Fully prefabricated decoupled data center construction method

Abstract

The present invention discloses a method for constructing a fully prefabricated decoupled data center, which relates to the technical field of data processing. The present invention breaks through the serial dependence of traditional civil engineering and equipment deployment through modular design of container components and factory prefabrication production, realizes the parallel advancement of civil engineering construction and module production, and optimizes the transportation plan and assembly sequence with the help of intelligent scheduling algorithms, thereby significantly shortening the construction period; in the module prefabrication and assembly stages, a unified interface standard is adopted, and in combination with digital twin technology, the compatibility and matching of the module interface are verified through real-time data comparison, effectively reducing the readjustment cost caused by on-site construction errors and improving the success rate of module assembly; in addition, in the operation stage of the data center, a deep learning algorithm is used to predict load dynamics and dynamically adjust the cooling power and path; at the same time, combined with reinforcement learning technology, the linkage resource allocation of power supply and distribution and refrigeration modules is realized to maximize the system energy efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for constructing a fully prefabricated decoupled data center. Background Art

[0002] With the rapid development of cloud computing, big data, and artificial intelligence, data centers, as the core infrastructure of the digital economy, have become increasingly large and complex. However, the construction and operation modes of traditional data centers have exposed many problems when meeting the current high demand. These problems run through the construction cycle and operation process, seriously restricting the efficiency, flexibility, and sustainable development of data centers.

[0003] First, the construction period of data centers is generally long and inefficient. The traditional construction model relies on the deployment of mechanical and electrical equipment after the completion of civil engineering. This sequential approach creates a large time gap between each link, resulting in resource waste and construction delays. In addition, the construction process requires frequent coordination among multiple teams such as civil engineering, power supply and distribution, and refrigeration. The slow response to sudden problems further slows down the project progress. Although the introduction of the modular concept has achieved the prefabrication of some equipment to a certain extent, the flexibility and efficiency of modular deployment have not yet reached the ideal level due to the lack of comprehensive decoupling design.

[0004] Secondly, during the operation stage of the data center, the energy consumption problem of the refrigeration system is particularly prominent; the traditional refrigeration system is mainly air-cooled. In the scenario of high-density server deployment, the energy efficiency is limited and it is easy to form a local heat island effect. The cooling demand rises rapidly, resulting in a high overall PUE value (power utilization efficiency); although some data centers have tried to use technologies such as liquid cooling or energy-saving air cooling, they lack intelligent management and it is difficult to dynamically adjust the cooling strategy according to the heat source distribution and equipment load. It is easy to have problems of overcooling or insufficient cooling, affecting the stability of equipment operation;

[0005] In addition, the power supply architecture of traditional data centers is usually designed with high redundancy as the goal, and the reliability of power supply is guaranteed through dual power supply and UPS (uninterruptible power supply); however, this high redundancy design often leads to idle equipment resources and reduces overall resource utilization; as clean energy is gradually introduced into data centers, the power supply and distribution system needs to dynamically dispatch multiple power sources, but the static design of the traditional power supply architecture is difficult to adapt to this demand, resulting in low utilization of renewable energy and exacerbating carbon emissions.

[0006] Although some data centers are currently trying to optimize the above problems through modular design, distributed temperature control and dynamic power supply management, most of these solutions remain at the level of local optimization and lack a global perspective. Therefore, there is an urgent need for a more systematic and intelligent data center construction solution to address the shortcomings of existing technologies and promote the development of data centers in an efficient, green and flexible direction. Summary of the invention

[0007] In view of the above existing problems, the present invention is proposed.

[0008] The present invention provides a method for constructing a fully prefabricated decoupled data center to solve the problems that the sequential construction mode forms a large time gap between each link, resulting in resource waste and construction delays, and it is difficult to deal with sudden problems; the refrigeration system has limited energy efficiency and is prone to form a local heat island effect, and the overall PUE value is high; the power supply architecture is difficult to adapt to the problem of dynamically scheduled electricity from multiple sources.

[0009] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0010] The embodiment of the present invention provides a method for constructing a fully prefabricated decoupled data center, which includes:

[0011] Step S1: All equipment in the data center is prefabricated in the factory using a container assembly model, and the deployment sequence of each module is generated using digital twin technology;

[0012] Step S2, civil construction is carried out simultaneously with module prefabrication production. Based on the deployment sequence, an intelligent scheduling algorithm is used to adjust the module transportation schedule, determine the module deployment schedule, and pre-embed the interface;

[0013] Step S3: After the modules are prefabricated in the factory, they are transported to the site in order for assembly according to the deployment schedule after the civil construction is completed. After the assembly is completed, the modules are connected in combination with the embedded interfaces to start the global joint debugging work;

[0014] Step S4, using the initial parameters of the global joint debugging phase as input, deploying distributed sensors to monitor heat source distribution and load changes, using a deep learning algorithm to predict load changes, and adjusting the cooling power and path based on load changes;

[0015] Step S5, monitor the data of each module during the operation phase of the entire data center, continuously update the digital twin model in real time, and verify the newly added modules when new modules are added to verify the compatibility between the modules.

[0016] As a preferred solution of the method for constructing a fully prefabricated decoupled data center described in the present invention, the equipment of the data center is divided into a power supply and distribution module, a refrigeration module, an IT equipment module and a fire protection module, and the production process formulates the module interface according to the interface standard;

[0017] The method of using digital twin technology to generate the deployment order of each module is as follows:

[0018] Digital twin technology is used to generate a virtual model of each module, simulating its operating characteristics, including load capacity, energy consumption indicators, heat generation and interface matching, verifying the compatibility between modules, generating the module deployment sequence, and coordinating the module production plan based on the deployment sequence.

[0019] As a preferred solution for the method of constructing a fully prefabricated decoupled data center described in the present invention, during the civil construction process, the interfaces required for the modules are embedded in advance to determine the module location planning. The embedded interfaces include power supply channels, cooling pipes, clean energy access points, energy storage systems, and signal connection points. The embedded interfaces match the interface standards.

[0020] As a preferred solution of the method for constructing a fully prefabricated decoupled data center described in the present invention, during the on-site assembly process, the on-site assembled interface data is compared with the digital twin model data of each module to detect the assembly matching and compatibility.

[0021] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, in which: in step S4, the power supply and distribution module is linked with the refrigeration module, and a reinforcement learning algorithm is used to allocate power and cooling resources;

[0022] The pre-buried clean energy access point and energy storage system in step S2 are put into use, and photovoltaic power generation and phase change energy storage modules are combined to achieve peak load shaving and valley filling of the grid load, thereby improving energy efficiency.

[0023] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, the steps of using digital twin technology to generate a virtual model of each module and simulating its operating characteristics are as follows:

[0024] Collect the design parameters and preliminary operation data of the module, define the input variables of the virtual model, the module includes module type, module size and operation characteristic parameters,

[0025] The module type is , They represent power supply and distribution module, cooling module, IT equipment module and fire protection module respectively.

[0026] The module size is , are the length, width and height of the module respectively.

[0027] The operating characteristic parameters include the maximum load capacity of , the static energy consumption of the module is , the module dynamic energy consumption is and the heat generation efficiency is ;

[0028] Combined with the collected parameters, simulation algorithms are used to simulate the operating characteristics of the module, including load capacity, energy consumption indicators and heat generation.

[0029] To simulate the load capacity, the formula is:

[0030] ,

[0031] in, is the actual load capacity of the module, is the maximum design load of the module, is the safety factor,

[0032] Carry out energy consumption module, the formula is:

[0033] ,

[0034] in, For the module at time The total energy consumption, is the basic energy consumption of the module in static standby state, is the dynamic power consumption of the module when running the load, depending on the load level, is the simulation running time,

[0035] Based on the energy consumption simulation, the heat generated by the calculation module during operation is calculated as follows:

[0036] ,

[0037] in, For the module at time of heat generation, is the heat generation efficiency, which indicates the ratio of energy consumption converted into heat. For the module at time The total energy consumption,

[0038] The steps of verifying the compatibility between modules are:

[0039] Define the interface parameters for each module:

[0040] The power interface standard is ,in Indicates voltage, represents the current,

[0041] Cooling interface standard is ,in Indicates the refrigerant flow rate, Indicates the refrigerant temperature.

[0042] The signal interface standard is , indicating the communication protocol of the module;

[0043] Check Module With module Interface compatibility,

[0044] like, and and ,but ,otherwise ,in, For modules and modules Interface compatibility flag, 1 means compatible, 0 means incompatible, For modules Power, cooling and signal interface parameters;

[0045] The steps of generating the module deployment sequence are:

[0046] According to the simulation results, the module's load capacity, energy consumption, heat generation, and interface compatibility are considered, and a priority sorting method is used to generate a deployment order. The priority sorting formula is:

[0047] ,

[0048] in, For modules The deployment priority score of is a weight parameter used to balance the impact of load capacity, energy consumption and heat on priority, satisfying , is the load capacity of the module, is the total energy consumption of the module, For the heat generation of the modules, the modules are sorted from high to low according to the priority value to generate the module deployment order.

[0049] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, the step of adjusting the module transportation schedule based on the deployment sequence and using an intelligent scheduling algorithm to determine the module deployment schedule is as follows:

[0050] Model the transport and deployment of modules and define variables and constraints, including module attributes, vehicle transport characteristics, and time variables.

[0051] Module attributes include: Modules , module Transportation distance , module Loading time And modules On-site deployment time ,

[0052] Vehicle transport characteristics include: Transport vehicles ,vehicle Loading capacity and vehicles Average transportation speed ,

[0053] Time variables include: Module Start time of transportation , module Time of arrival at the scene And modules Time when deployment is completed ;

[0054] Calculate the module's arrival time and deployment schedule based on the module's transportation distance and the vehicle's transportation characteristics,

[0055] Modules The formula for calculating the arrival time is:

[0056] ,

[0057] in, For modules Arrival time, For modules The start time of transportation, For modules The loading time, For modules The transport distance, For vehicles The average transport speed,

[0058] The calculation formula for module deployment completion time is:

[0059] ,

[0060] in, For modules Deployment completion time, For modules field deployment time;

[0061] The steps of using the intelligent scheduling algorithm to adjust the module transportation schedule are:

[0062] Genetic algorithm is used for transportation scheduling, and the goal is to minimize the overall deployment time, which is expressed as:

[0063] ,

[0064] in, The total time for all module deployments to complete, For modules The deployment completion time,

[0065] The defined constraints include vehicle capacity constraints, module deployment order constraints, and field deployment order constraints.

[0066] Vehicle capacity limits are:

[0067] ,

[0068] in, Assigned to vehicles A collection of modules,

[0069] The module deployment order constraint is: if the module Dependency modules , then it must satisfy ,

[0070] The on-site deployment sequence constraint is: the maximum number of modules deployed in parallel on-site in the same time period Limited, ,in, Representation Module At the moment Deployment is in progress;

[0071] Each chromosome of the genetic algorithm is an arrangement of the transport and deployment order of a module.

[0072] The fitness function is: ,in For fitness, maximize fitness ,

[0073] Cross-operation adjusts the transportation allocation of modules, and mutation operation changes the deployment order of modules.

[0074] Under the constraints, the module transportation and deployment schedule is gradually optimized through multiple generations of evolution to obtain the final module transportation and deployment schedule, including:

[0075] The start time of shipping each module , Arrival time and time to complete deployment , combined with the interface pre-embedded plan for on-site construction, coordinate the transportation and assembly of modules according to the schedule.

[0076] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, the step of comparing the interface data assembled on site with the digital twin model data of each module to detect the assembly matching and compatibility is as follows:

[0077] Collect actual parameters of the field interface through sensors and detection equipment, and define interface properties, including:

[0078] Power interface properties: actual voltage and the actual current ,

[0079] Cooling interface properties: The actual refrigerant flow is and actual refrigerant temperature ,

[0080] Signal interface properties: communication protocol ;

[0081] Extract the standard interface properties of the corresponding modules from the digital twin model, including:

[0082] Power interface standard: Model voltage and model current ,

[0083] Cooling interface standard: Model refrigerant flow and model refrigerant temperature ,

[0084] Signal interface standard: Model communication protocol is ;

[0085] According to the comparison between the actual power interface parameters and the model data, the matching degree of the interface voltage and current is calculated as follows:

[0086] like and ,but ,otherwise ,in, is the allowable voltage error range, is the allowable current error range;

[0087] Compare the flow rate and temperature of the cooling interface and calculate the matching degree. The calculation method is:

[0088] like and ,but ,otherwise ,in, is the allowable refrigerant flow error range, is the allowable refrigerant temperature error range;

[0089] Compare the communication protocols of the signal interfaces and check compatibility. The calculation method is:

[0090] like ,but ,otherwise ;

[0091] The weighted method is used to comprehensively evaluate the matching score, and the calculation method is as follows:

[0092] ,in, For modules The matching score of is the weight of matching detection, satisfying ,

[0093] The compatibility of the modules is determined based on the comprehensive score. The judgment method is as follows:

[0094] like ,but ,otherwise ,in, For modules The compatibility determination result is 1 for compatible and 0 for incompatible. is the minimum score threshold for compatibility;

[0095] If a module fails the matching or compatibility test , then take corresponding measures according to the mismatched items.

[0096] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, the step of using a deep learning algorithm to predict load changes and adjusting the cooling power and path based on load changes is as follows:

[0097] The initial operating parameters of the system are obtained from the joint debugging output of step S3, including:

[0098] Heat source distribution matrix , indicating that the data center Initial heat distribution at the grid points,

[0099] Load Distribution Matrix ,express Initial calculation loads at the grid points;

[0100] Deploy sensor networks to collect data in real time: The current heat source distribution is: , the current load distribution is and the ambient temperature is ,

[0101] The time series model is used to predict future load distribution. The model structure is:

[0102] Input: The current load distribution is , the heat source distribution is and the ambient temperature is ,

[0103] Output: Future The load distribution after time is ;

[0104] Construct a prediction model based on LSTM long short-term memory network:

[0105] ,

[0106] in, It is an LSTM network, which is used to capture nonlinear relationships in time series. is the load distribution matrix at the current moment, is the heat source distribution matrix at the current moment, is the ambient temperature;

[0107] According to the predicted load distribution, the cooling power is dynamically adjusted, and the adjustment formula is:

[0108] ,

[0109] in, For Time allocated to Cooling power at the grid point, For prediction Future loads at the grid point, For prediction Future heat sources at grid points, is the adjustment coefficient, which controls the weight of the load and heat source on the cooling power;

[0110] The optimization algorithm is used to dynamically adjust the refrigerant flow and refrigerant temperature. The adjustment formula is:

[0111] ,

[0112] in, To be allocated Refrigerant flow at the grid point, To be allocated Refrigerant temperature at the grid point, is the cooling path adjustment coefficient.

[0113] As a preferred solution of the method for constructing a fully prefabricated decoupled data center according to the present invention, the step of allocating power and cooling resources using a reinforcement learning algorithm is as follows:

[0114] Define the status as , including the current load distribution , Heat source distribution , Power supply and cooling power ,

[0115] Define the action as , indicating that at time Adjust the power supply and cooling power.

[0116] Define the reward as ,in, For energy consumption, is the temperature at each grid point, is the target temperature, is the load balance, reflecting the imbalance in the utilization of power and cooling resources. is the weight parameter;

[0117] Generate continuous actions based on the deep deterministic policy gradient DDPG algorithm , the goal is to maximize the cumulative reward ;

[0118] According to the RL strategy output, the power supply and cooling power distribution are adjusted in real time, and the new state Input into the RL model to continuously optimize the resource allocation strategy.

[0119] The beneficial effects of the present invention are as follows: the present invention breaks through the serial dependence of traditional civil engineering and equipment deployment through modular design of container components and factory prefabrication production, realizes the parallel advancement of civil engineering construction and module production, optimizes transportation plans and assembly sequences with the help of intelligent scheduling algorithms, and ensures that modules arrive efficiently on site according to the deployment schedule, thereby significantly shortening the construction period; in the module prefabrication and assembly stage, the present invention adopts a unified interface standard, and combines digital twin technology to verify the compatibility and matching of module interfaces through real-time data comparison to ensure the accuracy of the assembly process; effectively reduces the readjustment cost caused by on-site construction errors and improves the success rate of module assembly; in addition, in the operation stage of the data center, a deep learning algorithm is used to predict load dynamics and dynamically adjust the cooling power and path; at the same time, combined with reinforcement learning technology, the linkage resource allocation of power supply and distribution and refrigeration modules is realized, thereby maximizing the energy efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0121] Figure 1 This is a flow chart of the method for constructing a fully prefabricated decoupled data center in Example 1. DETAILED DESCRIPTION

[0122] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0123] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0124] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0125] Example 1, reference Figure 1 , is an embodiment of the present invention, which provides a method for constructing a fully prefabricated decoupled data center, including the following steps:

[0126] Step S1: All equipment in the data center is prefabricated in the factory using a container assembly model, and the deployment sequence of each module is generated using digital twin technology;

[0127] The equipment in the data center is divided into power supply and distribution modules, refrigeration modules, IT equipment modules and fire protection modules. The module interfaces are formulated according to the interface standards during the production process.

[0128] The method of using digital twin technology to generate the deployment order of each module is as follows:

[0129] Use digital twin technology to generate a virtual model of each module, simulate its operating characteristics, including load capacity, energy consumption indicators, heat generation and interface matching, verify the compatibility between modules, generate the module deployment sequence, and coordinate the module production plan based on the deployment sequence;

[0130] The steps of using digital twin technology to generate a virtual model of each module and simulate its operating characteristics are as follows:

[0131] Collect the design parameters and preliminary operation data of the module, define the input variables of the virtual model, the module includes module type, module size and operation characteristic parameters,

[0132] The module type is , They represent power supply and distribution module, cooling module, IT equipment module and fire protection module respectively.

[0133] The module size is , are the length, width and height of the module respectively.

[0134] The operating characteristic parameters include the maximum load capacity of , the static energy consumption of the module is , the module dynamic energy consumption is and the heat generation efficiency is ;

[0135] Combined with the collected parameters, simulation algorithms are used to simulate the operating characteristics of the module, including load capacity, energy consumption indicators and heat generation.

[0136] To simulate the load capacity, the formula is:

[0137] ,

[0138] in, is the actual load capacity of the module, is the maximum design load of the module, is the safety factor,

[0139] Carry out energy consumption module, the formula is:

[0140] ,

[0141] in, For the module at time The total energy consumption, is the basic energy consumption of the module in static standby state, is the dynamic power consumption of the module when running the load, depending on the load level, is the simulation running time,

[0142] Based on the energy consumption simulation, the heat generated by the calculation module during operation is calculated as follows:

[0143] ,

[0144] in, For the module at time of heat generation, is the heat generation efficiency, which indicates the ratio of energy consumption converted into heat. For the module at time The total energy consumption,

[0145] The steps of verifying the compatibility between modules are:

[0146] Define the interface parameters for each module:

[0147] The power interface standard is ,in Indicates voltage, represents the current,

[0148] Cooling interface standard is ,in Indicates the refrigerant flow rate, Indicates the refrigerant temperature.

[0149] The signal interface standard is , indicating the communication protocol of the module;

[0150] Check Module With module Interface compatibility,

[0151] like, and and ,but ,otherwise ,in, For modules and modules Interface compatibility flag, 1 means compatible, 0 means incompatible, For modules Power, cooling and signal interface parameters;

[0152] The steps of generating the module deployment sequence are:

[0153] According to the simulation results, the module's load capacity, energy consumption, heat generation, and interface compatibility are considered, and a priority sorting method is used to generate a deployment order. The priority sorting formula is:

[0154] ,

[0155] in, For modules The deployment priority score of is a weight parameter used to balance the impact of load capacity, energy consumption and heat on priority, satisfying , is the load capacity of the module, is the total energy consumption of the module, For the heat generation of the modules, the modules are sorted from high to low according to the priority value to generate the module deployment order;

[0156] Specifically, through the above steps, the virtual modeling of the module, the operation characteristic simulation, the interface compatibility verification are completed, and the module deployment sequence is generated.

[0157] Step S2, civil construction is carried out simultaneously with module prefabrication production. Based on the deployment sequence, an intelligent scheduling algorithm is used to adjust the module transportation schedule, determine the module deployment schedule, and pre-embed the interface;

[0158] During the civil construction process, the interfaces required for the modules are pre-buried in advance to determine the module location planning. The pre-buried interfaces include power supply channels, cooling pipes, clean energy access points, energy storage systems, and signal connection points. The pre-buried interfaces match the interface standards.

[0159] The steps of adjusting the module transportation schedule based on the deployment sequence and determining the module deployment schedule by using an intelligent scheduling algorithm are as follows:

[0160] Model the transport and deployment of modules and define variables and constraints, including module attributes, vehicle transport characteristics, and time variables.

[0161] Module attributes include: Modules , module Transportation distance , module Loading time And modules On-site deployment time ,

[0162] Vehicle transport characteristics include: Transport vehicles ,vehicle Loading capacity and vehicles Average transportation speed ,

[0163] Time variables include: Module Start time of transportation , module Time of arrival at the scene And modules Time when deployment is completed ;

[0164] Calculate the module's arrival time and deployment schedule based on the module's transportation distance and the vehicle's transportation characteristics,

[0165] Modules The formula for calculating the arrival time is:

[0166] ,

[0167] in, For modules Arrival time, For modules The start time of transportation, For modules The loading time, For modules The transport distance, For vehicles The average transport speed,

[0168] The calculation formula for module deployment completion time is:

[0169] ,

[0170] in, For modules Deployment completion time, For modules field deployment time;

[0171] The steps of using the intelligent scheduling algorithm to adjust the module transportation schedule are:

[0172] Genetic algorithm is used for transportation scheduling, and the goal is to minimize the overall deployment time, which is expressed as:

[0173] ,

[0174] in, The total time for all module deployments to complete, For modules The deployment completion time,

[0175] The defined constraints include vehicle capacity constraints, module deployment order constraints, and field deployment order constraints.

[0176] Vehicle capacity limits are:

[0177] ,

[0178] in, Assigned to vehicles A collection of modules,

[0179] The module deployment order constraint is: if the module Dependency modules , then it must satisfy ,

[0180] The on-site deployment sequence constraint is: the maximum number of modules deployed in parallel on-site in the same time period Limited, ,in, Representation Module At the moment Deployment is in progress;

[0181] Each chromosome of the genetic algorithm is an arrangement of the transport and deployment order of a module.

[0182] The fitness function is: ,in For fitness, maximize fitness ,

[0183] Cross-operation adjusts the transportation allocation of modules, and mutation operation changes the deployment order of modules.

[0184] Under the constraints, the module transportation and deployment schedule is gradually optimized through multiple generations of evolution to obtain the final module transportation and deployment schedule, including:

[0185] The start time of shipping each module , Arrival time and time to complete deployment , combined with the interface pre-embedded plan for on-site construction, coordinate the transportation and assembly of modules according to the schedule;

[0186] Specifically, through intelligent scheduling algorithms, the module transportation schedule and deployment timetable are optimized based on the module transportation distance, vehicle capacity and on-site deployment conditions. The final output schedule ensures that the modules arrive on site in order and are assembled efficiently.

[0187] Step S3: After the modules are prefabricated in the factory, they are transported to the site in order for assembly according to the deployment schedule after the civil construction is completed. After the assembly is completed, the modules are connected in combination with the embedded interfaces to start the global joint debugging work;

[0188] During the on-site assembly process, the interface data assembled on-site is compared with the digital twin model data of each module to detect assembly matching and compatibility;

[0189] The step of comparing the interface data assembled on site with the digital twin model data of each module to detect assembly matching and compatibility is:

[0190] Collect actual parameters of the field interface through sensors and detection equipment, and define interface properties, including:

[0191] Power interface properties: actual voltage and the actual current ,

[0192] Cooling interface properties: The actual refrigerant flow is and actual refrigerant temperature ,

[0193] Signal interface properties: communication protocol ;

[0194] Extract the standard interface properties of the corresponding modules from the digital twin model, including:

[0195] Power interface standard: Model voltage and model current ,

[0196] Cooling interface standard: Model refrigerant flow and model refrigerant temperature ,

[0197] Signal interface standard: Model communication protocol is ;

[0198] According to the comparison between the actual power interface parameters and the model data, the matching degree of the interface voltage and current is calculated as follows:

[0199] like and ,but ,otherwise ,in, is the allowable voltage error range, is the allowable current error range;

[0200] Compare the flow rate and temperature of the cooling interface and calculate the matching degree. The calculation method is:

[0201] like and ,but ,otherwise ,in, is the allowable refrigerant flow error range, is the allowable refrigerant temperature error range;

[0202] Compare the communication protocols of the signal interfaces and check compatibility. The calculation method is:

[0203] like ,but ,otherwise ;

[0204] The weighted method is used to comprehensively evaluate the matching score, and the calculation method is as follows:

[0205] ,in, For modules The matching score of is the weight of matching detection, satisfying ,

[0206] The compatibility of the modules is determined based on the comprehensive score. The judgment method is as follows:

[0207] like ,but ,otherwise ,in, For modules The compatibility determination result is 1 for compatible and 0 for incompatible. The minimum score threshold for compatibility; if a module fails the matching or compatibility test , then take corresponding measures according to the mismatched items;

[0208] Specifically, through multi-dimensional comparison of on-site interface data and digital twin model data, including power, cooling and signals, and combined with comprehensive matching scores and compatibility judgments, it is ensured that the modules can operate efficiently and collaboratively after assembly. If mismatches are found, targeted adjustment measures are taken to ensure the overall stability and reliability of the system.

[0209] Step S4, using the initial parameters of the global joint debugging phase as input, deploying distributed sensors to monitor heat source distribution and load changes, using a deep learning algorithm to predict load changes, and adjusting the cooling power and path based on load changes;

[0210] In step S4, the power supply and distribution module is linked with the refrigeration module to allocate power and cooling resources using a reinforcement learning algorithm;

[0211] The pre-buried clean energy access point and energy storage system in step S2 are put into use, and the photovoltaic power generation and phase change energy storage module are combined to realize peak load shaving and valley filling of the grid load, thereby improving energy efficiency;

[0212] The steps of using a deep learning algorithm to predict load changes and adjusting the cooling power and path based on the load changes are:

[0213] The initial operating parameters of the system are obtained from the joint debugging output of step S3, including:

[0214] Heat source distribution matrix , indicating that the data center Initial heat distribution at the grid points,

[0215] Load Distribution Matrix ,express Initial calculation loads at grid points;

[0216] Deploy sensor networks to collect data in real time: The current heat source distribution is: , the current load distribution is and the ambient temperature is ,

[0217] The time series model is used to predict future load distribution. The model structure is:

[0218] Input: The current load distribution is , the heat source distribution is and the ambient temperature is ,

[0219] Output: Future The load distribution after time is ;

[0220] Construct a prediction model based on LSTM long short-term memory network:

[0221] ,

[0222] in, It is an LSTM network, which is used to capture nonlinear relationships in time series. is the load distribution matrix at the current moment, is the heat source distribution matrix at the current moment, is the ambient temperature;

[0223] According to the predicted load distribution, the cooling power is dynamically adjusted, and the adjustment formula is:

[0224] ,

[0225] in, For Time allocated to Cooling power at the grid point, For prediction Future loads at the grid point, For prediction Future heat sources at grid points, is the adjustment coefficient, which controls the weight of the load and heat source on the cooling power;

[0226] The optimization algorithm is used to dynamically adjust the refrigerant flow and refrigerant temperature. The adjustment formula is:

[0227] ,

[0228] in, To be allocated Refrigerant flow at the grid point, To be allocated Refrigerant temperature at the grid point, is the cooling path adjustment coefficient;

[0229] The steps of allocating power and cooling resources using a reinforcement learning algorithm are:

[0230] Define the status as , including the current load distribution , Heat source distribution , Power supply and cooling power ,

[0231] Define the action as , indicating that at time Adjust the power supply and cooling power.

[0232] Define the reward as ,in, For energy consumption, is the temperature at each grid point, is the target temperature, is the load balance, reflecting the imbalance in the utilization of power and cooling resources. is the weight parameter;

[0233] Generate continuous actions based on the deep deterministic policy gradient DDPG algorithm , the goal is to maximize the cumulative reward ;

[0234] According to the RL strategy output, the power supply and cooling power distribution are adjusted in real time, and the new state Input into the RL model to continuously optimize the resource allocation strategy;

[0235] Specifically, deep learning algorithms are used to predict load changes and dynamically adjust cooling power and paths. At the same time, reinforcement learning is used to achieve the coordinated allocation of resources between power supply and distribution and refrigeration modules to maximize energy efficiency and ensure stable and efficient operation of the data center.

[0236] Step S5, monitor the data of each module during the operation phase of the entire data center, continuously update the digital twin model in real time, and verify the newly added modules when new modules are added to verify the compatibility between the modules.

[0237] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a fully prefabricated decoupled data center, characterized by: include, Step S1: All equipment in the data center is prefabricated in the factory using a container assembly model, and the deployment sequence of each module is generated using digital twin technology; Step S2, civil construction is carried out, and the civil construction is carried out simultaneously with the module prefabrication production. Based on the deployment sequence, an intelligent scheduling algorithm is used to adjust the module transportation schedule, determine the module deployment schedule, and pre-embed the interface; Step S3: After the modules are prefabricated in the factory, they are transported to the site in order for assembly according to the deployment schedule after the civil construction is completed. After the assembly is completed, the modules are connected in combination with the embedded interfaces to start the global joint debugging work; Step S4, using the initial parameters of the global joint debugging phase as input, deploying distributed sensors to monitor heat source distribution and load changes, using a deep learning algorithm to predict load changes, and adjusting the cooling power and path based on load changes; Step S5, monitoring the data of each module during the operation phase of the entire data center, continuously updating the digital twin model in real time, and verifying the newly added modules when adding them to verify the compatibility between the modules; The equipment in the data center is divided into power supply and distribution modules, refrigeration modules, IT equipment modules and fire protection modules. The module interfaces are formulated according to the interface standards during the production process. The method of using digital twin technology to generate the deployment order of each module is as follows: Digital twin technology is used to generate a virtual model of each module, simulating its operating characteristics, including load capacity, energy consumption indicators, heat generation and interface matching, verifying the compatibility between modules, generating the module deployment sequence, and coordinating the module production plan based on the deployment sequence.

2. The method for constructing a fully prefabricated decoupled data center according to claim 1, characterized in that: During the civil construction process, the interfaces required for the modules are embedded in advance to determine the module location planning. The embedded interfaces include power supply channels, cooling pipes, clean energy access points, energy storage systems, and signal connection points. The embedded interfaces match the interface standards.

3. The method for constructing a fully prefabricated decoupled data center according to claim 2, characterized in that: During the on-site assembly process, the on-site assembled interface data is compared with the digital twin model data of each module to detect assembly matching and compatibility.

4. The method for constructing a fully prefabricated decoupled data center according to claim 3, characterized in that: In step S4, the power supply and distribution module is linked with the refrigeration module, and a reinforcement learning algorithm is used to allocate power and cooling resources.

5. The method for constructing a fully prefabricated decoupled data center according to claim 4, characterized in that: The steps of using digital twin technology to generate a virtual model of each module and simulate its operating characteristics are as follows: Collect the design parameters and preliminary operation data of the module, define the input variables of the virtual model, the module includes module type, module size and operation characteristic parameters, The module type is , They represent power supply and distribution module, cooling module, IT equipment module and fire protection module respectively. The module size is , are the length, width and height of the module respectively. The operating characteristic parameters include the maximum load capacity of , the static energy consumption of the module is , the module dynamic energy consumption is and the heat generation efficiency is ; Combined with the collected parameters, simulation algorithms are used to simulate the operating characteristics of the module, including load capacity, energy consumption indicators and heat generation. To simulate the load capacity, the formula is: ， in, is the actual load capacity of the module, is the maximum design load of the module, is the safety factor, Carry out energy consumption module, the formula is: ， in, For the module at time The total energy consumption, is the basic energy consumption of the module in static standby state, is the dynamic power consumption of the module when running the load, depending on the load level, is the simulation running time, Based on the energy consumption simulation, the heat generated by the calculation module during operation is calculated as follows: ， in, For the module at time of heat generation, is the heat generation efficiency, which indicates the ratio of energy consumption converted into heat. For the module at time The total energy consumption, The steps of verifying the compatibility between modules are: Define the interface parameters for each module: The power interface standard is ,in Indicates voltage, represents the current, Cooling interface standard is ,in Indicates the refrigerant flow rate, Indicates the refrigerant temperature. The signal interface standard is , indicating the communication protocol of the module; Check Module With module Interface compatibility, like, and and ,but ,otherwise ,in, For modules and modules Interface compatibility flag, 1 means compatible, 0 means incompatible, For modules Power, cooling and signal interface parameters; The steps of generating the module deployment sequence are: According to the simulation results, the module's load capacity, energy consumption, heat generation, and interface compatibility are considered, and a priority sorting method is used to generate a deployment order. The priority sorting formula is: ， in, For modules The deployment priority score of is a weight parameter used to balance the impact of load capacity, energy consumption and heat on priority, satisfying , is the load capacity of the module, is the total energy consumption of the module, For the heat generation of the modules, the modules are sorted from high to low according to the priority value to generate the module deployment order.

6. The method for constructing a fully prefabricated decoupled data center according to claim 5, characterized in that: The steps of adjusting the module transportation schedule based on the deployment sequence and determining the module deployment schedule by using an intelligent scheduling algorithm are as follows: Model the transport and deployment of modules and define variables and constraints, including module attributes, vehicle transport characteristics, and time variables. Module attributes include: Modules , module Transportation distance , module Loading time And modules On-site deployment time , Vehicle transport characteristics include: Transport vehicles ,vehicle Loading capacity and vehicles Average transportation speed , Time variables include: Module Start time of transportation , module Time of arrival at the scene And modules Time when deployment is completed ; Calculate the module's arrival time and deployment schedule based on the module's transportation distance and the vehicle's transportation characteristics, Modules The formula for calculating the arrival time is: ， in, For modules Arrival time, For modules The start time of transportation, For modules The loading time, For modules The transport distance, For vehicles The average transport speed, The calculation formula for module deployment completion time is: ， in, For modules Deployment completion time, For modules field deployment time; The steps of using the intelligent scheduling algorithm to adjust the module transportation schedule are: Genetic algorithm is used for transportation scheduling, and the goal is to minimize the overall deployment time, which is expressed as: ， in, The total time for all module deployments to complete, For modules The deployment completion time, The defined constraints include vehicle capacity constraints, module deployment order constraints, and field deployment order constraints. Vehicle capacity limits are: ， in, Assigned to vehicles A collection of modules, The module deployment order constraint is: if the module Dependency modules , then it must satisfy , The on-site deployment sequence constraint is: the maximum number of modules deployed in parallel on-site in the same time period Limited, ,in, Representation Module At the moment Deployment is in progress; Each chromosome of the genetic algorithm is an arrangement of the transport and deployment order of a module. The fitness function is: ,in For fitness, maximize fitness , Cross-operation adjusts the transportation allocation of modules, and mutation operation changes the deployment order of modules. Under the constraints, the module transportation and deployment schedule is gradually optimized through multiple generations of evolution to obtain the final module transportation and deployment schedule, including: The start time of shipping each module , Arrival time and time to complete deployment , combined with the interface pre-embedded plan for on-site construction, coordinate the transportation and assembly of modules according to the schedule.

7. The method for constructing a fully prefabricated decoupled data center according to claim 6, characterized in that: Compare the interface data of on-site assembly with the digital twin model data of each module to detect assembly matching and compatibility. Collect actual parameters of the field interface through sensors and detection equipment, and define interface properties, including: Power interface properties: actual voltage and the actual current , Cooling interface properties: The actual refrigerant flow is and actual refrigerant temperature , Signal interface properties: communication protocol ; Extract the standard interface properties of the corresponding modules from the digital twin model, including: Power interface standard: Model voltage and model current , Cooling interface standard: Model refrigerant flow and model refrigerant temperature , Signal interface standard: Model communication protocol is ; According to the comparison between the actual power interface parameters and the model data, the matching degree of the interface voltage and current is calculated as follows: like and ,but ,otherwise ,in, is the allowable voltage error range, is the allowable current error range; Compare the flow rate and temperature of the cooling interface and calculate the matching degree. The calculation method is: like and ,but ,otherwise ,in, is the allowable refrigerant flow error range, is the allowable refrigerant temperature error range; Compare the communication protocols of the signal interfaces and check compatibility. The calculation method is: like ,but ,otherwise ; The weighted method is used to comprehensively evaluate the matching score, and the calculation method is as follows: ,in, For modules The matching score of is the weight of matching detection, satisfying , The compatibility of the modules is determined based on the comprehensive score. The judgment method is as follows: like ,but ,otherwise ,in, For modules The compatibility determination result is 1 for compatible and 0 for incompatible. is the minimum score threshold for compatibility; If a module fails to match or compatibility test , then take corresponding measures according to the mismatched items.

8. The method for constructing a fully prefabricated decoupled data center according to claim 7, characterized in that: The steps of using a deep learning algorithm to predict load changes and adjusting the cooling power and path based on the load changes are: The initial operating parameters of the system are obtained from the joint debugging output of step S3, including: Heat source distribution matrix , indicating that the data center Initial heat distribution at the grid points, Load Distribution Matrix ,express Initial calculation loads at the grid points; Deploy sensor networks to collect real-time data: The current heat source distribution is: , the current load distribution is and the ambient temperature is ; The time series model is used to predict future load distribution. The model structure is: Input: The current load distribution is , the heat source distribution is and the ambient temperature is , Output: Future The load distribution after time is ; Construct a prediction model based on LSTM long short-term memory network: ， in, It is an LSTM network, which is used to capture nonlinear relationships in time series. is the load distribution matrix at the current moment, is the heat source distribution matrix at the current moment, is the ambient temperature; According to the predicted load distribution, the cooling power is dynamically adjusted, and the adjustment formula is: ， in, For Time allocated to Cooling power at the grid point, For prediction Future loads at the grid point, For prediction Future heat sources at grid points, is the adjustment coefficient, which controls the weight of the influence of load and heat source on the cooling power; The optimization algorithm is used to dynamically adjust the refrigerant flow and refrigerant temperature. The adjustment formula is: ， in, To be allocated Refrigerant flow at the grid point, To be allocated Refrigerant temperature at the grid point, is the cooling path adjustment coefficient.

9. The method for constructing a fully prefabricated decoupled data center according to claim 8, characterized in that: The steps of allocating power and cooling resources using a reinforcement learning algorithm are: Define the state as , including the current load distribution , Heat source distribution , Power supply and cooling power , Define the action as , indicating that at time Adjust the power supply and cooling power. Define the reward as ,in, For energy consumption, is the temperature at each grid point, is the target temperature, is the load balance, reflecting the imbalance in the utilization of power and cooling resources. is the weight parameter; Generate continuous actions based on the deep deterministic policy gradient DDPG algorithm , the goal is to maximize the cumulative reward ; According to the RL strategy output, the power supply and cooling power distribution are adjusted in real time, and the new state Input into the RL model to continuously optimize the resource allocation strategy.

Images