Prefabricated container deployment system and method for high-density compute data centers

Abstract

The application provides a prefabricated container deployment system and method for high-density computing power data centers, relating to the technical field of data center construction, the system comprises: a data acquisition module divides the computing power equipment inside the container and acquires real-time operation data; a heat dissipation priority construction module constructs a heat dissipation priority list; a heat dissipation data import module reads heat dissipation path structures and path switching nodes; an adaptive reconstruction module performs heat dissipation path reconstruction according to the heat dissipation priority list and heat dissipation path data sets, configures multiple heat dissipation strategies; and an intelligent deployment module performs three-dimensional adaptive analysis on the multiple heat dissipation strategies and performs intelligent deployment management. The technical problem of poor reliability and stability caused by the lack of fine perception and dynamic response in the heat dissipation management of the high-density computing power data center construction in the prior art is solved. The technical effects of rapid deployment, intelligent reconstruction and dynamic optimization of the heat dissipation path of the high-density computing power data center are achieved, and the operation stability and reliability are enhanced.

Description

Technical Field

[0001] This invention relates to the field of data center construction technology, and more specifically to a prefabricated container deployment system and method for high-density computing data centers. Background Technology

[0002] With the rapid development of artificial intelligence, high-performance computing, and other related services, data centers are continuously evolving towards higher power density and higher computing power density. GPU clusters with single-rack power exceeding 120kW are gradually becoming common configurations. This places higher demands on the heat dissipation capabilities, construction cycles, and deployment flexibility of data centers. Traditional data centers typically adopt an on-site construction model based on civil engineering. From scheme design and construction to final operation, a long cycle is often required. In overseas regions or areas with relatively weak infrastructure, the problems of long construction cycles, complex implementation, and insufficient deployment flexibility are particularly prominent, making it difficult to meet the needs of artificial intelligence computing power for rapid delivery and flexible deployment.

[0003] In terms of thermal management, existing data centers mostly adopt fixed heat dissipation paths or control methods based on simple thresholds, typically managing racks or individual devices. They lack the ability to finely perceive and dynamically respond to the differences in thermal load across different spatial zones of high-density computing data centers. Furthermore, these solutions primarily adjust for current temperature conditions, failing to proactively optimize based on trends in computing load changes. This can easily lead to lag in thermal response under high load scenarios or excessive heat dissipation under low load scenarios, resulting in increased energy consumption, decreased system stability, and higher operational costs.

[0004] Existing technologies suffer from problems such as long construction cycles, inflexible deployment, and a lack of refined perception and dynamic response in heat dissipation management, resulting in poor reliability and stability. Summary of the Invention

[0005] The purpose of this application is to provide a prefabricated container deployment system and method for high-density computing data centers, in order to solve the technical problems of long construction cycles, inflexible deployment, and lack of fine-grained perception and dynamic response in heat dissipation management in existing high-density computing data centers, which lead to poor reliability and stability.

[0006] In view of the above problems, this application provides a prefabricated container deployment system and method for high-density computing data centers.

[0007] The first aspect of this application provides a prefabricated container deployment system for high-density computing data centers. The prefabricated container deployment system for high-density computing data centers includes: a data acquisition module, used to divide the computing devices inside the container according to preset spatial partitions during the operation of the prefabricated container, and collect real-time operating data of the corresponding computing devices in each spatial partition; a heat dissipation priority construction module, used to partition and normalize the real-time operating data, construct a computing power heat load characterization value, and sort the spatial partitions according to the computing power heat load characterization value to construct a heat dissipation priority list; a heat dissipation data import module, used to read the heat dissipation path structure inside the prefabricated container, and read path switching nodes to construct a heat dissipation path dataset; an adaptation and reconstruction module, used to perform heat dissipation path reconstruction according to the heat dissipation priority list and the heat dissipation path dataset, and configure multiple heat dissipation strategies according to the heat dissipation path reconstruction results; and an intelligent deployment module, used to perform three-dimensional adaptation analysis under energy consumption-stability-prediction adaptation on multiple heat dissipation strategies, and perform intelligent deployment management according to the three-dimensional adaptation analysis results.

[0008] Optionally, the path structure parsing unit is used to parse the heat dissipation path structure inside the prefabricated container based on the heat dissipation path dataset, and divide it into N standardized heat dissipation path units, each standardized heat dissipation path unit corresponding to at least one controllable flow guiding component; the topology graph construction unit is used to construct a heat dissipation path topology graph composed of the N standardized heat dissipation path units and their connection relationships based on the path switching nodes, wherein the topology nodes represent standardized heat dissipation path units and the topology edges represent path switching relationships; the heat dissipation path reconstruction unit is used to perform path activation combination solving on the heat dissipation path topology graph under the constraints of the heat dissipation priority list, generate multiple candidate heat dissipation path combination schemes, and use the multiple candidate heat dissipation path combination schemes to complete the heat dissipation path reconstruction.

[0009] Optionally, the following components are included: a quantization object definition component for defining each standardized heat dissipation path unit as a quantization object, with quantization indicators including heat flow capacity, flow resistance, switching delay, and the state of controllable flow guiding components; a node weight construction component for associating the quantization object with the corresponding spatial partition heat load and priority value to construct node weights; a weighted heat dissipation path topology graph construction component for modeling path switching nodes and physical connectivity as edge weights, wherein the edge weights are jointly determined by switching delay, flow resistance, and mutual exclusion constraints, and constructing a weighted heat dissipation path topology graph using the node weights and edge weights; a path combination activation solution component for performing path combination activation solution based on a heuristic search algorithm on the weighted heat dissipation path topology graph after configuring constraints, wherein the constraints include the requirement that paths corresponding to the first priority label must be activated, the requirement that paths sharing flow guiding components or airflow channels cannot be activated simultaneously under the same heat dissipation path, and the requirement that thermally coupled partition paths perform linkage activation or suppression constraints; and a candidate combination scheme construction component for constructing multiple candidate heat dissipation path combination schemes based on the combination solution results.

[0010] Optionally, the prediction unit is used to acquire the preset tasks of the computing power equipment inside the container, perform computing power heat load characterization prediction based on the preset tasks and real-time operating data, and establish prediction results; the mapping unit is used to map the real-time energy consumption data, spatial partition temperature fluctuations, and prediction adaptation of each heat dissipation strategy into a three-dimensional heat dissipation strategy tensor based on the prediction results; the scoring matrix construction unit is used to perform adaptive normalization processing on the three-dimensional heat dissipation strategy tensor to construct a weighted heat dissipation adaptation scoring matrix; and the adaptation filtering unit is used to use the weighted heat dissipation adaptation scoring matrix to perform adaptation filtering of multiple heat dissipation strategies and establish three-dimensional adaptation analysis results.

[0011] Optionally, the real-time feedback acquisition module is used to collect real-time temperature, computing load, and energy consumption data of each spatial partition during the strategy execution process to establish a dynamic feedback dataset; the feedback update module is used to update the dynamic feedback dataset to the three-dimensional heat dissipation strategy tensor and optimize the three-dimensional adaptation analysis results based on the update results.

[0012] Optionally, the anomaly identification module is used to perform anomaly event triggering judgment based on the reading result after reading the dynamic feedback dataset, and output the judgment result; the emergency heat dissipation processing module is used to configure an emergency trigger signal based on the reading result when the judgment result indicates the existence of an anomaly event, and use the emergency trigger signal to perform emergency heat dissipation strategy optimization and heat dissipation control management.

[0013] Optionally, the prefabricated container integrates a high-density liquid-cooled computing cabinet and an air-cooled storage cabinet, which perform heat dissipation through a heat dissipation path structure.

[0014] Alternatively, prefabricated containers can be prefabricated, internally integrated, tested, and tuned in the factory, and then assembled and deployed via quick docking interfaces after being transported to the target project site.

[0015] Optionally, the historical test correction module is used to call up the historical heat dissipation dataset, perform three-dimensional adaptation analysis based on the historical heat dissipation dataset, and perform intelligent deployment management based on the correction results.

[0016] A second aspect of this application provides a method for deploying prefabricated containers for high-density computing data centers. The method includes: during the operation of the prefabricated container, dividing the internal computing devices of the container according to preset spatial partitions, and collecting real-time operating data of the corresponding computing devices in each spatial partition; performing partition normalization on the real-time operating data to construct a computing power heat load characterization value; sorting the spatial partitions based on the computing power heat load characterization value to construct a heat dissipation priority list; reading the heat dissipation path structure within the prefabricated container and reading path switching nodes to construct a heat dissipation path dataset; performing heat dissipation path reconstruction based on the heat dissipation priority list and the heat dissipation path dataset; configuring multiple heat dissipation strategies based on the heat dissipation path reconstruction results; performing a three-dimensional adaptation analysis under energy consumption-stability-prediction adaptation on the multiple heat dissipation strategies; and performing intelligent deployment management based on the three-dimensional adaptation analysis results.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages: The system employs a data acquisition module to divide the computing power devices inside the prefabricated container according to preset spatial partitions during operation, and collects real-time operating data of the corresponding computing power devices within each spatial partition. A heat dissipation priority construction module normalizes the real-time operating data by partition, constructs a computing power heat load characterization value, and sorts the spatial partitions based on this characterization value to build a heat dissipation priority list. A heat dissipation data import module reads the heat dissipation path structure within the prefabricated container and reads path switching nodes to construct a heat dissipation path dataset. An adaptation and reconstruction module performs heat dissipation path reconstruction based on the heat dissipation priority list and the heat dissipation path dataset, and configures multiple heat dissipation strategies based on the reconstruction results. An intelligent deployment module performs three-dimensional adaptation analysis under energy consumption-stability-prediction adaptation on multiple heat dissipation strategies, and performs intelligent deployment management based on the analysis results. This achieves the technical effects of rapid deployment of high-density computing power data centers, intelligent reconstruction and dynamic optimization of heat dissipation paths, improved heat dissipation efficiency and energy utilization, and enhanced operational stability and reliability.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the prefabricated container deployment system for high-density computing data centers provided in this application.

[0021] Figure 2 This is a structural schematic diagram of the prefabricated container deployment method for high-density computing data centers provided in this application.

[0022] Figure labeling: Data acquisition module 11, heat dissipation priority construction module 12, heat dissipation data import module 13, adaptation and reconstruction module 14, intelligent deployment module 15. Detailed Implementation

[0023] This application provides a prefabricated container deployment system and method for high-density computing data centers, addressing the technical problems of long construction cycles, inflexible deployment, and lack of refined perception and dynamic response in heat dissipation management in existing high-density computing data centers, resulting in poor reliability and stability. It achieves the technical effects of rapid deployment of high-density computing data centers, intelligent reconfiguration and dynamic optimization of heat dissipation paths, improved heat dissipation efficiency and energy utilization, and enhanced operational stability and reliability.

[0024] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0025] Example 1, as Figure 1 As shown, this application provides a prefabricated container deployment system for high-density computing data centers, the prefabricated container deployment system for high-density computing data centers comprising: The data acquisition module 11 is used to divide the computing power equipment inside the container according to the preset spatial partitions during the operation of the prefabricated container, and to collect the real-time operating data of the corresponding computing power equipment in each spatial partition.

[0026] Specifically, during the operation of prefabricated containers, the computing equipment inside the containers is divided into multiple independent spatial partitions with clearly defined boundaries according to a pre-set spatial partitioning strategy. The pre-set spatial partitioning rules refer to the basis for logically dividing the computing equipment inside the prefabricated containers based on their physical structure and thermal characteristics. For example, partitioning can be based on the physical arrangement of the server racks, hot / cold aisles or air supply / return paths, the boundary between liquid-cooled and air-cooled areas, or functional partitioning based on computing density and historical heat load characteristics. This ensures consistency in heat dissipation conditions and thermal coupling relationships for equipment within the same spatial partition. Computing equipment refers to data processing units deployed within prefabricated containers to perform computing tasks. During operation, these units generate the main heat load and constitute the core objects for heat dissipation control, including server nodes, GPU / AI accelerator cards, CPU computing nodes in high-density liquid-cooled computing racks, and their associated power modules and high-speed interconnect components.

[0027] After spatial partitioning is completed, multi-source sensor interfaces, such as power metering sensors, temperature sensors, and airflow sensors, deployed on computing devices, racks, and environmental nodes, are used to collect real-time data on the operating status of computing devices within each spatial partition. This real-time operating data includes parameters such as power consumption, current, voltage, temperature, and flow rate. The real-time operating data reflects the current workload and heat generation status of the computing devices. Spatial partitioning decomposes the complex internal environment of the container into multiple independently analyzable and controllable thermal management units. During the data collection process, data from different sources are synchronized and fundamentally verified to ensure the timeliness and consistency of data within each spatial partition. For example, in a prefabricated container divided into three spatial partitions according to rack arrangement, the first spatial partition contains five high-density liquid-cooled computing racks. Temperature sensors installed in this partition collect the air inlet and outlet temperature data of the racks every minute. Power metering sensors collect the power consumption of each rack in real time. The second spatial partition contains three air-cooled storage racks, and their temperature and power consumption are also collected using corresponding sensors. The third spatial partition is a mixed area, with two high-density liquid-cooled computing racks and one air-cooled storage rack, which also collects data in the same way.

[0028] By partitioning and collecting real-time operating data of computing devices within each spatial partition, precise basis is provided for heat dissipation priority construction, heat dissipation path reconstruction, and intelligent deployment analysis, thereby ensuring that subsequent heat dissipation strategies and deployment decisions are real-time, targeted, and executable.

[0029] The heat dissipation priority construction module 12 is used to partition and normalize the real-time running data, construct a computing power heat load characterization value, and sort the spatial partitions based on the computing power heat load characterization value to construct a heat dissipation priority list.

[0030] Specifically, spatial partitioning is used as the basic processing unit. Real-time operating data of computing devices in each spatial partition from different sources and with different dimensions are subjected to partition normalization processing. Partition normalization refers to unifying real-time collected operating parameters such as power consumption, temperature and liquid cooling flow rate within the same spatial partition into a relatively consistent range and scale through mathematical methods, such as Z-score normalization or maximum-minimum normalization, to eliminate the influence of dimensions and make the data comparable.

[0031] After normalization, a weighted fusion is performed based on each normalization parameter. Based on the weighted fusion result, a computing power heat load characterization value is constructed to represent the overall heat generation intensity and heat dissipation pressure of the spatial partition. This value comprehensively reflects the intensity of heat generated by computing devices within a unit spatial partition and their demand for heat dissipation resources. The computing power heat load characterization value H... i =w P ×P i +w T ×T i +w F ×F i , where P i T is the normalized value of power consumption. i F is the normalized value of the equipment temperature or average temperature. i Normalized value of liquid cooling flow rate, w P w T w FThe weighting coefficients for the corresponding parameters, with the sum of the weights being 1, are used to characterize the relative influence of each parameter on the thermal load of the spatial partition. The weights can be adaptively set based on experience or expert rules. Using the thermal load characterization value corresponding to each spatial partition as the sorting criterion, all spatial partitions are sorted from high to low to form a clear order of spatial partition heat dissipation requirements, thereby constructing a heat dissipation priority list. This heat dissipation priority list indicates which spatial partitions should be prioritized for heat dissipation paths and resources under the current operating state. For example, a spatial partition may contain 5 high-density liquid-cooled computing racks with an average power consumption of 15kW, while the maximum power consumption of the data center is 30kW. After normalizing the power consumption range of the entire data center, the normalized power consumption value is 0.5, the normalized equipment temperature value is 0.6, and the normalized liquid cooling flow rate value is 0.5. Based on experience or expert rules, weighting coefficients are set: power consumption weight 0.5, temperature weight 0.3, and liquid cooling flow rate weight 0.2. By substituting these values ​​into the thermal load characterization value formula, the thermal load characterization value for this area is obtained as 0.53. After calculating other spatial partitions in the same way, they can be sorted according to the size of the characterization value. The larger the value, the higher the heat generation intensity and the greater the heat dissipation pressure of the partition.

[0032] By transforming real-time operational data into comparable and sortable heat dissipation demand indicators, a clear decision-making basis is provided for subsequent heat dissipation path reconstruction and heat dissipation strategy configuration, avoiding the blind allocation of heat dissipation resources, thereby improving the overall heat dissipation efficiency and operational stability of high-density computing containers.

[0033] The heat dissipation data import module 13 is used to read the heat dissipation path structure inside the prefabricated container and read the path switching nodes to build a heat dissipation path dataset.

[0034] Specifically, the heat dissipation path structure inside the prefabricated container is read through an interface or configuration file. The heat dissipation path structure refers to the physical and logical network composed of liquid cooling circuits, air-cooled airflow channels, and their connection relationships and flow directions between cabinets. It is used to describe the transmission path of heat from the point of generation of computing equipment to the heat dissipation outlet. For example, the topology information of liquid cooling circuits and air-cooled channels, including pipe numbers, flow directions, connection relationships, and sensor locations, can be directly obtained through the management and control system installed inside the container, such as the BMS / SCADA interface. Alternatively, a pre-designed pipe and airflow layout scheme can be called through an interface or configuration file to identify information such as liquid cooling pipes, air ducts, inter-row air conditioning locations, radiator and fan arrangements, and obtain the heat dissipation path structure.

[0035] Simultaneously, the status registers of valves or dampers are read through the control system interface, or the path switching node information is read by defining the controllable attributes of each node through the configuration file. The path switching nodes refer to the controllable flow guiding or switching component positions in the heat dissipation path, such as valves, dampers, adjustable air guides, or fan outlets. These nodes can dynamically adjust the path channels to change the direction or distribution ratio of heat flow. The heat dissipation path structure and switching node data are then integrated to form a heat dissipation path dataset. This dataset uses nodes and path units as basic elements, recording the physical connections, controllable attributes, controllable states, and corresponding spatial partitioning relationships of each path unit, providing complete and structured data input for subsequent heat dissipation path reconstruction.

[0036] Taking a double-layer prefabricated container as an example, the upper IT terminal box is a standard 40-foot ultra-high container, integrating high-density liquid-cooled computing cabinets and air-cooled storage cabinets, and equipped with inter-row chilled water air conditioning to form a closed cold aisle environment, responsible for bearing all IT loads and providing liquid-cooled terminal heat dissipation and air-cooled heat dissipation. The lower power chilled water tank is also a 40-foot ultra-high container, integrating a complete secondary-side liquid-cooled distribution unit. It converts the primary-side chilled water (12℃ / 18℃) of the external chiller unit into secondary-side coolant (15℃ / 21℃) that meets the requirements of liquid-cooled servers through a plate heat exchanger, and provides constant pressure, filtration, distribution and metering functions, while not including primary pumps, cold storage and UPS power supply. The heat dissipation data import module 13 reads the pipeline layout from the secondary-side distribution unit of the lower power chilled water tank to the liquid-cooled loop of the upper IT terminal box through the interface or configuration file, including the path of coolant flowing into the liquid cooling plates of each computing cabinet, return pipeline and bypass valve position, and records path switching nodes such as valve opening and closing status and fan diversion port settings. By acquiring this structural information, the heat dissipation data import module 13 integrates the liquid cooling pipeline, air cooling channel and controllable switching node into a heat dissipation path dataset.

[0037] By integrating the heat dissipation structure and path switching nodes inside the prefabricated container into a heat dissipation path dataset, the adaptation and reconfiguration module 14 can dynamically schedule and optimize the heat flow on a real and controllable physical basis, thereby ensuring the feasibility and efficiency of the heat dissipation strategy.

[0038] The adaptation and reconstruction module 14 is used to perform heat dissipation path reconstruction based on the heat dissipation priority list and the heat dissipation path dataset, and to configure multiple heat dissipation strategies based on the heat dissipation path reconstruction results.

[0039] Furthermore, the adaptation and reconstruction module 14 includes: a path structure parsing unit, used to parse the heat dissipation path structure inside the prefabricated container based on the heat dissipation path dataset, and divide it into N standardized heat dissipation path units, each standardized heat dissipation path unit corresponding to at least one controllable flow guiding component; a topology graph construction unit, used to construct a heat dissipation path topology graph composed of the N standardized heat dissipation path units and their connection relationships based on the path switching nodes, wherein topology nodes represent standardized heat dissipation path units and topology edges represent path switching relationships; and a heat dissipation path reconstruction unit, used to perform path activation combination solving on the heat dissipation path topology graph under the constraints of the heat dissipation priority list, generate multiple candidate heat dissipation path combination schemes, and complete the heat dissipation path reconstruction using the multiple candidate heat dissipation path combination schemes.

[0040] Specifically, based on the heat dissipation path dataset, and combining the function of the heat dissipation path with the location of controllable flow guiding components, the heat dissipation path structure within the prefabricated container is analyzed. Continuous heat dissipation pipes or airflow channels are divided into N standardized heat dissipation path units, where N is a positive integer. Each standardized heat dissipation path unit corresponds to at least one controllable flow guiding component, such as a valve, air valve, or adjustable air guide plate, facilitating subsequent refined control and activation combination analysis. Taking liquid cooling pipes as an example, in the complete liquid cooling loop from the power chilled water tank to the IT terminal box, the circuit is divided according to the water flow direction and valve control range. In the liquid inlet stage, the pipe from the power chilled water tank outlet to the first flow regulating valve is considered a standardized heat dissipation path unit. This standardized heat dissipation path unit is mainly responsible for delivering the low-temperature coolant to the subsequent distribution stage. The distribution section is the collection of pipes from the first flow regulating valve to the liquid inlets of each IT device. Each branch pipe can be further subdivided or merged into a standardized heat dissipation path unit according to actual control requirements, used to accurately distribute the coolant to each device. The return section is the pipe returning from the IT device outlet to the power chilled water tank, responsible for returning the coolant after absorbing heat to the cold source. For air-cooled airflow channels, similar divisions are made based on the position of the air valve or adjustable air guide plate. For example, the airflow channel between the air outlet and the first air valve is regarded as a standardized heat dissipation path unit, and each standardized heat dissipation path unit is equipped with an independent flow regulating valve.

[0041] The types and locations of path switching nodes are clearly defined. Detailed information for each node is extracted from the heat dissipation path dataset, including its number, type, number of operable states, and the flow distribution ratio or airflow guidance information corresponding to each state. For example, for a three-way valve node, the flow distribution ratio of coolant to the two branch pipes at different opening degrees is recorded. Using graph theory tools, such as the NetworkX library in Python, the path switching node information is used to construct a heat dissipation path topology graph from N standardized heat dissipation path units and their connections. The nodes in the heat dissipation path topology graph represent standardized heat dissipation path units, i.e., the N standardized heat dissipation path units obtained from the decomposition. Specifically, N corresponding nodes are created in the topology graph, and attributes are set for each node, including node number, the type of heat dissipation path unit it belongs to, and standardized parameters. For example, node 1 represents the inlet section standardized unit, and its attributes record that the flow range of this unit is 5-20 L / min. Based on the connection relationships of the path switching nodes, edges are created in the topology graph, with each edge connecting two related nodes, representing the connection relationship between heat dissipation path units. Simultaneously, attributes are set for each edge, including the path switching node number, the number of operable states of the node, and the switching constraints corresponding to each state, such as flow distribution strategies and airflow guidance restrictions, forming a heat dissipation path topology graph. This graph reflects the structure of the heat dissipation path within the prefabricated container, the connection relationships between its parts, and the control constraints. The edges of the heat dissipation path topology graph represent path switching relationships and controllable flow guidance switching constraints. For example, the edge connecting the inlet section node and the distribution section node records the valve node number connecting them, as well as the flow distribution ratio from the inlet section to the distribution section under different valve opening degrees.

[0042] Constraint analysis is performed on the topology graph based on the heat dissipation priority list. Paths in spatial partitions with high thermal loads are set as high-priority activation objects. Heuristic search or combinatorial optimization algorithms are used to solve the path activation combination, generating multiple candidate heat dissipation path combination schemes. Each candidate heat dissipation path combination scheme includes the activation state and flow allocation strategy of different path units to meet the heat dissipation requirements of spatial partitions while taking into account overall energy consumption and stability.

[0043] The adaptation and reconfiguration module 14 organically combines zoned thermal load with the container's controllable heat dissipation path. Through path topology modeling and activation combination solving, it achieves precise heat dissipation control for high-load areas, thereby providing efficient, executable, and diverse heat dissipation strategy options for subsequent intelligent deployment, improving deployment flexibility, and ensuring efficient heat dissipation and system reliability of computing devices inside the container under high-density operating conditions.

[0044] Furthermore, the heat dissipation path reconstruction unit includes: a quantization object definition component, used to define each standardized heat dissipation path unit as a quantization object, the quantization indicators including heat flow capacity, flow resistance, switching delay, and controllable flow guiding component status; a node weight construction component, used to associate the quantization object with the corresponding spatial partition heat load and priority value to construct node weights; a weighted heat dissipation path topology graph construction component, used to model path switching nodes and physical connectivity as edge weights, the edge weights being jointly determined by switching delay, flow resistance, and mutual exclusion constraints, and using the node weights and edge weights to construct a weighted heat dissipation path topology graph; a path combination activation solution component, used to perform path combination activation solution based on a heuristic search algorithm on the weighted heat dissipation path topology graph after configuring constraints, the constraints including the requirement that the path corresponding to the first priority label must be activated, the requirement that paths sharing flow guiding components or airflow channels cannot be activated simultaneously under the same heat dissipation path, and the requirement that thermally coupled partition paths perform linkage activation or suppression constraints; and a candidate combination scheme construction component, used to construct multiple candidate heat dissipation path combination schemes based on the combination solution results.

[0045] Specifically, each standardized heat dissipation path unit is treated as a quantified object, which describes the comprehensive capabilities of the standardized heat dissipation path unit in terms of heat transfer and control. Its quantification indicators include at least heat flow capacity, flow resistance, switching delay, and the state of controllable flow guiding components. Among them, heat flow capacity reflects the maximum heat transfer capacity that can be carried per unit time. This can be achieved by setting high-precision temperature sensors and flow meters at the inlet and outlet of the standardized heat dissipation path unit to collect the mass of fluid flowing through the unit per unit time and the temperature difference between the inlet and outlet. The maximum heat transfer capacity that the unit can carry per unit time can be calculated using the heat flow calculation formula Qmax=cmΔT, where Q is the heat flow rate, c is the specific heat capacity of the fluid, m is the mass of the fluid per unit time, and ΔT is the temperature difference. For air-cooled path units, the calculation can be performed based on air volume, specific heat of air, and channel cross-sectional area. Flow resistance reflects the resistance loss of fluid or airflow within a specific heat dissipation path unit. It can be quantified using differential pressure measurement. Pressure sensors are installed at both ends of a standardized heat dissipation path unit. Fluid is allowed to flow through the unit at a certain velocity, and the pressure difference ΔP between the two ends is measured. This pressure difference reflects the resistance loss of fluid or airflow within the standardized heat dissipation path unit; a larger pressure difference indicates greater flow resistance. Switching delay refers to the response time required for a controllable flow guide component to switch from one state to another. It can be measured using high-precision timing equipment. By installing a state sensor on the controllable flow guide component, changes in its state can be accurately detected. When a state switching command is issued, the timing equipment is started. When the state sensor detects that the state switching is complete, the timing equipment stops, and the recorded time is the response time required for the controllable flow guide component to switch from one state to another. The state of the controllable flow guide component is used to describe the current or achievable adjustment capability of the path unit. Its quantification can be achieved by mapping the opening degree, working mode or adjustable range of the flow guide component to discrete or continuous values. For example, the closed state is recorded as 0, the fully open state is recorded as 1, and the partially open state is assigned a value according to the opening degree ratio, which is used to represent the adjustable capability of the path unit in the current or available state.

[0046] The quantized object is associated with the corresponding spatial partition's thermal load and priority value. Values ​​are assigned to the quantized object based on the thermal load level and priority weight, forming node weights that reflect the importance of the path. Specifically, the thermal load and heat dissipation priority of the spatial partition are normalized to obtain the computational thermal load characterization value and heat dissipation priority value corresponding to different spatial partitions. These two values ​​are then weighted and fused based on preset weight coefficients. The node weight can be expressed as N = w1 × L + w2 × P, where w1 and w2 are configurable weight parameters with a sum of 1, L represents the spatial partition's thermal load characterization value, and P represents the heat dissipation priority value. For example, if the thermal load weight w1 = 0.6, the priority weight w2 = 0.4, and the normalized thermal load of a certain spatial partition is 0.8, and the normalized priority is 0.7, then the node weight of the standardized heat dissipation path unit corresponding to this spatial partition is N = 0.6 × 0.8 + 0.4 × 0.7 = 0.76. For spatial partitions with high thermal load and high priority, the node weight corresponding to the associated standardized heat dissipation path unit will be relatively large.

[0047] Simultaneously, path switching nodes and physical connectivity are modeled as edge weights. Physical connectivity refers to the actual connection between heat dissipation paths, representing the connecting edges between them. In the topology graph, each path switching node and physical connectivity corresponds to an edge, with both ends of the edge connecting to the relevant heat dissipation path unit nodes. The edge weight comprehensively considers switching delay, flow resistance, and mutual exclusion constraints to reflect the costs and limitations of standardized heat dissipation path units during activation and switching. For example, if there is a mutual exclusion constraint between two paths, the edge weight between these two paths will be increased due to this constraint, indicating that the difficulty of activating these two paths simultaneously increases. By normalizing the switching delay and flow resistance, denoted as T and R respectively, ranging from 0 to 1, for mutual exclusion constraints, if there is mutual exclusion between two paths, the mutual exclusion factor M=1 is set; otherwise, M=0. Then, the edge weight E is determined by weighted summation. The weight of switching delay is set as w3, the weight of flow resistance as w4, and the weight of the mutual exclusion factor as w5, with w3 + w4 + w5 = 1. These weights can be set based on practical experience. The edge weight calculation formula is E = w3 × T + w4 × R + w5 × M. For example, if the switching delay weight is 0.3, the flow resistance weight is 0.4, the mutual exclusion factor weight is 0.3, the normalized switching delay of a certain edge is 0.5, the normalized flow resistance is 0.6, and there is a mutual exclusion constraint, then the edge weight E = 0.3 × 0.5 + 0.4 × 0.6 + 0.3 × 1 = 0.69.

[0048] A weighted heat dissipation path topology is constructed using pre-defined node and edge weights. Based on this topology, various constraints are configured, including: the requirement that paths corresponding to the highest priority label must be activated; the restriction that paths sharing flow guides or airflow channels cannot be activated simultaneously within the same heat dissipation path; and the requirement for linked activation or suppression of paths in thermally coupled spatial partitions. Specifically, the requirement that paths corresponding to the highest priority label must be activated, and the restriction that paths sharing flow guides or airflow channels cannot be activated simultaneously within the same heat dissipation path, prevent resource conflicts. Thermally coupled partitions influence each other in heat dissipation, thus requiring linked activation or suppression. For example, given three standardized heat dissipation paths A, B, and C, where path A has the highest priority, it must be activated during the solution process. If paths B and C share a flow guide, they cannot be activated simultaneously in a heat dissipation path combination. If two partitions are thermally coupled, when one partition's heat dissipation path is activated, the other partition is activated or suppressed according to the linkage rules.

[0049] Under the premise of satisfying the above constraints, a heuristic search algorithm is used to perform path combination activation solving in the weighted heat dissipation path topology graph. An initial solution set is generated based on the initial condition that the first priority path must be activated. The comprehensive cost function composed of node weights and edge weights is used as the evaluation index. During the search process, different combinations of activation and suppression of path units are gradually tried. For each candidate combination generated, it is verified whether it satisfies mutual exclusion constraints and thermal coupling linkage constraints, and the comprehensive score of the combination in terms of heat dissipation capacity, switching cost, and energy consumption is calculated. Based on the evaluation results, the optimal state is selected from the neighborhood states as the next current state. The neighborhood state with the highest feasibility and lowest cost can be selected. If multiple states satisfy the conditions, a certain random strategy can be used to select the next state to avoid getting trapped in local optima. Through iterative search and pruning mechanisms, infeasible or excessively costly combinations are gradually eliminated, retaining several path activation combinations that meet the heat dissipation requirements and have a relatively good comprehensive cost. Finally, multiple candidate heat dissipation path combination schemes are formed. Each candidate scheme explicitly gives the activation state of the path unit and the corresponding heat flow allocation strategy for subsequent heat dissipation strategy configuration and intelligent deployment decisions.

[0050] By quantizing heat dissipation path units and constructing a weighted heat dissipation path topology graph, various attributes and interrelationships of heat dissipation paths can be accurately described. With appropriate constraints and heuristic search algorithms, multiple executable candidate heat dissipation path combinations that meet heat dissipation requirements can be efficiently generated. This provides a scientific, comparable, and optimizable decision-making basis for subsequent heat dissipation strategy configuration and intelligent deployment, thereby effectively improving the heat dissipation efficiency and operational stability of high-density computing power prefabricated containers under complex load conditions.

[0051] The intelligent deployment module 15 is used to perform three-dimensional adaptation analysis on multiple heat dissipation strategies under the conditions of energy consumption-stability-prediction adaptation, and to perform intelligent deployment management based on the results of the three-dimensional adaptation analysis.

[0052] Furthermore, the intelligent deployment module 15 includes: a prediction unit, used to acquire the preset tasks of the computing power equipment inside the container, perform computing power heat load characterization prediction based on the preset tasks and real-time operating data, and establish prediction results; a mapping unit, used to map the real-time energy consumption data, spatial partition temperature fluctuations, and prediction adaptation of each heat dissipation strategy into a three-dimensional heat dissipation strategy tensor based on the prediction results; a scoring matrix construction unit, used to perform adaptive normalization processing on the three-dimensional heat dissipation strategy tensor to construct a weighted heat dissipation adaptation scoring matrix; and an adaptation filtering unit, used to use the weighted heat dissipation adaptation scoring matrix to perform adaptation filtering of multiple heat dissipation strategies and establish three-dimensional adaptation analysis results.

[0053] Specifically, the pre-set tasks of the computing equipment inside the container are obtained through the task scheduling system or computing power management platform interface. These pre-set tasks refer to the types, durations, and resource occupancy of scheduled or planned computing operations, reflecting the changing trends of computing load over a future period. The pre-set tasks are combined with currently collected real-time operating data to predict the computing heat load of each spatial partition, establishing a prediction result. This prediction result is used to predict the heat generated by the computing equipment under different tasks and operating states. The prediction can use a Long Short-Term Memory (LSTM) network model to predict the computing heat load of each spatial partition. The process is as follows: First, the real-time operating data of the computing equipment inside the container is organized according to a time series. Historical data is divided into continuous windows, each containing information such as computing load, power consumption, temperature, and air / liquid cooling status for several consecutive time steps, serving as the main input features of the prediction model. Pre-set task information, including task type, expected duration, and resource occupancy ratio, is embedded as auxiliary features in the input vector to reflect the impact of future load changes on heat generation. The output is the computational heat load value of each spatial partition after the corresponding time window, which represents the intensity of heat generation per unit spatial partition in the future time period.

[0054] The prediction model consists of an input layer, several LSTM layers, a fully connected layer, and an output layer. The input layer receives the current real-time running data time series and the device's preset task information. The first LSTM layer contains 64 hidden units to learn the short-term and medium-term dependencies of the time series. Setting `return_sequences=True` retains the output of each time step for the next layer. The second LSTM layer contains 32 hidden units to further extract long-term dependency features and outputs the hidden state vector of the last time step by setting `return_sequences=False`. The fully connected layer maps the output vector of the second LSTM layer to an output vector, where each node represents the predicted heat load value of the corresponding spatial partition. The output layer uses a linear activation function to output continuous value heat loads. The loss function uses mean squared error to measure the difference between the predicted and the true values.

[0055] During the training phase, a training set is formed by collecting historical equipment preset tasks, real-time running data, and actual measured spatial partition heat load data. The training steps are as follows: Training samples are input into the LSTM network in batches for forward propagation. The mean squared error loss between the predicted output and the actual heat load value of each sample is calculated. The gradient is calculated using the backpropagation algorithm and passed back from the output layer to the fully connected layer, the second LSTM layer, and the first LSTM layer. All trainable parameters in the network are updated, including LSTM unit weights, biases, and fully connected layer weights. The Adam optimizer is used, combined with a learning rate of 0.001, the first moment (mean gradient), and the second moment (uncentered variance gradient) for adaptive parameter updates, so that the prediction model can more accurately predict the heat load in the next iteration. After training, an independent validation dataset is input into the trained prediction model to evaluate the prediction error and fitting effect. If the error meets the preset threshold, the current model parameters are saved; otherwise, the hyperparameters are adjusted and retraining is performed until the performance requirements are met. Then, the current real-time operating data and the device's preset task information are organized into the same time series input as during training. The trained LSTM model is then input, and the predicted values ​​of computing power and heat load for each spatial partition in the future time period are output. These values ​​are used for the predicted adaptation dimension of the three-dimensional adaptation analysis, guiding the selection of heat dissipation strategies and intelligent deployment.

[0056] After obtaining the heat load prediction results, for each candidate heat dissipation strategy, its real-time energy consumption data, temperature fluctuation levels of each spatial partition, and the degree of matching of the strategy with the predicted heat load are read under the current operating conditions. The real-time energy consumption data comes directly from the energy consumption monitoring module. The temperature fluctuation level is obtained by statistically analyzing the temperature change amplitude of each spatial partition under strategy simulation or short-term operation. The prediction adaptation index is quantified by comparing the heat dissipation capacity provided by the heat dissipation strategy with the degree of matching with the predicted heat load. The real-time energy consumption data, temperature fluctuation levels of each spatial partition, and prediction adaptation index corresponding to the degree of matching with the predicted heat load for each candidate heat dissipation strategy under the current operating conditions are uniformly mapped to construct a three-dimensional heat dissipation strategy tensor with three dimensions: energy consumption, stability, and prediction adaptation. Each dimension reflects the energy efficiency performance, temperature control stability, and adaptability to future load changes of the heat dissipation strategy.

[0057] The three-dimensional heat dissipation strategy tensor is adaptively normalized using a max-min normalization algorithm to eliminate the influence of differences in units and numerical ranges between different dimensions, making the data comparable. Then, the weights of each dimension are dynamically set according to the current operational goals or strategy preferences. For example, stability and prediction adaptation weights are increased in high-load scenarios, while energy consumption weights are increased in energy-saving priority scenarios. The normalized data of the three dimensions are multiplied by their corresponding weights and then summed to obtain a comprehensive score for each heat dissipation strategy. This constructs a weighted heat dissipation adaptation score matrix. By assigning different weights to different dimensions, the weighted heat dissipation adaptation score matrix comprehensively considers the performance of each heat dissipation strategy in multiple aspects. Each element in the weighted heat dissipation adaptation score matrix corresponds to a comprehensive score value for a heat dissipation strategy. Multiple heat dissipation strategies are comprehensively sorted and filtered using the weighted heat dissipation adaptation score matrix to generate corresponding three-dimensional adaptation analysis results. Based on the analysis results, the optimal heat dissipation strategy is selected for deployment management, including heat dissipation path activation, equipment control parameter distribution, and operation strategy switching. This achieves intelligent deployment management of the heat dissipation system, thereby ensuring thermal stability while optimizing energy consumption and proactively adapting to future load changes.

[0058] The intelligent deployment module 15 improves the selection of heat dissipation strategy from a single real-time response to a comprehensive optimization decision-making based on future load changes by introducing predictive information and multi-dimensional adaptation analysis. This reduces system energy consumption while ensuring temperature stability, and enables intelligent and highly reliable deployment and operation of high-density computing power prefabricated containers.

[0059] Furthermore, the system also includes: a real-time feedback acquisition module, used to acquire real-time temperature, computing load, and energy consumption data of each spatial partition during strategy execution, and establish a dynamic feedback dataset; and a feedback update module, used to update the dynamic feedback dataset to the three-dimensional heat dissipation strategy tensor, and optimize the three-dimensional adaptation analysis results based on the update results.

[0060] Specifically, during strategy execution, real-time temperature, computing load, and energy consumption data for each spatial partition are collected through multi-source sensors and management interfaces. For temperature data, real-time temperatures at air inlets, inside racks, and at air outlets are obtained using temperature sensors deployed in each spatial partition. Computing load data is obtained by accessing the management interfaces built into the computing devices, such as IPMI, SNMP, or server monitoring APIs, to obtain CPU / GPU utilization, task queue information, and computing resource usage ratios. Energy consumption data is collected in real-time using power sensors and smart meters deployed in each rack or power unit. Data from different sensors and interfaces is aggregated by spatial partition and timestamped to generate a dynamic feedback dataset. Each record includes a spatial partition identifier, collection time, temperature value, computing load value, and energy consumption value, forming a dynamic information library that can be directly mapped to the three-dimensional heat dissipation strategy tensor by the feedback update module, achieving real-time closed-loop monitoring of the status.

[0061] The feedback update module maps and updates the dynamic feedback dataset to the previously constructed three-dimensional heat dissipation strategy tensor. The three dimensions of this tensor correspond to energy consumption, temperature control stability, and prediction adaptability, respectively. The update operation can be achieved by replacing or weighted fusion of real-time collected data with the original tensor values, ensuring that the three-dimensional heat dissipation strategy tensor always reflects the latest actual operating conditions. After the update, the weighted heat dissipation adaptability scoring matrix is ​​recalculated based on the new three-dimensional tensor. Adaptability analysis is performed on multiple heat dissipation strategies, adjusting strategy priorities and combination schemes to optimize the three-dimensional adaptability analysis results, thereby achieving dynamic strategy adjustment and optimized deployment.

[0062] By collecting and updating data in real time, the selection of heat dissipation strategies can be continuously optimized based on the latest operational information, thereby improving the effectiveness and reliability of deployment strategies. This ensures that heat dissipation efficiency, temperature stability, and energy consumption optimization can still be maintained when computing load or environmental conditions change, thus enhancing the stable and reliable operation capability of high-density computing prefabricated containers under complex and dynamic load conditions.

[0063] Furthermore, the system also includes: an anomaly identification module, used to perform an anomaly event trigger judgment based on the reading result after reading the dynamic feedback dataset, and output the judgment result; and an emergency heat dissipation processing module, used to configure an emergency trigger signal based on the reading result when the judgment result indicates the existence of an anomaly event, and use the emergency trigger signal to perform emergency heat dissipation strategy optimization and heat dissipation control management.

[0064] Specifically, after reading the dynamic feedback dataset, the real-time temperature, computing load, and energy consumption data of each spatial partition are compared with preset thresholds. Combined with rule-based judgment methods, such as temperature exceeding the safety limit, abnormal spikes in computing load, or sudden increases in energy consumption, an anomaly event judgment is triggered, and the judgment result is output. When the judgment result indicates the existence of an anomaly event, a corresponding emergency trigger signal is configured based on the specific anomaly data read. The emergency trigger signal contains key information such as the type, severity, and location of the anomaly. The emergency trigger signal initiates an emergency cooling strategy optimization process, quickly invoking multiple pre-stored emergency cooling strategies. The applicability and effectiveness of each emergency cooling strategy are evaluated based on the current anomaly situation to obtain the optimal emergency cooling strategy best suited to the current anomaly. Then, based on the selected optimal emergency cooling strategy, heat dissipation control management is executed, such as immediately opening key liquid cooling valves to full capacity, running fans at full speed, or adjusting the heat dissipation path combination. This achieves rapid temperature reduction and heat dissipation pressure relief for the abnormal spatial partition, completing safe and efficient emergency heat dissipation management.

[0065] By reading dynamic feedback datasets and determining abnormal events, potential problems during operation can be detected in a timely manner. Emergency trigger signals can be configured and emergency heat dissipation strategies can be optimized. This allows for rapid response when anomalies occur, effectively preventing further deterioration of the abnormal situation, ensuring the stable operation of computing equipment, and improving the reliability of the prefabricated container deployment system.

[0066] Furthermore, the prefabricated container integrates a high-density liquid-cooled computing cabinet and an air-cooled storage cabinet, which perform heat dissipation through a heat dissipation path structure.

[0067] Specifically, the prefabricated container integrates high-density liquid-cooled computing racks and air-cooled storage racks, with thermal management achieved through a unified and coordinated heat dissipation path structure. The high-density liquid-cooled computing racks are specifically designed for high-performance computing devices, employing liquid cooling. Internally installed liquid cooling plates or cold heads directly contact the server processors and high-power accelerator cards, directly absorbing heat generated by the computing equipment with coolant. This method boasts high heat dissipation efficiency and meets the cooling requirements of high-power devices. The liquid cooling pipeline circulates secondary coolant supplied by a lower-level power chiller tank, achieving efficient heat transfer from the heat source to the cold source. The air-cooled storage racks rely on in-row air conditioning and duct ventilation for air convection cooling. Cool air enters the rack through the inlet, absorbs heat through the duct, and is exhausted through the outlet. The heat dissipation capabilities of both types of racks are logically and physically integrated through the heat dissipation path structure, including liquid cooling loops, duct channels, and controllable airflow guiding components, forming a programmable and adjustable heat dissipation path that enables the coordinated operation of liquid and air cooling.

[0068] By integrating high-density liquid-cooled computing cabinets and air-cooled storage cabinets within prefabricated containers and designing a rational heat dissipation path structure, the advantages of different heat dissipation methods can be fully utilized to meet the heat dissipation requirements of different types of equipment. Liquid cooling can efficiently handle the large amount of heat generated by high-performance computing equipment, while air cooling can ensure the normal operation of storage equipment at a lower cost. The two work together to improve the heat dissipation efficiency and stability of the entire container, ensuring the reliable realization of computing and storage functions, thereby improving deployment effectiveness and reliability.

[0069] Furthermore, prefabricated containers are prefabricated, integrated, tested, and tuned in the factory. After being transported to the target project site, they are assembled and deployed through quick docking interfaces.

[0070] Specifically, the prefabricated container undergoes internal integration, testing, and calibration in the factory. This involves installing and interconnecting high-density liquid-cooled computing cabinets, air-cooled storage cabinets, power chilled water tanks, liquid-cooled circuits, air duct ventilation systems, and various sensors and control units according to design specifications. At the same time, functional testing and calibration are carried out, including computing equipment operation verification, heat dissipation path activation testing, temperature control response correction, and energy consumption monitoring calibration, to ensure that the prefabricated container has complete and stable operating capabilities before leaving the factory.

[0071] After passing factory testing and calibration, the prefabricated containers are transported to the target project site and assembled for deployment via quick-connect interfaces. These quick-connect interfaces, defined on the exterior of the prefabricated container, include coolant input / output ports, power ports, and network ports. All interfaces utilize quick connectors or flange designs, enabling connections between the prefabricated container and external infrastructure, such as chillers and power distribution systems, to be completed in a very short time, typically 5-30 minutes. Once delivered, the prefabricated containers, through these quick-connect interfaces, can rapidly deploy individual computing nodes of 1.2MW or higher with the on-site infrastructure, achieving plug-and-play assembly of racks, liquid cooling systems, and air cooling systems without complex on-site wiring or system debugging, thus shortening the deployment cycle and reducing installation complexity. Furthermore, multiple nodes can be expanded into larger-scale data center clusters through parallel or series configurations. External infrastructure includes chillers providing primary-side cooling for the entire system, cold storage devices, and backup power supplies. These facilities are deployed separately from the prefabricated container modules and can be centrally configured according to the overall project scale, providing greater redundancy and maintainability.

[0072] By prefabricating, integrating, and testing in the factory, each prefabricated container can be put into operation quickly, stably, and predictably after being transported to the site. This reduces the amount of on-site debugging work, provides a basic guarantee for the rapid expansion and standardized construction of high-density computing data centers, improves deployment efficiency, and reduces the difficulty and cost of on-site construction.

[0073] Furthermore, the system also includes a historical test correction module, which is used to call up historical heat dissipation dataset, perform three-dimensional adaptation analysis based on the historical heat dissipation dataset, and perform intelligent deployment management based on the correction results.

[0074] Specifically, the historical heat dissipation dataset is accessed through the historical test correction module. This dataset records the temperature distribution, energy consumption, and heat dissipation path response of the prefabricated container under different computing loads, environmental conditions, and heat dissipation strategies. The various indicators in the historical heat dissipation dataset, such as the temperature, energy consumption, and heat dissipation path response of each spatial partition under different computing loads, are organized according to timestamps and spatial partitions, and associated with the heat dissipation strategy information executed at that time. Each historical record is mapped to a corresponding three-dimensional heat dissipation strategy tensor: the energy consumption dimension uses actual power consumption data, the stability dimension uses temperature control deviation or temperature fluctuation amplitude, and the prediction adaptation dimension uses the difference between the predicted and actual thermal load under the task load.

[0075] After mapping, historical predicted values ​​are compared with actual measured values ​​to calculate the deviation for each spatial partition and each dimension, forming an error correction coefficient. This coefficient is the ratio of the actual value to the measured value, used to quantify the prediction error and deviation trend, providing a basis for subsequent correction. Based on the correction coefficient, the current 3D adaptation analysis results are dynamically adjusted, including correcting strategy scoring weights, adjusting heat dissipation path priorities, or optimizing heat dissipation strategy combinations. For example, in the energy consumption dimension, deviations are corrected by increasing or decreasing the strategy energy efficiency scoring weights; in the stability dimension, the temperature control sensitivity weights of each spatial partition are adjusted to improve the response accuracy of the temperature control strategy; and in the prediction adaptation dimension, the priority of heat dissipation paths or strategy combinations are optimized, such as setting paths in historically high-deviation partitions to higher priorities or adjusting path combinations, thereby improving the accuracy and applicability of strategy prediction. The corrected 3D adaptation analysis results are then used for intelligent deployment management, guiding heat dissipation path activation, airflow component adjustment, and heat dissipation strategy distribution, achieving optimal heat dissipation management under current operating conditions. Simultaneously, historical experience is incorporated into strategy prediction to improve the accuracy and applicability of strategy selection.

[0076] By correcting predictions and strategy selections using historical heat dissipation data, it is possible to more accurately match actual heat dissipation performance and spatial zoning requirements, thereby improving the reliability, energy efficiency, and temperature control stability of heat dissipation strategy execution, and ensuring that high-density computing power prefabricated containers achieve the expected thermal management effect in actual operation.

[0077] Example 2, based on the same inventive concept as the prefabricated container deployment system for high-density computing data centers in the foregoing examples, such as... Figure 2As shown, this application provides a method for deploying prefabricated containers for high-density computing data centers, wherein the method includes: During the operation of the prefabricated container, the computing devices inside the container are divided according to preset spatial partitions, and real-time operating data of the corresponding computing devices are collected in each spatial partition. The real-time operating data is partitioned and normalized to construct a computing power heat load characterization value. The spatial partitions are sorted based on the computing power heat load characterization value to construct a heat dissipation priority list. The heat dissipation path structure inside the prefabricated container is read, and the path switching nodes are read to construct a heat dissipation path dataset. Heat dissipation path reconstruction is performed according to the heat dissipation priority list and the heat dissipation path dataset. Multiple heat dissipation strategies are configured according to the heat dissipation path reconstruction results. Three-dimensional adaptation analysis under energy consumption-stability-prediction adaptation is performed on multiple heat dissipation strategies, and intelligent deployment management is performed according to the three-dimensional adaptation analysis results.

[0078] Furthermore, heat dissipation path reconstruction is performed based on the heat dissipation priority list and the heat dissipation path dataset. Multiple heat dissipation strategies are configured based on the reconstruction results, including: analyzing the heat dissipation path structure within the prefabricated container based on the heat dissipation path dataset, dividing it into N standardized heat dissipation path units, each corresponding to at least one controllable flow guiding component; constructing a heat dissipation path topology graph based on the path switching nodes, consisting of the N standardized heat dissipation path units and their connections, where topology nodes represent standardized heat dissipation path units and topology edges represent path switching relationships; and performing path activation combination solving on the heat dissipation path topology graph under the constraints of the heat dissipation priority list to generate multiple candidate heat dissipation path combination schemes, and using these multiple candidate heat dissipation path combination schemes to complete the heat dissipation path reconstruction.

[0079] Furthermore, under the constraints of the heat dissipation priority list, path activation combination solving is performed on the heat dissipation path topology to generate multiple candidate heat dissipation path combination schemes. This includes: defining each standardized heat dissipation path unit as a quantization object, with quantization indicators including heat flow capacity, flow resistance, switching delay, and controllable flow guiding component status; associating the quantization object with the corresponding spatial partition heat load and priority value to construct node weights; modeling path switching nodes and physical connectivity as edge weights, whereby the edge weights are jointly determined by switching delay, flow resistance, and mutual exclusion constraints; and constructing a weighted heat dissipation path topology using the node weights and edge weights; after configuring constraints, performing path combination activation solving on the weighted heat dissipation path topology based on a heuristic search algorithm, whereby the constraints include the requirement that paths corresponding to the first priority label must be activated, the requirement that paths sharing flow guiding components or airflow channels cannot be activated simultaneously under the same heat dissipation path, and the requirement that thermally coupled partition paths perform linkage activation or suppression constraints; and constructing multiple candidate heat dissipation path combination schemes based on the combination solution results.

[0080] Furthermore, the three-dimensional adaptation analysis of multiple heat dissipation strategies under the energy consumption-stability-prediction adaptation includes: obtaining the preset tasks of the computing power equipment inside the container; performing computing power heat load characterization prediction based on the preset tasks and real-time operating data, and establishing prediction results; mapping the real-time energy consumption data, spatial partition temperature fluctuations, and prediction adaptation of each heat dissipation strategy into a three-dimensional heat dissipation strategy tensor based on the prediction results; performing adaptive normalization processing on the three-dimensional heat dissipation strategy tensor to construct a weighted heat dissipation adaptation scoring matrix; and using the weighted heat dissipation adaptation scoring matrix to perform adaptation screening of multiple heat dissipation strategies and establish three-dimensional adaptation analysis results.

[0081] Furthermore, the method also includes: during the strategy execution process, collecting real-time temperature, computing load, and energy consumption data of each spatial partition to establish a dynamic feedback dataset; updating the dynamic feedback dataset to the three-dimensional heat dissipation strategy tensor; and optimizing the three-dimensional adaptation analysis results based on the update results.

[0082] Furthermore, the method also includes: after reading the dynamic feedback dataset, performing an abnormal event triggering judgment based on the reading result and outputting the judgment result; when the judgment result indicates that an abnormal event exists, configuring an emergency trigger signal based on the reading result, using the emergency trigger signal to perform emergency heat dissipation strategy optimization, and performing heat dissipation control management.

[0083] Furthermore, the prefabricated container integrates a high-density liquid-cooled computing cabinet and an air-cooled storage cabinet, which perform heat dissipation through a heat dissipation path structure.

[0084] Furthermore, prefabricated containers are prefabricated, integrated, tested, and tuned in the factory. After being transported to the target project site, they are assembled and deployed through quick docking interfaces.

[0085] Furthermore, the method also includes: after calling the historical heat dissipation dataset, performing three-dimensional adaptation analysis and correction based on the historical heat dissipation dataset, and performing intelligent deployment management based on the correction results.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0087] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A prefabricated container deployment system for high-density computing data centers, characterized in that: The system includes: The data acquisition module is used to divide the computing power equipment inside the prefabricated container according to the preset spatial partitions during the operation of the container, and to collect the real-time operating data of the corresponding computing power equipment in each spatial partition. The heat dissipation priority construction module is used to partition and normalize the real-time running data, construct a computing power heat load characterization value, and sort the spatial partitions based on the computing power heat load characterization value to construct a heat dissipation priority list. The heat dissipation data import module is used to read the heat dissipation path structure inside the prefabricated container, read the path switching nodes, and construct a heat dissipation path dataset. The adaptation and reconstruction module is used to perform heat dissipation path reconstruction based on the heat dissipation priority list and the heat dissipation path dataset, and to configure multiple heat dissipation strategies based on the heat dissipation path reconstruction results. The intelligent deployment module is used to perform three-dimensional adaptation analysis on multiple heat dissipation strategies under the conditions of energy consumption, stability and prediction, and to perform intelligent deployment management based on the results of the three-dimensional adaptation analysis.

2. The prefabricated container deployment system for high-density computing data centers as described in claim 1, characterized in that, Adapted refactoring modules, including: The path structure parsing unit is used to parse the heat dissipation path structure inside the prefabricated container based on the heat dissipation path dataset, and divide it into N standardized heat dissipation path units, each of which corresponds to at least one controllable flow guiding component. The topology graph construction unit is used to construct a heat dissipation path topology graph composed of the N standardized heat dissipation path units and their connection relationships based on the path switching nodes, wherein the topology nodes represent standardized heat dissipation path units and the topology edges represent path switching relationships. The heat dissipation path reconstruction unit is used to perform path activation combination solving on the heat dissipation path topology graph under the constraints of the heat dissipation priority list, generate multiple candidate heat dissipation path combination schemes, and complete the heat dissipation path reconstruction using multiple candidate heat dissipation path combination schemes.

3. The prefabricated container deployment system for high-density computing data centers as described in claim 2, characterized in that, The heat dissipation path reconfiguration unit includes: The quantization object definition component is used to define each standardized heat dissipation path unit as a quantization object. The quantization indicators include heat flow capacity, flow resistance, switching delay, and the status of controllable flow guiding components. The node weight construction component is used to associate quantized objects with corresponding spatial partition thermal load and priority values ​​to construct node weights; A weighted heat dissipation path topology graph construction component is used to model path switching nodes and physical connectivity as edge weights. The edge weights are jointly determined by switching delay, flow resistance and mutual exclusion constraints. The weighted heat dissipation path topology graph is constructed using the node weights and edge weights. The path combination activation solution component is used to perform path combination activation solution based on a heuristic search algorithm on a weighted heat dissipation path topology map after configuring constraints. The constraints include the requirement that the path corresponding to the first priority label must be activated, the requirement that paths sharing a flow guiding component or airflow channel cannot be activated simultaneously under the same heat dissipation path, and the requirement that thermally coupled partitioned paths perform linkage activation or suppression constraints. The candidate combination scheme construction component is used to construct multiple candidate heat dissipation path combination schemes based on the combination solution results.

4. The prefabricated container deployment system for high-density computing data centers as described in claim 1, characterized in that, The intelligent deployment module includes: The prediction unit is used to obtain the preset tasks of the computing power equipment inside the container, perform computing power heat load characterization prediction based on the preset tasks and real-time operating data, and establish prediction results. The mapping unit is used to map the real-time energy consumption data, spatial partition temperature fluctuations and prediction adaptation of each heat dissipation strategy into a three-dimensional heat dissipation strategy tensor based on the prediction results. The scoring matrix construction unit is used to adaptively normalize the three-dimensional heat dissipation strategy tensor and construct a weighted heat dissipation adaptation scoring matrix. The adaptation filtering unit is used to perform adaptation filtering of multiple heat dissipation strategies using the weighted heat dissipation adaptation scoring matrix and establish three-dimensional adaptation analysis results.

5. The prefabricated container deployment system for high-density computing data centers as described in claim 4, characterized in that, The system also includes: The real-time feedback acquisition module is used to collect real-time temperature, computing load, and energy consumption data of each spatial partition during the strategy execution process, and to establish a dynamic feedback dataset. The feedback update module is used to update the dynamic feedback dataset into the three-dimensional heat dissipation strategy tensor and optimize the three-dimensional adaptation analysis results based on the update results.

6. The prefabricated container deployment system for high-density computing data centers as described in claim 5, characterized in that, The system also includes: The anomaly detection module is used to perform anomaly event trigger judgment based on the reading result after reading the dynamic feedback dataset, and output the judgment result; The emergency heat dissipation module is used to configure an emergency trigger signal based on the reading result when the determination result indicates the existence of an abnormal event, and to perform emergency heat dissipation strategy optimization and heat dissipation control management using the emergency trigger signal.

7. The prefabricated container deployment system for high-density computing data centers as described in claim 1, characterized in that, The prefabricated container integrates a high-density liquid-cooled computing cabinet and an air-cooled storage cabinet, which perform heat dissipation through a heat dissipation path structure.

8. The prefabricated container deployment system for high-density computing data centers as described in claim 1, characterized in that, Prefabricated containers are prefabricated, integrated, tested, and tuned in the factory. After being transported to the target project site, they are assembled and deployed through quick docking interfaces.

9. The prefabricated container deployment system for high-density computing data centers as described in claim 1, characterized in that, The system also includes: The historical test correction module is used to call up historical heat dissipation datasets, perform three-dimensional adaptation analysis based on the historical heat dissipation datasets, and perform intelligent deployment management based on the correction results.

10. A method for deploying prefabricated containers for high-density computing data centers, characterized in that, The method for deploying prefabricated containers for high-density computing data centers, performed by any one of claims 1 to 9, comprises: During the operation of prefabricated containers, the computing devices inside the containers are divided according to the preset spatial partitions, and real-time operating data of the corresponding computing devices are collected in each spatial partition. The real-time running data is partitioned and normalized to construct a computing power heat load characterization value. The computing power heat load characterization value is then used to spatially partition and sort the data to construct a heat dissipation priority list. Read the heat dissipation path structure inside the prefabricated container and read the path switching nodes to construct a heat dissipation path dataset; Perform heat dissipation path reconstruction based on the heat dissipation priority list and the heat dissipation path dataset, and configure multiple heat dissipation strategies based on the heat dissipation path reconstruction results. Perform a three-dimensional adaptation analysis on multiple heat dissipation strategies under the conditions of energy consumption, stability, and prediction, and perform intelligent deployment management based on the results of the three-dimensional adaptation analysis.

Images