, in addition to temperature data and power load data, air flow information, humidity information, or pressure information may be additionally included, and such data may be mapped to the same three-dimensional cube structure to form an extended data tensor. For example, temperature, power load, air flow rate, and humidity values ​​may be stored together for a specific cube, thereby enabling more precise thermal behavior analysis to be performed in a subsequent step.

[0333] Next, in the AI-based heat island prediction step (S902), the time-series multidimensional data tensor generated in the 3D heat volume data collection step (S901) is input into an artificial intelligence model to predict future temperature changes for each 3D cube zone, calculate a dynamic threshold that is variably set according to the hardware limit temperature of the server equipment and the power load fluctuation rate, and identify zones predicted to exceed the dynamic threshold as heat island-prone zones.

[0334] In addition, by analyzing the connectivity status and time-dependent change patterns of the temperature distribution within a three-dimensional space, the artificial intelligence model can preemptively detect signs of topological anomalies, such as areas where air flow is stagnant or areas where the temperature rises intensively in specific regions.

[0335] More specifically, if the temperature distribution between adjacent three-dimensional cubes is continuously concentrated in a specific area, that area may be determined as a heat island formation zone, and if the temperature change remains locally without spreading to surrounding areas, it may be identified as an air conditioning convection stagnation zone. The heat island potential zone and convection stagnation zone identified in this way may be set as target areas for air conditioning equipment control in a subsequent step.

[0336] Additionally, the method may include the step of setting a reference threshold temperature using the maximum allowable temperature and current temperature of the server equipment corresponding to each three-dimensional cube; and the step of variably adjusting the reference threshold temperature according to the rate of change of power load of the server equipment.

[0337] Here, the reference threshold temperature may refer to an initial reference value set based on the difference between the server equipment's maximum allowable temperature and the current temperature, and the dynamic threshold may refer to a final judgment reference value adjusted by reflecting the rate of change in the server equipment's power load to the reference threshold temperature. In other words, the reference threshold temperature can be set as a static reference value, and the dynamic threshold can be calculated by adjusting the reference threshold temperature upward or downward in response to power load conditions that change over time.

[0338] For example, if the allowable temperature of a specific server device is 80°C and the current temperature is 60°C, the reference threshold temperature can be set to a certain level, and if the power load increases rapidly at the same time, the reference threshold temperature is adjusted to decrease, so that the dynamic threshold is set to a level of 70°C or lower. Conversely, if the power load is maintained stably, the reference threshold temperature is maintained or changes gradually, so that the dynamic threshold is set to a relatively high value.

[0339] Therefore, even under the same temperature conditions, the reference threshold temperature can be adjusted according to the rate of change of power load to determine the dynamic threshold, and finally, the determination of the heat island-prone area can be performed based on the dynamic threshold.

[0340] Additionally, the step of identifying a heat island-prone area may include: a step of predicting temperature changes over time using an artificial intelligence model that takes as input temperature data and power load data of three-dimensional cubes collected over multiple time points; and a step of identifying a high-temperature area that is continuously formed in a specific area based on the connectivity of the temperature distribution between adjacent three-dimensional cubes. More specifically, if a temperature rise occurs continuously between adjacent cubes within a three-dimensional space, it may be determined that a spatially extended heat accumulation area is formed rather than a single point. For example, if a temperature rise starting from the front of a server rack spreads and connects to the upper exhaust area, the entire area may be identified as a single heat island-prone area.

[0341] In addition, the artificial intelligence model can be configured to take a time-series multidimensional data tensor as input and simultaneously reflect spatial and temporal characteristics. For example, a convolution-based structure reflecting a three-dimensional spatial structure and a time-series processing structure reflecting temporal change characteristics can be combined, and in certain embodiments, an attention structure that assigns weights to areas where temperature changes are concentrated may be included. Through this, rather than a simple comparison of temperature thresholds, a prediction can be performed that reflects heat accumulation patterns due to increased load and heat stagnation phenomena within the air layer.

[0342] In addition, the occurrence of a heat island can be determined by combining the temperature prediction results calculated by an artificial intelligence model with a dynamic threshold. More specifically, if the predicted temperature value in a specific three-dimensional cube zone exceeds the dynamic threshold of that zone, that zone can be identified as a heat island-prone zone where heat accumulation intensifies after a certain period of time. For example, if the predicted temperature of a specific zone is 72°C and the dynamic threshold is set to 68°C, that zone can be determined as a danger zone.

[0343] In addition, information regarding areas where heat islands are expected to occur can be stored in the form of structured data according to a three-dimensional coordinate system and can be directly input into local cooling load calculations and HVAC control calculations performed in subsequent stages. More specifically, for each three-dimensional cube, a predicted temperature value, a dynamic threshold value, and status information indicating whether a heat island occurs can be configured as a single data record, and the data can be stored in a form that includes spatial coordinates and time information. For example, data for a specific cube can be stored in a structure such as [coordinates (x, y, z), time t, predicted temperature value, dynamic threshold, risk flag], and a three-dimensional heat risk distribution map can be formed by collectively configuring this data for the entire space.

[0344] In addition, data regarding areas expected to develop into urban heat islands can be accumulated and stored in a time-series format, and patterns of heat risk change over time can be analyzed. More specifically, predicted temperature values ​​and dynamic thresholds can be recorded across multiple time points for the same three-dimensional cube, thereby identifying trends in which the likelihood of urban heat island development increases or decreases in a specific area. For example, if the predicted temperature in a specific area continuously rises for a certain period and the extent of the excess relative to the dynamic threshold widens, that area can be classified as a high-risk area and set as a priority control target area in a subsequent stage.

[0345] In addition, data regarding areas expected to experience heat islands can be managed using a spatial indexing structure, which can improve the speed of data access to specific areas. More specifically, a hash-based index or tree structure can be configured based on 3D spatial coordinates, allowing for the rapid retrieval of sets of cubes included within a specific coordinate range. For example, cube data included within a certain radius around a specific rack can be extracted in real time to immediately analyze the thermal risk status of that area.

[0346] In addition, data regarding areas expected to generate urban heat islands can be extended with additional parameters for cooling load calculations performed in subsequent steps. More specifically, for each cube, the rate of change in temperature, the rate of change in power load, or the temperature difference value with adjacent cubes can be stored together, and these values ​​can be combined and used to calculate the local cooling requirement. For example, even if two areas have the same temperature level, the area with a faster rate of temperature increase may be determined to be the area requiring a higher cooling load.

[0347] In addition, according to various embodiments, a correction process using the difference between the predicted value and the actual measured value may be included to improve the accuracy of the prediction result, thereby allowing the parameters of the artificial intelligence model to be continuously updated. More specifically, the error between the temperature data actually measured after a certain period of time and the temperature data predicted at the previous point in time may be calculated, and the error may be reused as model training data to adjust the internal parameters of the model. For example, if repeated under-prediction occurs in a specific area, the model weights may be adjusted so that the sensitivity to temperature rise for that area increases.

[0348] Furthermore, this correction process can be configured as a closed-loop learning structure and can be performed iteratively in a manner where prediction results are reflected back into the input data. More specifically, prediction results from previous time points, actual measurement results, and information on power load changes can be combined to form new training data, thereby allowing the model to be updated to adapt to environmental changes. For example, if temperature distribution characteristics change due to seasonal variations or changes in data center operating conditions, these changes can be reflected in the model training process to maintain subsequent prediction accuracy.

[0349] In addition, according to various embodiments, if the prediction error for a specific area is repeated beyond a certain threshold, a separate correction factor may be applied to that area, thereby enabling correction that reflects spatial characteristics. For example, since a temperature rise may be greater in a specific rear rack area where airflow is stagnant even under the same conditions, a weighted correction may be applied to the prediction result for that area.

[0350] In addition, the prediction result data can be provided in a format compatible with external systems, allowing the thermal risk status to be checked in real time on an administrator terminal or control system. For example, a 3D thermal risk distribution map can be output in a visualized form, displaying the risk level of a specific zone by distinguishing it with colors, which enables the operator to make immediate response decisions.

[0352] Next, in the local cooling load calculation step (S903), the difference between the predicted temperature and the target maintenance temperature is calculated for the heat island area identified in the AI-based heat island prediction step (S902), and the cooling requirement for each cube unit can be calculated using the volume of the 3D cube and the time-series temperature change value.

[0353] More specifically, the local cooling load calculation step (S903) may include a step of calculating the difference between the predicted temperature and the target maintenance temperature. In this case, the predicted temperature may be a value derived by an artificial intelligence model that takes a time-series multidimensional data tensor as input, and the target maintenance temperature may be a reference temperature set according to the data center operation policy. For example, if the predicted temperature in a specific zone is 72 ℃ and the target maintenance temperature is 65 ℃, the temperature difference may be calculated as 7 ℃. The temperature difference calculated in this way may be used as a basic indicator representing the amount of heat to be removed in that zone.

[0354] Additionally, the local cooling load calculation step (S903) may include a step of calculating the cooling requirement for each cube unit using the volume of the three-dimensional cube corresponding to the difference and the time-series temperature change value. More specifically, each three-dimensional cube may be defined as a spatial unit having a certain volume, and the cooling requirement may be calculated by reflecting the thermal characteristics of the air and equipment contained in the volume. For example, even with the same temperature difference, a higher cooling load may be calculated if the cube volume is larger, as more heat removal is required.

[0355] Additionally, in the local cooling load calculation step (S903), time-series temperature change values ​​are also reflected to model the dynamic thermal state. More specifically, if the temperature rises rapidly in a specific zone, that zone may be determined as a zone where heat accumulation will increase rapidly in the future, and accordingly, the cooling demand may be calculated in a direction that increases. For example, if one of two zones with the same temperature difference is in a stable state with almost no temperature change and the other is in a state of rapid increase, a higher cooling load may be allocated to the latter zone.

[0356] In addition, in the local cooling load calculation step (S903), a heterogeneous thermal state can be modeled by reflecting the temperature distribution and air flow characteristics between adjacent three-dimensional cubes in a three-dimensional space. More specifically, a tendency for heat to accumulate when air flow is stagnant in a specific area can be reflected, and a characteristic of heat to spread rapidly in an area where air flow is smooth can be reflected. For example, since high-temperature air may be concentrated in the exhaust area at the rear of the server rack and heat accumulation may occur, the cooling load in that area can be calculated to increase.

[0357] Additionally, in the local cooling load calculation step (S903), power load information included in the time series multidimensional data tensor is reflected together so that the amount of heat generated can be reflected more accurately. More specifically, if the power load of server equipment placed in a specific zone is high, it can be determined that the amount of heat generated in that zone increases, and accordingly, the cooling requirement can be calculated in a direction that increases. For example, if the server power load is 2 kW in one of two zones with the same temperature state and 5 kW in the other, a higher cooling load can be calculated for the latter zone.

[0358] In addition, in the local cooling load calculation step (S903), the cooling requirements calculated for each three-dimensional cube can be spatially combined to calculate the total cooling load for the entire area where a heat island is expected to occur. More specifically, after the cooling requirements are calculated for each of the multiple cubes constituting a specific heat island area, the values ​​are summed to derive the total cooling capacity required for that area. For example, if a cooling load of 0.5 kW is calculated for each of the 20 cubes constituting a specific heat island area, a cooling requirement of approximately 10 kW can be derived for the entire area.

[0359] In addition, the local cooling load calculation step (S903) can be structured so that the calculated cooling load can be directly linked to the air conditioning equipment control performed in subsequent steps. More specifically, the cooling requirement for each 3D cube or heat island area can be stored along with coordinate information, thereby allowing the required cooling capacity for the air conditioning equipment corresponding to a specific location to be quantitatively transmitted. For example, if a cooling load of 8 kW is required for a specific coordinate area, that value can be directly reflected in the fan speed or cooling water supply setting of the air conditioning equipment.

[0360] In addition, according to various embodiments, the local cooling load calculation process can be performed repeatedly, and the cooling requirement can be updated whenever new temperature data and power load data are input. More specifically, the cooling load can be recalculated at regular intervals to reflect thermal conditions that change over time, thereby enabling dynamic control that responds to the data center environment changing in real time. For example, if the server load in a specific zone decreases, the cooling load in that zone can be updated in a direction that reduces it.

[0361] In addition, according to various embodiments, the result of the cooling load calculation can be used as an input for a targeting control calculation performed in a subsequent step, and the required cooling capacity for each air conditioning unit can be directly set. For example, the fan speed and cooling water valve opening rate of the air conditioning unit can be set as values ​​corresponding to the cooling load calculated for a specific zone, thereby enabling precise control at the local unit level rather than the entire air conditioning system.

[0363] Next, in the step of calculating the time of pre-operation of the air conditioning equipment (S904), adjacent air conditioning equipment is identified using the three-dimensional spatial coordinates of the area where the heat island is expected to occur identified in the AI-based heat island prediction step (S902), and the time of pre-operation can be calculated using time values ​​corresponding to the mechanical operation delay time and the rate of temperature change of each air conditioning equipment.

[0364] More specifically, the step of calculating the time of pre-operation of the air conditioning equipment (S904) may include a step of identifying air conditioning equipment within a certain distance based on the 3D coordinates of the area where the heat island is expected to occur. At this time, the air conditioning equipment may be equipment capable of performing local cooling for a specific area, such as an in-low cooler, an air conditioner, or a cooling unit placed within the data center.

[0365] For example, if a specific heat island is expected to form within a coordinate range of (x, y, z), multiple air conditioning units located within a certain radius centered on those coordinates may be selected as candidates, and the unit to be controlled may be determined based on the unit's installation location, air discharge direction, and capacity. Additionally, if multiple air conditioning units exist within the same distance, the spatial range that the units can affect is analyzed, and the unit capable of providing the most direct cooling effect to the specific area may be selected as a priority.

[0366] In addition, the step of calculating the time of operation in advance for the air conditioning equipment (S904) may include a step of calculating the time of operation start using a time value corresponding to the mechanical operation delay time and the rate of temperature change of each air conditioning equipment. More specifically, the air conditioning equipment may experience a certain time delay from the input of a control signal until actual cooling performance is manifested, and this delay time may be determined by the fan driving characteristics of the equipment, cooling water circulation delay, or heat exchange delay. For example, in a specific air conditioning equipment, there may be a delay of about 10 seconds from the increase in fan speed until cooling air is actually discharged, and an additional tens of seconds may be required until the cooling water temperature stabilizes.

[0367] In addition, the rate of temperature change can be calculated using the rate of change of temperature and the rate of change of power load included in the time-series multidimensional data tensor, through which the remaining time until the temperature reaches the target temperature in the area where the heat island is expected to occur can be estimated. For example, if the temperature in a specific area rises by 0.5°C per second, and the difference between the current temperature and the target maintenance temperature is 5°C, it can be predicted that the area will reach a dangerous state in about 10 seconds.

[0368] In addition, the pre-operation timing can be calculated by combining the expected temperature rise time and the mechanical delay time of the air conditioning equipment. More specifically, the operation start time can be set so that control begins at a point equal to the delay time of the air conditioning equipment before the scheduled time of heat island occurrence. For example, if it is predicted that a heat island condition will form in a specific zone after 60 seconds and the total delay time of the air conditioning equipment is 20 seconds, the air conditioning equipment can be set to operate in advance at approximately 40 seconds. In this way, preemptive cooling can be performed at a stage prior to the occurrence of heat accumulation.

[0369] Additionally, in the step of back-calculating the pre-operation time of the air conditioning equipment (S904), thermal inertia characteristics within the space may be taken into account. More specifically, temperature changes may be delayed or accelerated depending on the amount of air contained in a specific zone, the heat storage characteristics of surrounding structures, or air flow characteristics, and the start time of operation may be adjusted by reflecting these characteristics. For example, since heat may accumulate rapidly in zones where air flow is stagnant, the air conditioning equipment may be set to operate at an earlier time.

[0370] Additionally, in the step of calculating the timing of the pre-operation of the air conditioning equipment (S904), information on the cooling load corresponding to the area where the heat island is expected to occur can be reflected to determine the control priority. More specifically, the air conditioning equipment can be set to operate at an earlier time for areas with a higher cooling demand calculated in the local cooling load calculation step (S903), thereby enabling the efficient distribution of limited equipment resources. For example, if multiple areas where the heat island is expected to occur exist simultaneously, pre-operation can be performed preferentially for areas with high cooling loads.

[0371] In addition, according to various embodiments, the pre-start time can be repeatedly recalculated and dynamically updated whenever new temperature data and power load data are input. More specifically, as time-series data is updated, the rate of temperature rise and the predicted time of heat island occurrence may change, and accordingly, the start time of operation of the air conditioning equipment can also be adjusted in real time. For example, if the server load decreases rapidly, the time of heat island occurrence may be delayed, so the start time of operation of the air conditioning equipment may also be delayed accordingly.

[0372] In addition, according to various embodiments, the result of the pre-operation timing calculation can be used as an input for coordinated control of the air conditioning equipment performed in a subsequent step, and each air conditioning equipment can be set to operate sequentially at different times. For example, if two air conditioning equipments are adjacent to each other in a specific area, one equipment can be set to operate first and the other equipment to operate after a certain period of time, thereby minimizing airflow collision.

[0373] In addition, the process of calculating the pre-operation timing can be configured as a data-based control structure, and previous control results can be reflected in subsequent calculations. More specifically, temperature change data resulting from the operation of a specific air conditioning system can be included in a time-series data tensor and reflected in subsequent predictions and timing calculations, thereby gradually improving control accuracy. For example, if the cooling effect of a specific system appears later than expected, the delay time of that system can be reflected as an increased value in subsequent calculations.

[0375] Finally, in the step of performing asymmetric cooperative air conditioning control (S905), the selected air conditioning equipment is driven in advance according to the start time of operation calculated in the step of inverse calculation of the air conditioning equipment start time (S904), and when multiple air conditioning equipment are operating simultaneously, the air volume or direction of each air conditioning equipment is controlled asymmetrically to suppress the generation of turbulence caused by collision between airflows.

[0376] More specifically, in the step of performing asymmetric cooperative air conditioning control (S905), air conditioning equipment that has reached a pre-operation time may be operated preemptively, and local cooling may be initiated for the area where a heat island is expected to occur. For example, if a heat island is predicted to occur in a specific area, an inlow cooler adjacent to that area may be controlled to operate in advance before the heat island occurs to lower the air temperature. In this way, unnecessary energy consumption can be suppressed by selectively operating only specific equipment rather than the entire air conditioning system.

[0377] Additionally, in the step of performing asymmetric cooperative air conditioning control (S905), asymmetric control may be performed to prevent collisions between airflows when multiple air conditioning facilities operate simultaneously. More specifically, the step of performing cooperative control may include: a step of reducing the airflow of some of the air conditioning facilities or changing the airflow direction; and a step of maintaining or increasing the airflow of other air conditioning facilities to disperse the airflow so that the air conditioning airflows do not collide at the same point. For example, if two air conditioning facilities installed in positions facing each other discharge a strong airflow in the same direction, the airflow may collide in the central area and turbulence may occur; therefore, the airflow may be controlled to disperse the airflow by reducing the airflow of one facility and changing the airflow direction of the other facility to a lateral direction.

[0378] Additionally, in the step of performing asymmetric cooperative air conditioning control (S905), the cooling load amount calculated in the step of calculating the local cooling load (S903) may be directly reflected as the air conditioning equipment control value. More specifically, the step of performing cooperative control may include: a step of setting the airflow or cooling water control amount as a value corresponding to the cooling load amount for each of the plurality of air conditioning equipment; and a step of adjusting the control value so that the direction of airflow movement between the air conditioning equipment is formed in different directions. For example, if a cooling load of 8 kW is required for a specific area, a fan speed and a cooling water valve opening rate proportional to the load amount may be set, and a method may be applied in which the airflow direction of some equipment is adjusted to the left and the airflow direction of other equipment is adjusted to the right, taking into account airflow interference with adjacent equipment.

[0379] In addition, in the step of performing asymmetric cooperative air conditioning control (S905), a control strategy can be determined by reflecting together the temperature distribution and power load information included in the three-dimensional thermal volume data and the time-series multidimensional data tensor. More specifically, if the power load in a specific zone continuously increases, a higher cooling capacity may be allocated to that zone, and if there is a large temperature difference with adjacent zones, the direction of airflow may be adjusted to promote heat diffusion. For example, if a high-temperature area is formed at the rear of a specific rack and heat diffuses into an adjacent passage, the airflow direction of the air conditioning equipment may be adjusted toward that passage to efficiently remove heat.

[0380] In addition, in the step of performing asymmetric cooperative air conditioning control (S905), space-based control can be performed to minimize mutual interference between air conditioning equipment. More specifically, the range of influence of each air conditioning equipment can be analyzed based on three-dimensional spatial coordinates, and control values ​​can be adjusted so that excessive airflow is not concentrated in the same area. For example, if multiple air conditioning equipment simultaneously provides a strong airflow to a specific area, balanced control can be performed in such a way that the output of some equipment is limited and the output of other equipment is maintained.

[0381] Additionally, in the step of performing asymmetric cooperative air conditioning control (S905), a closed-loop control structure may be formed in which the control results of the air conditioning equipment are reflected back into the time-series data. More specifically, actual temperature change data after air conditioning equipment control may be collected again and utilized as input data for an artificial intelligence model, thereby allowing for correction of the control results. For example, if the control effect of a specific piece of equipment is lower than expected, the control strategy may be modified in such a way that the output of that equipment is increased or another piece of equipment is additionally selected during subsequent control.

[0382] In addition, according to various embodiments, the control method of the air conditioning system may be configured as a multi-stage control structure that is adjusted stepwise according to the passage of time and changes in thermal state. More specifically, in the initial control section, the airflow and cooling water control amounts may be set within a limited range to prevent excessive cooling in the area where a heat island is expected to occur, and subsequently, a method may be applied in which the control intensity gradually increases according to changes in time-series temperature data and power load data.

[0383] For example, if the predicted temperature in a specific zone differs from the target holding temperature by 2°C or less, the fan speed may be set to approximately 60% of the rated level; if the temperature difference in the same zone expands to 5°C or more, the fan speed may be increased to 80% or more, and a method may be applied in which the cooling water valve opening rate is gradually expanded from 20% to 50%. In addition, if the rate of temperature rise continuously increases over a certain period of time, complex control may be performed in which not only the airflow volume but also the cooling water circulation flow rate is additionally increased.

[0384] In addition, time-based control parameters may be applied during the stepwise control process, and the transition point between each control stage may be dynamically adjusted according to time-series data. More specifically, if the rate of temperature change in a specific zone increases above a certain threshold, the transition to the next control stage may be performed early, and if the temperature change stabilizes, the current control state may be maintained or lowered. For example, if the rate of temperature change is maintained at 0.3°C or less per second, the current control stage may be maintained, and if it increases to 0.6°C or more per second, it may be immediately switched to the next control stage to apply a higher cooling capacity.

[0385] In addition, the results of HVAC control can be provided in a visualized form combined with three-dimensional thermal volume data, allowing the thermal and control states within the space to be checked simultaneously. More specifically, for each three-dimensional cube, temperature values ​​can be represented by color, while the operating status of the HVAC equipment affecting that area can be superimposed and displayed. For example, a cube of a specific area may be displayed in shades of red according to temperature, the airflow level of the HVAC equipment cooling that area may be represented by the length of an arrow, and the airflow direction may be visualized as a vector within the space.

[0386] In addition, the operating status of the HVAC equipment can be visualized on an individual equipment basis, and the fan speed, cooling water valve opening rate, and operation status of each piece of equipment can be displayed together. More specifically, an icon indicating the equipment status can be placed at the location of each HVAC piece of equipment; when the fan speed is high, it is displayed in a highlighted color, and when the cooling water flow rate is increased, a separate display element can be added. For example, equipment operating at a fan speed of 90% or higher can be displayed in red, while equipment operating at 50% or lower can be displayed in a green hue.

[0387] In addition, airflow distribution can be represented in the form of a vector field within the space, which allows for the intuitive identification of airflow collisions between HVAC equipment and the airflow paths. More specifically, the direction of airflow can be indicated by arrows for each 3D cube, and if arrows pointing in opposite directions converge in the same area, it can be determined that there is a high probability of airflow collision in that zone. For example, if the airflow from the equipment on the left and the airflow from the equipment on the right collide in the center of a specific passageway section, that area can be highlighted to indicate that control adjustments are required.

[0388] In addition, visualization data can be provided in a format that allows playback along a time axis, enabling the continuous observation of temperature and airflow changes resulting from HVAC control. More specifically, a function to play back data for a specific time interval can be provided, and the temperature distribution and operating status of the HVAC equipment at each point in time can be displayed sequentially. For example, the process of the temperature gradually decreasing after pre-operation is performed in a specific heat island zone can be observed in chronological order.

[0389] In addition, visualization results can be provided in conjunction with external terminals or management systems, allowing operators to check the HVAC status even in a remote environment. More specifically, 3D spatial data can be transmitted over a network and rendered as the same spatial structure on a user terminal, and functions for zooming, rotating, or cross-sectional analysis of specific areas can be provided. For example, when a specific rack area is selected, the temperature distribution, power load status, and HVAC equipment operation status of that area can be displayed together.

[0391] FIG. 10 is a block diagram illustrating the overall configuration of a device for performing an air conditioning control method using a three-dimensional thermal volume according to an embodiment of the present invention.

[0392] Referring to FIG. 10, an apparatus (600) for performing an air conditioning control method using a three-dimensional thermal volume according to an embodiment of the present invention may include a three-dimensional data collection unit (610), an AI prediction unit (620), a cooling load calculation unit (630), and a targeting control unit (640).

[0393] The 3D data collection unit (610) can perform the role of constructing a time-series-based multidimensional data tensor by combining multiple 3D cube unit thermal volume data generated through the 3D thermal volume generation step (S306) and IT power load data collected from server equipment within the facility space.

[0394] More specifically, the 3D data collection unit (610) can receive 3D volumetric thermal data input from the visualization unit (560) or an external storage device, and can collect a data structure including coordinate information (x, y, z) corresponding to each cube, temperature value, measurement time information, and status information indicating whether measurement or interpolation has taken place. At this time, each cube data can be aligned according to an absolute coordinate system in space, and temperature values ​​for the same coordinates can be sequentially accumulated and stored. For example, it can be configured so that up to 100 temperature values ​​collected at 0.1-second intervals for a specific cube are stored, thereby allowing the temperature change pattern for a period of about 10 seconds to be maintained in a time series form.

[0395] Additionally, the 3D data collection unit (610) can receive IT power load data from server equipment or a power monitoring system. More specifically, for each server rack or individual server equipment, information such as power consumption, CPU usage, GPU usage, and power supply status may be included, and this data may be mapped to the 3D spatial coordinates where the equipment is located. For example, if the central location of a specific rack corresponds to the coordinates (x=12, y=5, z=2), the average power load value of that rack may be linked to the cube data of that location. Additionally, if multiple servers exist in a single rack, the power load of each server may be aggregated in the form of an average value or a weighted sum and distributed to the corresponding cube or multiple cubes.

[0396] Additionally, the 3D data collection unit (610) can synchronize and combine thermal volume data and IT power load data based on the time axis. More specifically, if the collection cycle of thermal volume data and the collection cycle of power load data are different, the data can be aligned based on the same time stamp through time interpolation or resampling processing. For example, if thermal volume data is collected at 10 frames per second and power data is collected once per second, the power data can be repeatedly applied to the 10 frames or an intermediate value can be generated using a linear interpolation method.

[0397] Additionally, the 3D data collection unit (610) can structure each data to form a multidimensional data tensor. More specifically, the data tensor can be composed of a 5-dimensional structure of (time, x, y, z, channel), and each channel may include a temperature value, a power load value, a rate of change of temperature, an interpolation flag, etc. For example, one tensor element may include a temperature value of 28.5℃, a power load of 3.2kW, a rate of change of temperature of 0.3℃ / s, and a value of 0 or 1 for interpolation for coordinates (x, y, z) at a specific time t.

[0398] In addition, in certain embodiments, data normalization and preprocessing processes may be performed together. More specifically, temperature values ​​may be normalized to a range of 0 to 1 based on a range of minimum and maximum values, and power load data may also be converted into a ratio value relative to the maximum capacity of the equipment. For example, if 40°C is set as the upper limit, 28°C may be converted into a normalized value of approximately 0.7, and when consuming 2.5kW at a maximum equipment of 5kW, it may be converted into a value of 0.5.

[0399] Additionally, the 3D data collection unit (610) can perform correction processing for missing data or outliers. More specifically, if data reception is missing in a specific time interval, data from a previous time point or data from an adjacent time point may be interpolated and filled in, and if an abnormally rapid temperature change is detected, the data may be filtered or smoothed. For example, if a rapid change of 5°C or more occurs at a 0.1-second interval, the value may be determined to be an outlier and replaced with the average value of adjacent data.

[0400] Additionally, the 3D data collection unit (610) can perform spatial unit grouping processing. More specifically, when multiple cubes correspond to a single rack or a specific area, the average value or maximum value of the cubes can be calculated separately and generated as group unit data. For example, for 20 cubes corresponding to a single rack area, an average temperature of 30.2°C and a maximum temperature of 34.1°C can be calculated and stored as rack unit status information.

[0401] Additionally, the 3D data collection unit (610) may include an interface for transmitting the generated data tensor to the AI ​​prediction unit (620). More specifically, the data tensor may be stored in a memory buffer and then transmitted in batches at regular intervals, and in a real-time processing environment, it may be transmitted continuously in a streaming manner. For example, data from the last 50 frames may be composed of a single input block and provided sequentially to the AI ​​prediction unit (620).

[0403] The AI ​​prediction unit (620) receives a time-series multidimensional data tensor transmitted from the 3D data collection unit (610) and can perform the role of detecting topological heat islands that may occur in a specific 3D cube area within the facility space in advance and identifying a risk area expected to exceed a dynamic threshold.

[0404] The AI ​​prediction unit (620) can predict the possibility of a future temperature rise in a specific area by receiving inputs such as the 3D temperature distribution accumulated over a certain time interval, the time change of IT power load, positional relationships within the space, temperature deviation between adjacent areas, and the allowable operating conditions of server equipment, rather than simply determining the temperature distribution at the current time. For example, even if the temperature of a specific cube area has not yet reached a dangerous temperature at the current time, if a continuous temperature rise occurs in adjacent cubes and the power load of the server rack corresponding to the same area increases rapidly, the AI ​​prediction unit (620) can determine that there is a high probability that the area will transition into a heat island state within seconds to minutes.

[0405] More specifically, the AI ​​prediction unit (620) can receive the input data tensor as a multidimensional structure including a time axis, a spatial axis, and an attribute axis. At this time, the AI ​​prediction unit (620) can read together for each tensor slice the 3D temperature distribution at a specific point in time, the IT power load corresponding to each cube, the rate of temperature change, an interpolation flag, and facility location information. For example, 60 frames stored at 0.5-second intervals during the last 30-second period may be input, and each frame may include a cube array in the form of X×Y×Z and multiple channel data corresponding to each cube.

[0406] Additionally, a temperature value may be stored in one channel, a power consumption value in another channel, and a change in temperature relative to a previous point in time in yet another channel. By simultaneously processing the input data from these multiple channels, the AI ​​prediction unit (620) can consider both the spatial cumulative heat generation pattern and the temporal upward trend, which are difficult to capture with a single temperature value.

[0407] Additionally, the AI ​​prediction unit (620) can determine whether a specific cube area develops into a topologically connected high-temperature section. More specifically, the AI ​​prediction unit (620) can analyze not only the absolute temperature of individual cubes, but also the connectivity between adjacent cubes, the area or volume of a cube cluster belonging to the same temperature range, and the continuity of heat diffusion in a specific direction.

[0408] For example, if 31°C is measured temporarily in only one cube, it may be classified as a single local heat, but if a temperature of 30°C or higher is continuously formed in 12 or more adjacent cubes and such a cluster of cubes expands in the direction of the passageway or the upper direction of the rack, it may be determined as a topological heat island. In addition, the AI ​​prediction unit (620) can subdivide the heat diffusion pattern by distinguishing whether the shape of the heat island cluster is formed as a long band, expands as a mass surrounding a specific center point, or accumulates as a column in the upward direction. In a specific embodiment, the topological characteristics of the heat island can be quantified by simultaneously using the connectivity of the cube cluster, the temperature difference at the boundary, and the temperature gradient between the center and the outer part.

[0409] Additionally, the AI ​​prediction unit (620) can calculate dynamic thresholds that reflect the hardware limitations of the server equipment. More specifically, the AI ​​prediction unit (620) can receive information on the acceptable temperature range, cooling structure, power consumption limit, and recommended operating temperature for each server rack or group of equipment, and can use this to set different risk judgment values ​​depending on the location, even within the same space.

[0410] For example, a rack area where general storage equipment is placed may be determined to be a dangerous area if it exceeds 35°C, and a rack area where high-heat GPU servers are densely concentrated may be classified as a dangerous state at a level of 32°C or 33°C, as the allowable temperature according to the equipment specifications is lower. Additionally, the AI ​​prediction unit (620) may set a lower dynamic threshold when the rate of increase in power load of a specific piece of equipment is large. For example, for an equipment area where power consumption has increased from 3.0 kW to 4.2 kW in the last 10 seconds, the allowable temperature may be adjusted to a value lowered by 1°C to 2°C. Conversely, for an area where the load remains stable and there is sufficient cooling margin, the threshold may be maintained slightly higher.

[0411] In addition, the AI ​​prediction unit (620) can predict the possibility of future temperature rise by reflecting not only the absolute value of the IT power load but also the rate of change over time. More specifically, the AI ​​prediction unit (620) can read power consumption data from a recent period of time in chronological order and extract increasing trends, decreasing trends, surge patterns, periodic oscillation patterns, etc. For example, if the power consumption of a specific server area increases by 0.2 kW per second for the past 5 seconds, the AI ​​prediction unit (620) can determine that additional heat generation will soon occur in that area.

[0412] Additionally, if a specific piece of equipment periodically repeats the load, for example, fluctuating between 2.5 kW and 4.0 kW in 20-second cycles, the AI ​​prediction unit (620) can predict the likelihood of a recurrence of a temperature rise at the next high load time. In a specific embodiment, the rate of change in power, the acceleration of the change in power, the recent maximum load holding time, and the total load value per rack unit may be configured together as input channels, and the AI ​​prediction unit (620) can combine these values ​​to determine the likelihood of reaching a dangerous temperature in the near future rather than just the current state.

[0413] Additionally, the AI ​​prediction unit (620) can process spatial and temporal characteristics simultaneously using a deep learning model. More specifically, the AI ​​prediction unit (620) may include a spatial-temporal combined model that receives 3D temperature data and power load data as input, and this model may be configured in a form that combines a 3D convolution operation structure, a time series prediction structure, or an attention-based structure.

[0414] For example, in the initial processing unit, spatial characteristics can be extracted using cube blocks of size 3×3×3 or 5×5×5 as units, and in the subsequent processing unit, the amount of change over the last 10 to 60 points in time can be analyzed as a time series to reflect the cumulative heat generation pattern over time. In addition, in a specific embodiment, a Transformer structure that tokenizes spatial zones and learns the relationship between each token may be applied, and in this case, the AI ​​prediction unit (620) can also analyze the thermal correlation between rack zones that are far apart from each other. For example, if a pattern in which the accumulation of high-temperature air generated in the upstream passage moves to the downstream rack zone is repeatedly observed, the relationship can be learned as a long-range dependency.

[0415] In addition, the AI ​​prediction unit (620) can not only determine the risk zone in a binary manner but also subdivide the risk level into multiple stages and output it. More specifically, the AI ​​prediction unit (620) can calculate a risk score by considering the predicted temperature value, the difference from the dynamic threshold, the size of the heat island cluster, the rate of increase in power load, and the predicted duration together. For example, a zone approaching the dynamic threshold by 0.5 ℃ or less can be classified as the "concern stage," a zone expected to exceed 0.5 ℃ to 2.0 ℃ can be classified as the "warning stage," and a zone expected to exceed 2.0 ℃ while the heat island cluster expands simultaneously can be classified as the "risk stage." In addition, the AI ​​prediction unit (620) can also calculate the expected time of risk occurrence for each risk zone. For example, it can generate prediction results such as reaching the threshold after 15 seconds for a specific zone and reaching the threshold after 90 seconds for another zone.

[0416] Additionally, the AI ​​prediction unit (620) can identify multiple spatially contiguous cubes by grouping them into a single risk cluster. More specifically, among adjacent cubes, cubes whose predicted temperature values ​​are expected to be above a dynamic threshold can be clustered to form a single high-temperature prediction zone. For example, if 18 cubes are predicted to be in a state exceeding the threshold consecutively, the AI ​​prediction unit (620) can define these cubes as a first risk cluster. Additionally, if 7 cubes form a separate cluster in another zone, this can be defined as a second risk cluster. In a specific embodiment, center coordinates, volume, maximum predicted temperature, average predicted temperature, rise rate, and adjacent facility information can be stored together for each cluster. This cluster information can subsequently be used by the cooling load calculation unit (630) and the targeting control unit (640) to select a control target.

[0417] Additionally, the AI ​​prediction unit (620) may include an internal data structure for systematically storing model input and output data. More specifically, the AI ​​prediction unit (620) may include an input tensor buffer, a model inference result buffer, a risk zone table, and a dynamic threshold table. For example, the risk zone table may store fields such as [cluster ID, center coordinates, number of cubes, expected maximum temperature, dynamic threshold, expected exceedance, expected time of risk occurrence]. Additionally, the dynamic threshold table may store fields such as [equipment ID, current power load, recent load fluctuation rate, allowable temperature, adjustment threshold]. In a specific embodiment, the model inference results may be configured to be stored in a frame-by-frame manner so as to compare whether repeated risks occur in the same zone. For example, the last 100 inference results may be stored to statistically verify whether a specific zone is repeatedly determined to be in a risk state.

[0418] Additionally, the AI ​​prediction unit (620) may perform additional correction operations to reduce false positives. More specifically, if a threshold exceedance is expected in only a single frame, it may not be immediately confirmed as a risk zone, but may be confirmed only when the same trend is confirmed in multiple consecutive frames. For example, it may be marked as a final risk zone only when a threshold exceedance is expected in the same zone for three consecutive frames. Furthermore, even if the expected temperature of a specific zone is high, the risk determination may be mitigated if the actual load in that zone decreases rapidly. Conversely, if the current temperature is below the threshold but the power load rise rate and adjacent heat gradient increase simultaneously, it may be designated as a preemptive risk zone early. In certain embodiments, the final judgment may be corrected by reflecting the variance of the predicted value, consistency within the last 5 frames, and whether there is a simultaneous rise with adjacent zones.

[0419] Additionally, the AI ​​prediction unit (620) can generate different types of prediction outputs according to various embodiments. More specifically, it can output a map of predicted temperature values ​​in a single cube unit, generate event outputs in the form of risk cluster units, and output results converted into risk scores in the form of equipment or rack units. For example, for a specific rack A, summary information such as "maximum predicted temperature 34.8 ℃, dynamic threshold 33.0 ℃, expected time after exceeding 20 seconds" can be generated. In addition, for a specific passageway area, spatial prediction information such as "heat island expansion direction: top and left, predicted diffusion volume: 1.2 m³" can be generated together.

[0421] The cooling load calculation unit (630) receives information on the risk zone and the predicted maximum temperature transmitted from the AI ​​prediction unit (620) and can perform the role of calculating the required local cooling load for each risk zone. More specifically, the cooling load calculation unit (630) can calculate the difference (△T) between the predicted maximum temperature for a specific three-dimensional cube zone or cube cluster and the target temperature that must be maintained for that zone, and can calculate the local cooling load in kcal / h units by reflecting the calculated △T together with the spatial volume, air density, air specific heat of the air-conditioned zone, air circulation conditions, and time conditions.

[0422] For example, if the expected maximum temperature of a specific risk zone is 34.5 ℃ and the target maintenance temperature of the zone is 28.0 ℃, the cooling load calculation unit (630) can calculate △T as 6.5 ℃, and it can be determined that the larger this value, the greater the amount of cooling required for the zone. In addition, even with the same △T, a larger amount of cooling load can be calculated if the volume of the target zone is large or if high-temperature air is continuously flowed into the zone.

[0423] More specifically, the cooling load calculation unit (630) can first identify the spatial range of the risk zone. At this time, the risk cluster information transmitted from the AI ​​prediction unit (620) may include center coordinates, number of cubes, spatial volume, expected maximum temperature, amount exceeding dynamic threshold, etc., and the cooling load calculation unit (630) can set the air conditioning target zone using the spatial volume and location range among this information.

[0424] For example, if a specific cluster is formed over a range of 1.2 m in the x-axis direction, 0.8 m in the y-axis direction, and 2.0 m in the z-axis direction, the cooling load calculation unit (630) can calculate a space of approximately 1.92 m³ as the volume to be cooled. Additionally, even for the same risk zone, the actual volume to be cooled may be set differently depending on whether it includes only an air layer or an air layer near the outer surface of the rack, and the cooling load calculation unit (630) can recalculate the effective volume by reflecting information on the structure occupancy ratio or the air layer occupancy ratio together. For example, even if the volume of a specific cluster is 2.0 m³, if 25% of it is a structure occupancy area, the effective volume used for calculating the actual cooling load may be adjusted to approximately 1.5 m³.

[0425] Additionally, the cooling load calculation unit (630) can calculate the difference (△T) between the expected maximum temperature and the target maintenance temperature for each risk zone for each zone. More specifically, the target maintenance temperature can be set differently depending on the zone characteristics or equipment type, and the cooling load calculation unit (630) can apply different target temperatures even within the same space.

[0426] For example, a general server rack area can be set to a target maintenance temperature of 28.0°C, and a high-heat GPU server rack area can be set to a target maintenance temperature of 26.5°C or 27.0°C. Additionally, the air layer surrounding the battery facility can be set to a target temperature of 24.0°C to 25.0°C. In this case, the cooling load calculation unit (630) can calculate △T as 6.8°C when the expected maximum temperature in a specific GPU rack area is 33.8°C and the target temperature is 27.0°C, and can calculate △T as 3.0°C when the expected maximum temperature in the general server area is 31.0°C and the target temperature is 28.0°C at the same time. As △T is calculated differently for each area in this way, the risk and cooling requirements can be distinguished by area.

[0427] In addition, the cooling load calculation unit (630) can convert the local cooling load amount into kcal / h units by considering air characteristics and time conditions together, rather than simply reflecting △T. More specifically, the cooling load calculation unit (630) can calculate the amount of air included in the effective volume, and can use an air density value in the range of about 1.15 kg / m³ to 1.25 kg / m³. In addition, an air specific heat value in the range of about 0.23 kcal / kg·℃ to 0.24 kcal / kg·℃ can be applied.

[0428] For example, if the effective volume is 1.5 m³, the air density is 1.2 kg / m³, and △T is 6.5 ℃, the basic cooling energy required to lower the area to a target temperature within a short period of time can be calculated. If a target time to reach cooling, such as 30 seconds, 60 seconds, or 120 seconds, is added to this, the cooling load calculation unit (630) can convert this into an hourly equivalent amount and calculate the required cooling amount in kcal / h units. For example, if rapid cooling within a short period of time is required, a load of 4,000 kcal / h to 8,000 kcal / h can be calculated, and if relatively gradual cooling is allowed, a load of 1,500 kcal / h to 3,000 kcal / h can be calculated.

[0429] Additionally, the cooling load calculation unit (630) can adjust the cooling load amount by reflecting the thermal inertia characteristics within the space. More specifically, even if the same △T and the same volume are present, the actual amount of cooling required may vary depending on the thermal storage characteristics of surrounding structures, rack surfaces, cable bundles, floor panels, etc. To reflect these differences, the cooling load calculation unit (630) may apply a thermal inertia coefficient for each zone, for example, a correction coefficient of 1.0 for the central air layer zone, 1.15 to 1.30 for the rack-dense zone, and 1.20 to 1.35 for the zone adjacent to the battery facility.

[0430] For example, even if the basic calculation result is calculated as 3,200 kcal / h, the final cooling load can be adjusted to approximately 4,000 kcal / h if a thermal inertia coefficient of 1.25 is applied to the rack-dense zone. Additionally, if there is a history of slow temperature drop response in the zone over the past 5 minutes, an additional correction factor may be applied, and consequently, a higher cooling amount may be repeatedly calculated for the same zone.

[0431] Additionally, the cooling load calculation unit (630) can adjust the cooling requirement for each zone by reflecting the direction of air intake and the arrangement of the air conditioning equipment. More specifically, even for zones with the same △T, if they are located close to the air conditioning equipment outlet, the target temperature can be reached with a small amount of cooling, and if they are far from the air conditioning equipment or there is an airflow blind spot, a larger amount of cooling may be required.

[0432] For example, if the danger zone is located within 1.0 m of the air conditioning equipment, the distance correction factor can be set to 1.0, and if it is located between 3.0 m and 5.0 m away, a correction factor of 1.15 to 1.25 can be applied. Additionally, if the air flow analysis results determine that the cold air flow entering the zone is weak, the cooling load calculation unit (630) can calculate an additional airflow requirement for the same zone or increase the cooling amount. For example, if it is determined to be an airflow stagnation zone, an additional cooling amount of 10% to 25% of the basic calculated value can be reflected.

[0433] Additionally, the cooling load calculation unit (630) can calculate a local cooling load that changes dynamically over time. More specifically, for the same zone, the expected maximum temperature at the current time, the expected maximum temperature at the previous time, the recent power load increase rate, and the recent cooling response result can be reflected together, and accordingly, the cooling load can be calculated as a dynamic value that is updated at each time point rather than a fixed value.

[0434] For example, if the expected maximum temperature in a specific zone was 31.0 ℃ 10 seconds ago but has now risen to 33.5 ℃ and the power load has increased by 20% at the same time, the cooling load calculation unit (630) can calculate a higher load amount than at the previous point in time. Conversely, if prior cooling has already been performed and the expected maximum temperature has decreased from 34.0 ℃ to 30.5 ℃, the cooling requirement can be lowered immediately. In a specific embodiment, the load amount update cycle can be set to 0.5 seconds, 1 second, or 5 seconds, and the cooling load calculation unit (630) can recalculate a new load amount at each cycle.

[0435] Additionally, the cooling load calculation unit (630) can convert the load calculation results into a form that can be distributed not only by zone unit but also by facility unit. More specifically, when a total cooling amount of 6,000 kcal / h is required for a specific risk cluster, the cooling load calculation unit (630) can distribute the load by reflecting the location, capacity, and current operating rate of adjacent multiple air conditioning facilities.

[0436] For example, if the left air conditioning unit can handle a maximum of 3,500 kcal / h and the right air conditioning unit can handle a maximum of 4,000 kcal / h, a method of distributing 2,500 kcal / h to the left and 3,500 kcal / h to the right may be applied. Additionally, if a specific unit is already operating at 85% or more, the additional load added to that unit may be limited, and a greater weight may be allocated to other units with spare capacity. The results of this distribution may be transmitted to a targeting control unit (640) and converted into fan speed, valve opening rate, airflow direction control amount, etc.

[0437] Additionally, the cooling load calculation unit (630) can generate and maintain various internal data structures during the calculation process. More specifically, the cooling load calculation unit (630) may include a risk zone table, a target temperature table, a △T calculation table, a volume and air characteristics table, and a cooling load result table. For example, the △T calculation table may store fields [cluster ID, expected maximum temperature, target maintenance temperature, △T, calculation time], and the cooling load result table may store [cluster ID, effective volume, thermal inertia coefficient, distance correction coefficient, final kcal / h, list of facilities to be distributed]. Additionally, the history of the last 100 to 500 calculations may be maintained as a history table, through which a pattern of repeated increases in cooling demand in a specific zone can be analyzed.

[0438] Additionally, the cooling load calculation unit (630) can perform limiting calculations to prevent outliers or excessive control. More specifically, if the cooling load becomes excessively large due to a temporary surge in the prediction result, an upper limit or a rate of change limit may be applied. For example, if the cooling demand increases by more than 50% compared to the previous value in a single update cycle, the increase may be limited and adjusted to a range of 20% to 30%. In addition, for very small △T, for example, 0.3°C or less, it may be set to maintain only the minimum load value, and if the danger zone does not persist for a certain period of time, the cooling demand may be lowered to near 0. Through these limiting calculations, frequent increase and decrease operations of unnecessary air conditioning equipment may be reduced.

[0440] The targeting control unit (640) receives the local cooling load amount transmitted from the cooling load calculation unit (630) and the risk zone information transmitted from the AI ​​prediction unit (620), selects the air conditioning equipment adjacent to the risk zone, and dynamically calculates the pre-operation time and control parameters for the selected air conditioning equipment to perform local air conditioning control.

[0441] The targeting control unit (640) can select the equipment that should actually be controlled among a number of air conditioning equipment by considering the location coordinates of the danger zone, the cluster range, the expected temperature rise time, the required cooling load, and the surrounding equipment placement information. For example, if a specific danger zone is formed in the range of x=10~12, y=4~6, and z=1~3, the targeting control unit (640) can identify 2 to 4 In-Row Coolers closest to the zone as priority candidates, and among these equipment, the final equipment to be controlled can be determined by considering the actual available capacity, current operating status, and wind direction conditions.

[0442] More specifically, the targeting control unit (640) can analyze the distance and direction relationship between the physical location of each air conditioning facility and the danger zone. For example, by calculating the distance between the center coordinates of the danger zone and the location of the discharge port of each air conditioning facility, facilities located within 1.5 m can be set as a primary candidate group, and facilities located within the range of 1.5 m to 3.0 m can be set as a secondary candidate group. Additionally, by determining whether the discharge direction of the air conditioning facility is facing the danger zone, facilities with suitable airflow directions can be selected preferentially even under the same distance conditions. For example, among two facilities located at the same distance of 2.0 m, if one has an airflow direction directly facing the danger zone and the other is set in the opposite direction, the targeting control unit (640) can set the former as the priority selection target. In a specific embodiment, the angle difference between the facility and the danger zone can be calculated, and facilities having an angle difference of 30° or less can be selected preferentially.

[0443] Additionally, the targeting control unit (640) can dynamically calculate a pre-operation time (T-minus) by reflecting the mechanical operation delay time of the equipment and the thermal inertia coefficient of the space. More specifically, each air conditioning equipment may take a certain amount of time from power application until actual cold air discharge begins, and this delay time may be set to a range of 2 to 10 seconds depending on the type of equipment.

[0444] For example, fan-based cooling equipment may have a response delay of about 2 to 3 seconds, whereas refrigerant compressor-based equipment may have a delay of 5 to 10 seconds. Additionally, additional time may be required for the temperature to drop depending on the thermal inertia of the space, and in certain rack-dense areas, an additional delay of 10 to 30 seconds may occur due to the thermal inertia coefficient. The targeting control unit (640) can combine these two factors to calculate a pre-operation time so that the equipment is operated at a certain time earlier than the predicted risk occurrence time. For example, if a critical temperature exceeding is expected in a specific area after 20 seconds, the equipment delay time is 5 seconds, and the thermal inertia response time is 10 seconds, the targeting control unit (640) can control the equipment to operate at a time approximately 15 seconds earlier. The T-minus value calculated in this way can be set differently for each piece of equipment and can be individually adjusted according to the response characteristics of each piece of equipment.

[0445] Additionally, the targeting control unit (640) can convert the required cooling amount transmitted from the cooling load calculation unit (630) into equipment-specific control parameters. More specifically, if the required cooling load for a specific piece of equipment is calculated to be 3,000 kcal / h, the targeting control unit (640) can convert the value into control values ​​such as fan speed, refrigerant flow rate, and valve opening rate.

[0446] For example, if the maximum cooling capacity of the equipment is 5,000 kcal / h, a required load of 3,000 kcal / h can be met with approximately 60% output, and accordingly, the fan speed can be set to 60% of the maximum speed. In addition, when multiple equipment is cooperatively controlled for the same zone, the total required cooling amount can be distributed among the equipment; for example, a requirement of 6,000 kcal / h can be distributed to two equipment as 3,500 kcal / h and 2,500 kcal / h, respectively. In this case, the current operating status, remaining capacity, and maintenance status of each equipment can be taken into account.

[0447] Additionally, the targeting control unit (640) can perform cooperative control logic to prevent airflow collision and turbulence formation that may occur when multiple air conditioning facilities are operated simultaneously. More specifically, when two or more air conditioning facilities installed in positions facing each other discharge a strong airflow in the same direction, the airflow may collide in the central area, forming turbulence and reducing cooling efficiency. To prevent this phenomenon, the targeting control unit (640) can asymmetrically adjust the wind direction and wind speed for each facility.

[0448] For example, the left unit can be configured to deflect the wind direction to the left by approximately 30° to 45°, and the right unit can be configured to reduce the wind speed to 50% to 70% to avoid a straight-line collision. Additionally, in certain embodiments, one unit can be configured to maintain maximum output, while the adjacent unit can be configured to operate at a low output state as an auxiliary role. This allows the airflow to be induced in a circulating manner so that the cold air passes uniformly through the entire rack.

[0449] Additionally, the targeting control unit (640) can dynamically update the control strategy over time. More specifically, during the initial pre-operation phase, the equipment can be operated at a high output to achieve a rapid temperature drop, and when approaching the target temperature, the output value can be gradually lowered to prevent excessive cooling. For example, the equipment can be operated at 80% output during the initial 10-second interval, then lowered to 60% during the subsequent 20-second interval, and adjusted to 40% just before reaching the target temperature. Additionally, if the temperature drop proceeds slower than expected in a specific zone, the output value can be increased again. These control values ​​can be updated at intervals of 0.5 to 2 seconds and can be configured to respond to real-time environmental changes.

[0450] Additionally, the targeting control unit (640) can perform control by setting priorities when multiple risk zones exist simultaneously. More specifically, for each risk zone, a risk score is calculated based on △T, expected excess temperature, time of occurrence, and cluster size, and the zone with the higher score can be controlled first. For example, if zone A has a △T of 7.0 ℃, risk occurs after 15 seconds, and cluster size of 20 cubes, and zone B has a △T of 3.0 ℃, risk occurs after 60 seconds, and cluster size of 8 cubes, the targeting control unit (640) can set zone A as the priority control target. Additionally, if a specific facility affects multiple zones, the output value of the facility can be determined by weighting the required loads of each zone.

[0451] In addition, the targeting control unit (640) can store control history and status information to reflect in subsequent control. More specifically, for each facility, the start time of operation, output value, wind direction setting, control duration, etc., can be recorded, and a more appropriate control strategy can be applied by referring to previous control results regarding heat problems that occur repeatedly in the same area. For example, if sufficient cooling was not achieved with 60% output during previous control in a specific area, the initial output value can be set to 70% or more during the next control. In addition, if an overload condition occurs repeatedly in a specific facility, the usage ratio of that facility can be lowered and the load distributed to adjacent facilities.

[0452] Additionally, the targeting control unit (640) may include an interface for transmitting control commands to actual equipment. More specifically, fan speed control signals, valve opening control signals, wind direction control signals, etc., may be transmitted through communication with each air conditioning unit, and current status information may be received from the equipment in the form of feedback. For example, whether the equipment is actually operating at a set output value, current airflow, temperature sensor values, etc., may be received periodically, and this information may be reflected back into the control loop.

[0454] FIG. 11 is a flowchart illustrating the process of combining equipment data with three-dimensional thermal volume data according to an embodiment of the present invention to form time-series data.

[0455] Referring to FIG. 11, the process of combining equipment data with three-dimensional thermal volume data according to one embodiment of the present invention to form time-series data can be performed by a three-dimensional thermal volume data collection step (S1101), a coordinate-based synchronization and mapping step (S1102), and a time-series multidimensional data tensor construction step (S1103).

[0456] First, in the 3D thermal volume data collection step (S1101), real-time thermal volume data in the form of 3D cubes generated through a physical-data fusion interpolation-based 3D thermal volume generation method and IT power load data collected in real-time from server equipment can be collected, respectively.

[0457] More specifically, temperature data for each 3D cube may be configured to include temperature values ​​corresponding to each coordinate within space and may have a time-series structure that is continuously updated over time. Additionally, IT power load data collected from server equipment may include power consumption, computing resource utilization, or power fluctuation information measured at each server or rack unit, and may be stored along with the physical installation location of each piece of equipment. For example, if a power load of 3 kW is measured at a specific server rack, the data may be recorded along with spatial coordinate information where the rack is located.

[0458] Next, in the coordinate-based synchronization and mapping step (S1102), thermal volume data in the form of a 3D cube and IT power load data collected from server equipment can be aligned and a correspondence relationship formed within the same 3D coordinate system. More specifically, the physical location coordinates of each server equipment can be converted into 3D cube coordinates, and power load data can be mapped to the cube corresponding to the coordinates.

[0459] In this case, if a single server device is placed across multiple cube areas, the power load value of the device can be allocated in a manner that distributes it to each cube, and the distribution ratio can be set according to the actual physical occupied area or volume of the device. For example, if a specific rack occupies four cube areas, the total power load can be distributed evenly or weighted to areas with concentrated heat generation. Additionally, a time synchronization process can be performed to align temperature data and power data based on the same time period, and if they have different collection cycles, they can be reconstructed into a structure with the same time step through interpolation or resampling.

[0460] Additionally, in the coordinate-based synchronization and mapping step (S1102), corrections for position errors and time errors that may occur during the data matching process may be performed. More specifically, a coordinate correction algorithm may be applied to correct minute discrepancies between the server equipment location and cube coordinates, and a time correction coefficient may be applied to data with a time delay. For example, if power data is collected about 1 second later than temperature data, the time axis may be corrected by taking the delay into account.

[0461] Next, in the time series multidimensional data tensor construction step (S1103), the 3D solid cube unit data and power load data, which have been synchronized and mapped, can be combined to create a time series multidimensional data tensor. More specifically, the data tensor can be constructed in a structure including a time axis (T), spatial coordinate axes (X, Y, Z), and a data channel axis (C), and each channel may include various physical characteristics such as temperature values, power load values, temperature change rates, and power change rates. For example, data corresponding to a specific coordinate at a specific time point can be constructed in the form of a vector including [temperature, power, temperature change rate, power change rate], and such data can be accumulated along the time axis to form a multidimensional tensor.

[0462] In addition, in the time-series multidimensional data tensor configuration step (S1103), the data structure can be configured to be optimized for AI-based heat island prediction and HVAC control operations performed in subsequent steps. More specifically, the tensor data can be stored in a continuous array structure considering memory access efficiency, and an index structure can be formed together to enable rapid access to data for specific coordinates or specific time intervals. For example, a function can be implemented to quickly extract and analyze only the last 10 seconds of data for a specific rack area.

[0463] In addition, according to various embodiments, additional environmental data may be included in the time-series multidimensional data tensor, thereby enabling more precise analysis. More specifically, air flow velocity, humidity, pressure, or equipment operating status information may be combined into the same cube structure, and this data may be composed of a single integrated tensor. For example, for a specific cube, air flow velocity and humidity values ​​may be stored together with temperature, allowing for analysis that reflects the air circulation status.

[0464] In addition, time-series multidimensional data tensors can be continuously updated and expanded by adding to the existing tensor whenever new data is input. More specifically, a sliding window structure can be applied where the latest time step data is added to the tensor and data from a certain time prior is removed or moved to a separate storage. For example, the tensor can be managed by maintaining only the data from the last 60 seconds.

[0466] FIG. 12 is a flowchart illustrating the process of calculating a dynamic threshold based on three-dimensional thermal volume data according to an embodiment of the present invention.

[0467] Referring to FIG. 12, the process of calculating a dynamic threshold based on three-dimensional thermal volume data according to one embodiment of the present invention can be performed by a data tensor input step (S1201), a hardware operating temperature analysis step (S1202), a power load fluctuation rate analysis step (S1202-1), a dynamic threshold setting step (S1203), a topological heat island temperature prediction step (S1204), and a danger zone identification step (S1205).

[0468] First, in the data tensor input step (S1201), a time-series multidimensional data tensor may be input, which combines three-dimensional thermal volume data generated by dividing a three-dimensional space into multiple three-dimensional cube units and IT power load data collected from server equipment. More specifically, multi-channel data including temperature values, temperature change rates, power load values, and power change rates over time may be input for each three-dimensional cube, and such data may be composed of time-series data over a specific time interval, for example, the interval of the last 30 to 120 seconds. Additionally, the data tensor may be structured to include spatial coordinates and time axes simultaneously, and may be provided in a form that allows for the simultaneous analysis of temperature and power change patterns at specific coordinates.

[0469] Next, in the hardware operating temperature analysis step (S1202), a reference threshold temperature may be set based on the maximum allowable temperature and current operating temperature of the server equipment corresponding to each three-dimensional cube. More specifically, an allowable temperature range defined by the server equipment manufacturer, for example, an allowable operating condition in the range of 40°C to 80°C, may be reflected, and a safety margin temperature may be calculated based on the difference from the currently measured temperature. In addition, for specific server equipment, the allowable temperature may be set differently depending on the high-performance computing unit or high-density layout structure, and an individual reference threshold temperature reflecting these equipment-specific characteristics may be calculated. For example, a method may be applied where the allowable temperature is set to 75°C for a high-performance GPU server and 65°C for a general server.

[0470] In addition, in the power load fluctuation rate analysis step (S1202-1), the temporal variation characteristics of IT power load data collected from server equipment can be analyzed. More specifically, the power load increase or decrease rate during a specific time interval can be calculated, and this rate of change can be used as a key indicator to determine the possibility of future temperature rise. For example, if the power load increases by more than 5% per second, it can be determined that there is a high probability of a temperature rise occurring in that area within a short period of time; conversely, if the power load is maintained stably, it can be determined that the temperature change will also be limited. Furthermore, the power load data can be analyzed with noise removed by smoothing using a moving average or exponentially weighted moving average method.

[0471] Next, in the dynamic threshold setting step (S1203), a dynamic threshold for each 3D cube can be set by combining the hardware operating temperature analysis result and the power load fluctuation rate analysis result. More specifically, a final threshold can be calculated by reflecting a correction value based on the power load growth rate to the reference threshold temperature, and a method can be applied in which the threshold is lowered when the power load growth rate is high and the threshold is set relatively high when the power load is stable.

[0472] For example, if the reference threshold temperature is 70°C, the threshold can be lowered to 65°C in situations where the power load growth rate is high, and raised to 72°C when the power load change is minimal. In this way, by setting a threshold that changes variably according to environmental conditions rather than a fixed threshold, it becomes possible to make more proactive risk assessments.

[0473] Next, in the topological heat island temperature prediction step (S1204), future temperature changes for each 3D cube zone can be predicted using an artificial intelligence model that takes a time-series multidimensional data tensor as input. More specifically, the connectivity of temperature distributions between adjacent cubes and power load change patterns are considered together to analyze the possibility of a topological heat island phenomenon forming in a specific zone where heat is concentrated. Additionally, based on the time-series data, temperature values ​​for a certain period of time, for example, 10 to 60 seconds later, can be predicted, and these prediction results can be compared with a dynamic threshold to determine whether there is a risk. For example, if the predicted temperature of a specific zone is calculated to be 68 ℃ and the dynamic threshold for that zone is set to 65 ℃, that zone can be determined to be in a dangerous state.

[0474] Finally, in the danger zone identification step (S1205), a three-dimensional cube zone where the predicted temperature exceeds a dynamic threshold can be identified as a heat island-prone zone. More specifically, for each cube, the predicted temperature and the dynamic threshold can be compared to determine whether they are exceeded, and the cubes that are exceeded can be grouped into one continuous danger zone if they are spatially adjacent.

[0475] Additionally, in certain embodiments, the risk level may be graded based on the size, duration, and rate of temperature rise of the risk zone, thereby determining the priority of air conditioning control. For example, a high risk grade may be assigned if a continuous temperature rise occurs over a wide area, and a low risk grade may be assigned if the rise is localized and temporary.

[0476] In addition, the results of identifying hazardous zones can be stored based on 3D coordinates and used as input data for local cooling load calculations and HVAC equipment control performed in subsequent steps. More specifically, for each 3D cube, predicted temperature values, dynamic thresholds, risk status, and risk grades can be stored together, thereby allowing the thermal risk distribution over the entire space to be configured in a 3D form.

[0478] FIG. 13 is a flowchart illustrating the process of calculating a pre-operation time based on three-dimensional thermal volume data according to an embodiment of the present invention.

[0479] Referring to FIG. 13, the process of calculating a pre-operation time based on three-dimensional thermal volume data according to one embodiment of the present invention can be performed by a target temperature setting step (S1301), a local cooling load calculation step (S1302), a control parameter collection step (S1303), a thermal inertia coefficient calculation step (S1304-1), an equipment operation delay time calculation step (S1304-2), a pre-operation time inversion step (S1305), and a preemptive air conditioning control execution step (S1306).

[0480] First, in the target temperature setting step (S1301), a target temperature to be maintained for the heat island-prone zone may be set. More specifically, for each three-dimensional cube zone, the target temperature may be individually set based on the normal operating conditions of the server equipment, service level requirements, and facility operation policies. For example, for a general server zone, a range of approximately 24°C to 27°C may be set, and for a zone densely populated with high-performance computing equipment, a lower temperature range, for example, 20°C to 23°C, may be set. Additionally, the target temperature may be dynamically changed according to specific time zones or workload conditions, and the target temperature may be raised during low-load conditions at night, while stricter temperature standards may be applied during high-load conditions.

[0481] Next, in the local cooling load calculation step (S1302), the amount of local cooling load required for each three-dimensional cube zone can be calculated based on the difference between the predicted temperature and the target temperature. More specifically, the amount of heat to be removed per unit time can be calculated by reflecting the temperature difference value, the volume of the corresponding cube, air density, and specific heat coefficient. For example, if the predicted temperature in a specific cube is 35 ℃ and the target temperature is 25 ℃, a cooling load corresponding to a temperature difference of about 10 ℃ can be calculated, and the amount of the load can be expressed in kcal / h or W units. In addition, the time-series temperature change rate is also reflected, so that a higher cooling load can be calculated for zones with a faster rate of temperature rise.

[0482] Next, in the control parameter collection step (S1303), control-related parameters for calculating the pre-operation time may be collected. More specifically, thermal inertia characteristics within the space, response characteristics of the air conditioning equipment, air flow characteristics, and structural heat transfer characteristics may be included, and these parameters may be used as input variables required for calculating the pre-operation time. In addition, equipment information such as the installation location, cooling capacity, and controllable range of each air conditioning equipment may be collected together.

[0483] Next, in the thermal inertia coefficient calculation step (S1304-1), a thermal inertia coefficient representing the heat accumulation and release characteristics of the space can be calculated. More specifically, the thermal inertia coefficient can be calculated based on the temperature change rate and heat accumulation characteristics of a specific zone, and this can be used as an indicator of how quickly or slowly the temperature change occurs. For example, a zone inside a rack with a closed structure may have a high thermal inertia coefficient, while a passageway zone with smooth air circulation may have a relatively low thermal inertia coefficient. Additionally, the thermal inertia coefficient can be composed of a value that changes over time and can be continuously updated based on real-time temperature data.

[0484] In addition, in the step of calculating the equipment operation delay time (S1304-2), the mechanical operation delay time required for each air conditioning equipment to produce an actual cooling effect can be calculated. More specifically, the time required to start fan operation, the time required to start cooling water circulation, and the time required to form airflow may be included, and these elements may be combined to calculate the total delay time. For example, in the case of a specific inlow cooler, if it takes about 2 seconds to start the fan and about 5 seconds to stabilize the cooling water flow rate, the total delay time may be calculated as about 7 seconds. Additionally, the delay time may vary depending on the condition of the equipment, and if it is already in a partially operating state, the delay time may be shortened.

[0485] Next, in the pre-operation time inversion step (S1305), the pre-operation time (T-minus) can be inversionly calculated based on the thermal inertia coefficient and the equipment operation delay time. More specifically, by considering the time required to reach the target temperature in a specific zone and the equipment response delay time, the operation time can be calculated so that the air conditioning equipment starts operating at a time before the actual temperature rise occurs. For example, if it is predicted that the temperature in a specific zone will reach a critical value after about 15 seconds and the equipment response delay time is 5 seconds, the T-minus value can be set so that the equipment starts operating at a time about 10 seconds earlier. In addition, for zones with a high thermal inertia coefficient, pre-operation can be set to be performed at an earlier time.

[0486] Finally, in the preemptive air conditioning control execution step (S1306), the target air conditioning equipment may be pre-operated and local cooling control may be performed according to the calculated pre-operation time. More specifically, air conditioning equipment adjacent to the area where a heat island is expected to occur may be selectively activated, and the fan speed or cooling water control amount of each equipment may be set to a value corresponding to the local cooling load. In addition, when multiple air conditioning equipment operates simultaneously, the airflow direction or airflow volume may be asymmetrically adjusted to prevent collision between airflows, thereby inducing the airflow to concentrate in a specific area. For example, a method to avoid airflow collision may be applied by setting the airflow volume of the left equipment to 70% and adjusting the airflow volume of the right equipment to 50%.

[0488] FIG. 14 is a diagram illustrating a concept of analyzing air flow distribution based on three-dimensional thermal volume data according to an embodiment of the present invention.

[0489] Referring to FIG. 14, multiple server racks may be arranged at regular intervals within the space, and a passage area for air circulation may be formed between each server rack. In this structure, cold air discharged from the air conditioning equipment flows in a specific direction, and an upward airflow is formed by the heat generated from the server equipment, thereby forming a complex air flow pattern within the space.

[0490] In addition, three-dimensional thermal volume data can be configured to include temperature values ​​for each unit of three-dimensional cubes, and the direction and intensity of airflow can be estimated based on this temperature distribution. More specifically, a thermal gradient based on the temperature difference between adjacent three-dimensional cubes can be calculated, and the direction of heat transfer from a high-temperature area to a low-temperature area can be used as a key indicator of the airflow direction. For example, if a high-temperature area is formed at the top of a specific server rack, an upward airflow may occur from that area, and a flow that diffuses into an adjacent passage area may be formed.

[0491] In addition, according to various embodiments, air flow can be represented in the form of a vector field, and vector information indicating the direction and magnitude of the air flow can be generated for each three-dimensional cube. More specifically, for each cube, the rate of temperature change, the temperature difference with adjacent cubes, and the temperature change pattern over time can be comprehensively considered to calculate the air flow vector. For example, if a temperature decrease occurs from left to right in a specific area, it can be determined that air flow is formed in that direction, and the length of the vector can be set in proportion to the rate of temperature change or the air movement speed.

[0492] In addition, the process of how air flow distribution changes depending on the operating status of the air conditioning equipment can be analyzed. More specifically, when the fan speed of a specific air conditioning unit increases, a strong airflow may be formed near that unit, which may change the direction of air flow in adjacent areas. For example, when the airflow of a specific in-row cooler increases, a strong cold air flow is formed from that unit toward the passageway, and whether there is an airflow collision with the unit on the opposite side can be analyzed.

[0493] In addition, when multiple air conditioning systems operate simultaneously, turbulence may occur due to collisions between airflows, and such turbulent regions may be represented in the form highlighted in Fig. 14. More specifically, when airflows moving in opposite directions meet in the same area, a turbulent region in which the airflow changes irregularly may be formed, which may lead to a decrease in cooling efficiency. For example, if an airflow discharged from the left system toward the right and an airflow discharged from the right system toward the left collide in the center of the passage, the temperature distribution in that area may appear uneven.

[0494] Furthermore, in the present invention, an air conditioning equipment control strategy can be established based on the results of such air flow distribution analysis. More specifically, airflow collisions can be avoided in areas with a high probability of turbulence by reducing the airflow of some air conditioning equipment or changing the airflow direction, and control can be performed to induce airflow in a specific direction. For example, airflow collisions can be mitigated by maintaining the airflow of the left equipment while reducing the airflow of the right equipment.

[0495] Furthermore, air flow distribution can consist of dynamic data that changes over time, and airflow patterns that occur repeatedly in specific areas can be identified through time-series analysis. More specifically, if turbulence repeatedly occurs at the same location within a specific timeframe, structural problems or issues with the placement of HVAC equipment in that area can be analyzed as the cause, and based on this, equipment relocation or changes to control strategies can be implemented.

[0496] In addition, according to various embodiments, air velocity sensor, pressure sensor, or humidity sensor data may be utilized in addition to temperature data for air flow analysis, thereby enabling more precise flow analysis. For example, if the air velocity is low and the temperature is high in a specific area, it may be determined to be an air stagnation area, and intensive cooling control may be required for that area.

[0498] FIG. 15 is a flowchart illustrating the entire process of analyzing heat distribution based on three-dimensional thermal volume data according to one embodiment of the present invention.

[0499] Referring to FIG. 15, the entire process of analyzing heat distribution based on three-dimensional thermal volume data according to one embodiment of the present invention can be performed by a three-dimensional thermal volume data collection step (S1501), an AI-based heat island prediction temperature calculation step (S1502), a dynamic threshold exceeding determination step (S1503), a pre-operation time calculation step (S1504), and a local pre-control step (S1505).

[0500] First, in the 3D thermal volume data collection step (S1501), real-time thermal volume data for each 3D cube formed by a physical-data fusion interpolation-based 3D thermal volume generation method and real-time IT power data of server equipment corresponding to each spatial coordinate can be collected. More specifically, the facility space can be maintained in a state divided into a plurality of 3D cubes, and temperature values ​​for each time point can be stored in each 3D cube.

[0501] In this case, since the temperature values ​​may include both values ​​obtained from the direct shooting area and values ​​interpolated for the air layer area, continuous spatial temperature information can be maintained even for sections not included in the camera's field of view. Additionally, IT power data of server equipment can be collected at the rack level or at the individual equipment level, and may include values ​​such as power consumption, computational resource usage, load growth rate, or load maintenance time. For example, if a specific rack corresponds to multiple 3D cubes, the power data of that rack can be distributed and recorded among the cubes included in the occupied area. Furthermore, if the collection cycles of temperature data and power data differ, reordering or interpolation processing can be performed to align them to the same time axis, thereby forming a time-series data structure that includes both temperature and power status at each time step.

[0502] Next, in the AI-based heat island prediction temperature calculation step (S1502), a data tensor combining 3D thermal volume data and server IT power data is input into an artificial intelligence model to calculate the predicted temperature of a topological heat island that may be formed in the future in a specific 3D cube area. More specifically, the artificial intelligence model receives the temperature change pattern and power load change pattern accumulated over multiple time points together to analyze the possibility of temperature rise in individual cubes and the possibility of forming a high-temperature region that continuously expands between adjacent cubes.

[0503] For example, if a temperature rise occurs simultaneously in the rear of a specific rack and the upper air layer, and the server load corresponding to that area increases significantly within a short period, that area can be predicted to become a region where heat accumulation is concentrated at a later point in time. Furthermore, the AI ​​model can calculate whether a specific area expands into a topological heat island by analyzing not only the absolute temperature of a single cube but also the connection structure of multiple adjacent cubes and changes in temperature distribution over time. In addition, a reference threshold temperature can be set using the maximum allowable temperature of the server equipment and the current operating temperature, and since the reference threshold temperature can be variably adjusted according to the rate of change in power load, different risk assessment conditions can be applied to load surge zones and load stability zones even under the same current temperature conditions.

[0504] Next, in the step for determining whether the dynamic threshold is exceeded (S1503), it can be determined whether the predicted temperature derived in the AI-based heat island predicted temperature calculation step (S1502) exceeds the dynamic threshold. More specifically, for each area consisting of a 3D cube or multiple cubes, the predicted temperature value and the dynamic threshold can be compared, and if the predicted temperature does not exceed the dynamic threshold, it is determined that the area is not immediately subject to control, and the process can return to the 3D heat volume data collection step (S1501).

[0505] For example, if the predicted temperature of a specific zone is 31°C and the dynamic threshold is 33°C, it may not be classified as a risk zone, and data from a later time point can be collected again to repeat the prediction process. On the other hand, if the predicted temperature exceeds the dynamic threshold, the zone may be identified as a risk zone. For example, if the predicted temperature of a specific zone is 36°C and the dynamic threshold is set to 33°C, the zone may be identified as a heat island-prone zone, and pre-control for local cooling may be performed in a later stage. Furthermore, since the dynamic threshold is a value that varies according to the allowable temperature of the server equipment and the rate of increase in power load, risk zones can be identified at an earlier point in time than with a simple fixed threshold method.

[0506] Next, in the pre-operation time calculation step (S1504), a pre-operation time T-minus can be calculated by reflecting the equipment delay time and the thermal inertia characteristics of the space for the area identified as a danger zone. More specifically, candidates for air conditioning equipment located within a certain distance can be identified using the 3D spatial coordinates of the danger zone, and the mechanical response time of each equipment, the time from the start of fan operation to the discharge of cold air, the cooling water control response time, and the temperature rise rate and heat accumulation characteristics of the danger zone can be used together.

[0507] For example, if a specific inlow cooler takes 8 seconds from receiving a control signal to actually supplying cold air, and a critical state is expected to be reached after 20 seconds based on the temperature rise rate in the hazardous zone and spatial heat accumulation characteristics, the pre-operation point can be calculated as approximately 12 seconds prior. Additionally, even with the same temperature difference, areas with severe air stagnation may experience a faster temperature rise, so the system may be configured to perform pre-operation at an earlier point. Conversely, in areas where cold air diffusion is relatively good, the pre-operation point may be delayed slightly. The T-minus calculated in this way is not a simple fixed value but can be updated at specific points based on spatial and equipment conditions.

[0508] Finally, in the local pre-control step (S1505), the target equipment is selectively operated according to the time calculated in the pre-operation time calculation step (S1504), and local pre-control with an airflow collision avoidance algorithm applied can be performed when multiple equipment operates simultaneously. More specifically, the air conditioning equipment closest to the danger zone or multiple equipment capable of having a substantial cooling effect on the danger zone may be selected, and the fan speed and cooling water control amount may be set to a value corresponding to the result of the local cooling load calculation.

[0509] For example, if a cooling requirement of 6 kW is calculated for a specific hazardous area, a fan speed and valve opening rate proportional to this can be set. In addition, when multiple air conditioning units discharge air simultaneously toward the same area, asymmetric control can be performed to prevent turbulence from occurring due to airflow colliding at the same point.

[0510] As another example, a method can be applied where the left unit maintains the airflow while slightly diverting the airflow direction outward, and the right unit reduces the airflow; or, a method can be applied where one unit maintains the upward discharge component and the other significantly increases the lateral discharge component. By adjusting the airflow or direction of each unit differently in this way, the air movement paths are formed differently, and consequently, the generation of turbulence caused by airflow collision can be suppressed. Furthermore, since the results of the pre-control can be reflected back into the thermal volume and power data and input into the next prediction cycle, a closed-loop structure can be formed in which prediction and control are repeatedly updated.

[0512] FIG. 16 is a diagram illustrating the concept of visualizing three-dimensional thermal volume data according to an embodiment of the present invention.

[0513] Referring to FIG. 16, the data center space has a structure in which a plurality of server racks are arranged at regular intervals, and the three-dimensional space including the top surface of each rack and the passage area can be divided and represented in units of three-dimensional cubes (Voxels). At this time, temperature data calculated through a method for generating a three-dimensional thermal volume based on physical-data fusion interpolation can be mapped to correspond to each three-dimensional cube, and the temperature value at each location can be converted into color and visualized in the form of a three-dimensional heatmap.

[0514] For example, if the temperature is relatively high in a specific rack surface area, that area may be displayed in a color indicating high temperature, and areas with low temperatures may be displayed in a color indicating low temperature. In this way, as the temperature distribution of each rack surface and aisle area is continuously represented corresponding to spatial coordinates, the temperature distribution status within the space can be intuitively verified.

[0515] In addition, each 3D cube can be configured to include coordinate information and temperature values, and multiple cubes can be combined to form a continuous set of temperature data for the entire space. Through this, a structure can be formed that allows simultaneous verification of the temperature distribution over the entire data center space, without being limited to a specific area.

[0516] Additionally, according to a specific embodiment, three-dimensional cubes of an area selected by a user may be highlighted, thereby allowing for focused verification of the temperature distribution of a specific rack surface or aisle area. For example, when a specific coordinate range is selected, the temperature values ​​of the cubes included in that range may be displayed in a highlighted color.

[0517] In addition, when data is updated over time, the temperature value of each 3D cube can be updated at regular intervals, and accordingly, the temperature distribution in 3D space can be expressed in a form that changes in real time. For example, as new temperature data is reflected at regular intervals, the trend of temperature increase or decrease in a specific area can be visually confirmed.

[0519] FIG. 17 is a flowchart illustrating the process of generating and processing three-dimensional thermal volume data according to an embodiment of the present invention.

[0520] Referring to FIG. 17, the process of generating and processing three-dimensional thermal volume data according to one embodiment of the present invention can be performed in a preprocessing step (S1701), a numerical analysis step (S1702), and a postprocessing step (S1703).

[0521] First, in the preprocessing step (S1701), geometric modeling of the data center space may be performed. More specifically, the spatial structure including server racks, aisles, floor surfaces, and air conditioning equipment may be defined as three-dimensional shape data, and each component may be set to correspond to actual placement coordinates. For example, the spacing between racks, aisle width, rack height, and the location of air conditioning equipment may be reflected identically in the numerical model. Additionally, in the preprocessing step, a computational grid for numerical analysis may be generated. More specifically, the entire space may be divided into a plurality of three-dimensional solid cubes or grid cells, and each grid cell may be set as a unit in which physical quantities such as temperature, flow rate, or pressure are defined. For example, the grid density may be differentially set such that the area around the rack top surface is subdivided into a relatively small grid, and the area with relatively little change is composed of a large grid.

[0522] Next, in the numerical analysis step (S1702), numerical analysis of heat and flow can be performed based on the generated grid structure. More specifically, governing equations to represent heat transfer and air flow within the space can be established, and a numerical solution can be applied. For example, heat transfer and air flow based on the temperature difference between each grid cell can be calculated, and iterative calculations can be performed to reflect changes over time. Additio...

Claims

Claim 1 A method performed on one or more processors of a computing device comprises: a step of constructing a time-series multidimensional data tensor by collecting three-dimensional thermal volume data generated by dividing a three-dimensional space into multiple three-dimensional cube (Voxel) units and IT power load data collected from server equipment corresponding to said three-dimensional cubes; a step of inputting said time-series multidimensional data tensor into an artificial intelligence model to predict future temperature changes for each three-dimensional cube area, calculating a dynamic threshold that is variably set by reflecting the hardware limit temperature and power load fluctuation rate of the server equipment, and identifying an area predicted to exceed said dynamic threshold as a heat island-prone area; a step of calculating the difference between the predicted temperature and the target maintenance temperature for said heat island-prone area, and calculating a local cooling load by reflecting the time-series temperature change characteristics of said three-dimensional thermal volume data; a step of identifying a plurality of adjacent air conditioning facilities based on the three-dimensional spatial coordinates of said heat island-prone area, and inversely calculating a pre-operation time (T-minus) by reflecting the mechanical operation delay time of each air conditioning facility and the thermal inertia coefficient of the space. A method for 3D volume thermal volume prediction-based data center targeting air conditioning control, comprising: a step of pre-operating selected air conditioning equipment according to the pre-operation time and performing cooperative control to suppress the generation of turbulence caused by collision between airflows by asymmetrically controlling the air volume or direction of each air conditioning equipment when multiple air conditioning equipment operates simultaneously; and a step of inversely calculating the pre-operation time, comprising: a step of identifying air conditioning equipment within a certain distance based on the 3D coordinates of a heat island-prone area; and a step of calculating the operation start time using time values ​​corresponding to the mechanical operation delay time and temperature change rate of each air conditioning equipment. Claim 2 A data center targeting air conditioning control method based on 3D thermal volume prediction, wherein the step of constructing the time series multidimensional data tensor comprises: a step of collecting temperature data for each 3D solid cube; and a step of collecting IT power load data collected from server equipment corresponding to the spatial coordinates of the 3D solid cube and combining them into a multidimensional array form arranged in chronological order. Claim 3 A data center targeting air conditioning control method based on 3D thermal volume prediction, wherein the step of identifying the heat island-prone zone comprises: a step of setting a reference threshold temperature using the maximum allowable temperature and current temperature of server equipment corresponding to each 3D cube; and a step of variably adjusting the reference threshold temperature according to the rate of change of power load of the server equipment. Claim 4 A method for air conditioning targeting a data center based on 3D thermal volume prediction, wherein the step of identifying the area where a heat island is expected to occur comprises: a step of predicting temperature changes over time using an artificial intelligence model that takes as input temperature data and power load data of a 3D cube collected over multiple time points; and a step of identifying a high-temperature area that is continuously formed in a specific area based on the connectivity status of the temperature distribution between adjacent 3D cubes. Claim 5 A method performed on one or more processors of a computing device comprises: a step of constructing a time-series multidimensional data tensor by collecting three-dimensional thermal volume data generated by dividing a three-dimensional space into multiple three-dimensional cube (Voxel) units and IT power load data collected from server equipment corresponding to said three-dimensional cubes; a step of inputting said time-series multidimensional data tensor into an artificial intelligence model to predict future temperature changes for each three-dimensional cube area, calculating a dynamic threshold that is variably set by reflecting the hardware limit temperature and power load fluctuation rate of the server equipment, and identifying an area predicted to exceed said dynamic threshold as a heat island-prone area; a step of calculating the difference between the predicted temperature and the target maintenance temperature for said heat island-prone area, and calculating a local cooling load by reflecting the time-series temperature change characteristics of said three-dimensional thermal volume data; a step of identifying a plurality of adjacent air conditioning facilities based on the three-dimensional spatial coordinates of said heat island-prone area, and inversely calculating a pre-operation time (T-minus) by reflecting the mechanical operation delay time of each air conditioning facility and the thermal inertia coefficient of the space. A method for 3D thermal volume prediction-based data center targeting air conditioning control, comprising: a step of pre-operating selected air conditioning equipment according to the pre-operation time and performing cooperative control to suppress the generation of turbulence caused by collision between airflows by asymmetrically controlling the air volume or air direction of each air conditioning equipment when multiple air conditioning equipment operates simultaneously; and a step of calculating the local cooling load, comprising: a step of calculating the difference between the predicted temperature and the target maintenance temperature; and a step of calculating the cooling requirement for each cube unit using the volume of the 3D solid cube and the time-series temperature change value corresponding to the difference. Claim 6 delete Claim 7 A method performed on one or more processors of a computing device comprises: a step of constructing a time-series multidimensional data tensor by collecting three-dimensional thermal volume data generated by dividing a three-dimensional space into multiple three-dimensional cube (Voxel) units and IT power load data collected from server equipment corresponding to said three-dimensional cubes; a step of inputting said time-series multidimensional data tensor into an artificial intelligence model to predict future temperature changes for each three-dimensional cube area, calculating a dynamic threshold that is variably set by reflecting the hardware limit temperature and power load fluctuation rate of the server equipment, and identifying an area predicted to exceed said dynamic threshold as a heat island-prone area; a step of calculating the difference between the predicted temperature and the target maintenance temperature for said heat island-prone area, and calculating a local cooling load by reflecting the time-series temperature change characteristics of said three-dimensional thermal volume data; a step of identifying a plurality of adjacent air conditioning facilities based on the three-dimensional spatial coordinates of said heat island-prone area, and inversely calculating a pre-operation time (T-minus) by reflecting the mechanical operation delay time of each air conditioning facility and the thermal inertia coefficient of the space. A method for air conditioning control targeting a data center based on three-dimensional thermal volume prediction, comprising: a step of pre-operating selected air conditioning equipment according to the pre-operation time and, when multiple air conditioning equipment is operating simultaneously, asymmetrically controlling the air volume or direction of each air conditioning equipment to suppress the generation of turbulence caused by collision between airflows; wherein the step of performing the cooperative control comprises: a step of reducing the air volume or changing the direction of some of the multiple air conditioning equipment; and a step of maintaining or increasing the air volume of other air conditioning equipment to disperse the air conditioning airflow so that it does not collide at the same point. Claim 8 A method performed on one or more processors of a computing device comprises: a step of constructing a time-series multidimensional data tensor by collecting three-dimensional thermal volume data generated by dividing a three-dimensional space into multiple three-dimensional cube (Voxel) units and IT power load data collected from server equipment corresponding to said three-dimensional cubes; a step of inputting said time-series multidimensional data tensor into an artificial intelligence model to predict future temperature changes for each three-dimensional cube area, calculating a dynamic threshold that is variably set by reflecting the hardware limit temperature and power load fluctuation rate of the server equipment, and identifying an area predicted to exceed said dynamic threshold as a heat island-prone area; a step of calculating the difference between the predicted temperature and the target maintenance temperature for said heat island-prone area, and calculating a local cooling load by reflecting the time-series temperature change characteristics of said three-dimensional thermal volume data; a step of identifying a plurality of adjacent air conditioning facilities based on the three-dimensional spatial coordinates of said heat island-prone area, and inversely calculating a pre-operation time (T-minus) by reflecting the mechanical operation delay time of each air conditioning facility and the thermal inertia coefficient of the space. A method for air conditioning control targeting a data center based on three-dimensional thermal volume prediction, comprising: a step of pre-operating selected air conditioning equipment according to the pre-operation time and performing cooperative control to suppress the generation of turbulence caused by collision between airflows by asymmetrically controlling the airflow volume or airflow direction of each air conditioning equipment when multiple air conditioning equipment operates simultaneously; wherein the step of performing cooperative control includes: a step of setting the airflow volume or cooling water control amount to a value corresponding to the cooling load for each of the multiple air conditioning equipment; and a step of adjusting the control value so that the airflow movement directions between the air conditioning equipment are formed in different directions. Claim 9 An apparatus comprising: a memory for storing one or more instructions; and a processor for executing the one or more instructions stored in the memory, wherein the processor performs any one of the methods of claims 1, 5, 7 and 8 by executing the one or more instructions. Claim 10 A computer program stored on a computer-readable recording medium that is combined with a computer, which is hardware, to enable the execution of any one of the methods of claims 1, 5, 7, and 8.

Citation Information

Patent Citations

  • Optimized thermal control of data center

    US20200100394A1