Method, system and device for finely adjusting and intelligently processing and storing machine room dynamic environment data in combination with large model

Abstract

The invention discloses an intelligent processing and storage method, system and device for machine room dynamic environment data combined with large model fine tuning, and the method comprises the steps: carrying out the grid division of the machine room dynamic environment data collected through a plurality of network transmission protocols through an intelligent grid segmentation method, and adjusting the size of a grid based on the use density of the grid; calculating the comprehensive score of each grid and the comprehensive score of each area to obtain a difference value between the comprehensive score of the grid and the comprehensive score of each area, and evaluating the performance state of the grid according to the difference value; and storing the processed data in a database, training a large model by using the processed data, performing first-stage fine adjustment and second-stage fine adjustment, storing the whole-process data, and performing regular backup. The system comprises a data acquisition and grid division module, a performance analysis module, a data caching module and a large model training and safety fine tuning module. According to the invention, high-efficiency collection, intelligent analysis and accurate control of machine room dynamic environment data are realized, and the intelligent level of energy efficiency management and the stability of system operation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of edge intelligence technology, specifically to a method, system, and device for intelligent processing and storage of data from a computer room's dynamic and environmental systems, combined with a large model for fine-tuning. Background Technology

[0002] With the rapid development of the internet and the digital economy, the energy consumption of data centers and communication equipment rooms has become increasingly prominent, becoming one of the major sources of global energy consumption. Especially in some regions, the air conditioning in data center base stations typically runs for nearly 16 hours a day, resulting in huge energy consumption. Currently, energy-saving goals are mainly achieved by manually adjusting the air conditioning temperature multiple times, which not only increases the workload of maintenance personnel but also has limited adjustment effects, making it difficult to achieve precise and intelligent control.

[0003] In existing technologies, although environmental systems (such as FSUs) have been deployed in data centers to collect environmental and equipment data and support simple remote control of equipment such as air conditioners, including starting / stopping and temperature setting, their automation level is low, lacking the ability to dynamically perceive and respond to actual needs such as local heat sources and equipment loads within the data center. Furthermore, traditional artificial intelligence methods suffer from insufficient accuracy and weak model generalization ability in equipment control, and large models face high computational pressure and response latency when deployed at the edge, making it difficult to meet real-time control requirements.

[0004] On the other hand, existing data acquisition relies on various short-range communication protocols and lacks a unified data fusion and intelligent partitioning mechanism, resulting in low data acquisition efficiency and uneven coverage, which further limits the improvement of energy efficiency optimization and control accuracy.

[0005] Therefore, there is an urgent need for a system and method that can fine-tune large models and process dynamic gridded data to improve the efficiency of data acquisition, processing accuracy and control intelligence of computer room environmental data. Summary of the Invention

[0006] The purpose of this invention is to provide a method for intelligent processing and storage of computer room dynamic and environmental data combined with large model fine-tuning, in order to solve problems such as insufficient control accuracy of equipment and high computational pressure on modules.

[0007] To achieve the above objectives, the technical solution provided by this invention is: a method for intelligent processing and storage of data from a computer room's environmental monitoring system, combining large-scale model fine-tuning, comprising the following steps: Step 1: Use the intelligent grid segmentation method to divide the environmental data of the computer room collected through network transmission protocols into grids; dynamically adjust the grid size of each grid based on the grid usage density; Step 2: Calculate the overall score of each grid cell and the overall score of the area, and obtain the difference between the overall score of each grid cell and the overall score of the area; evaluate the overall grid performance based on the difference between the overall score of the grid cells and the overall score of the area. Step 3: Store the data after mesh generation and mesh synthesis performance analysis into the database; Step 4: Train the large model using the database, and perform a first-stage fine-tuning of the large model based on ridge regression overfitting loss control, as well as a second-stage fine-tuning that incorporates a perceptual analysis safety mechanism; store the entire process data of the large model training, and perform regular data backup and recovery operations.

[0008] To optimize the above technical solution, the specific measures also include: In step 1, the network transmission protocols include: SNMP, Wi-Fi, ZigBee, Bluetooth, UWB, and RFID; The formula for calculating the density of the grid is:

[0009] in, Indicates the grid density used; Indicates user group density; This indicates that the transmission protocols within the mesh originate from the same channel density; The value represents the concurrent backhaul density; Wu, Wt, and Wc represent the weights used for user group density, transmission protocol density from the same channel, and concurrent backhaul density, respectively.

[0010] Furthermore, in step 1, the grid size of each grid is dynamically adjusted based on the grid density, and the adjusted grid size is calculated using the following formula:

[0011] in, This is the adjusted grid size; This is the initial grid size; It is the network usage density for each grid; It is the average network usage density of the entire backhaul data grid.

[0012] In step 2, the formula used to calculate the overall score for each grid cell is:

[0013] in, It is the overall score of the i-th grid. and The first Mean values ​​of SS-RSRP and SS-SINR for each grid cell These are the weights of SS-RSRP and SS-SINR, respectively.

[0014] In step 2, the formula used to calculate the comprehensive score of the area is:

[0015] in, It is the overall score of the area. It is the overall score of the i-th grid cell within the grid, where n is the number of grid cells in the area. This represents the geographic location weight of the i-th grid cell; The difference between the grid's overall score and the area's overall score = - .

[0016] Furthermore, if the difference between the overall score of the grid and the overall score of the area is negative, it is determined that the overall performance of the grid needs to be improved; if the difference between the overall score of the grid and the overall score of the area is zero, it is determined that the overall performance of the grid is at a balanced level; if the difference between the overall score of the grid and the overall score of the area is positive, it is determined that the wireless data integrity and overall performance of the grid are good.

[0017] In step 4, the one-stage fine-tuning based on ridge regression overfitting loss control specifically includes: The data in the database is clustered according to local regional characteristics to generate fine-tuned clusters. The overfitting loss of ridge regression is calculated, and its loss function is:

[0018]

[0019] Where X represents the input; y represents the predicted output; || represents the regularization operation; I represents the identity matrix; and θ is the fitting hyperparameter. is a weight constant; 'a' is the weight of the identity matrix; This means finding the value of θ given a fixed value for a.

[0020] In step 4, the two-stage fine-tuning that integrates the perception and analysis security mechanism specifically includes: Parse the associated path file to read the associated data and path of the current data; The fanotify component, deployed on the large model computing node, monitors the log file operation events corresponding to the large model in real time, and the perception component aggregates and processes the events. The decision component determines whether the large model has been compromised by a virus based on the aggregated event information, the number of file operations, and the consistency of query conditions. If the system is determined to have been compromised by a virus, the execution component will isolate or reset the large model's operations.

[0021] As another important technical solution, this invention also provides a data center environmental data integration and large-scale model fine-tuning intelligent processing and storage system, comprising: The data acquisition and grid division module is used to divide the environmental data of the computer room collected through network transmission protocols into grids using an intelligent grid segmentation method; and dynamically adjusts the grid size of each grid based on the grid usage density. The grid performance analysis and evaluation module is used to calculate the difference between the comprehensive score of each grid and the comprehensive score of the area by calculating the comprehensive score of each grid and the comprehensive score of the area; and to evaluate the overall grid performance based on the difference between the comprehensive score of the grid and the comprehensive score of the area. The large model fine-tuning training module is used to train large models using a database, and to perform one-stage fine-tuning of the large models based on ridge regression overfitting loss control, as well as two-stage fine-tuning that incorporates a perceptual analysis security mechanism. The data caching module is used to store the data that has undergone gridding and performance analysis to the database; it stores the entire process data in the training of large models and performs regular data backup and recovery operations.

[0022] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent processing and storage method for fine-tuning a large model of combined environmental data in a computer room, as described above.

[0023] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute the above-described intelligent processing and storage method for combining environmental data of a computer room with a large model for fine-tuning.

[0024] Compared with the prior art, the beneficial effects of the present invention are: This invention dynamically adjusts the grid size through dynamic grid division and grid density scoring mechanism. This not only ensures comprehensive coverage of different areas and types of environmental data in the computer room and reduces blind spots, but also reduces redundancy by using large grids in high-density areas and small grids in low-density areas to ensure accuracy. It also optimizes the acquisition layout of multi-protocol data to ensure comprehensive data coverage and efficient transmission, significantly improving data integrity and availability.

[0025] This invention employs a two-stage fine-tuning mechanism. The first-stage fine-tuning utilizes ridge regression to effectively control the risk of overfitting in specific data center scenarios, thereby improving the model's generalization ability. The second-stage fine-tuning further learns on complex data and incorporates a perception analysis security mechanism. This not only consolidates and deepens the learning results of the first stage but also corrects errors and reinforces difficult points in real time, making the final large model more accurate in capturing the features of data center environmental data and more precise and reliable in prediction and control decisions.

[0026] This invention, based on comprehensive analysis of environmental data from data centers and large-scale model inference, enables the system to accurately identify local hotspots and uneven equipment load within the data center. By combining the trained and optimized large-scale model with inference analysis, it can achieve precise on-demand and regional control of energy-consuming equipment such as air conditioners, significantly reducing energy consumption and improving electrical efficiency.

[0027] This invention monitors the model's operating status in real time through sensing components, enabling rapid detection and response to potential security threats such as virus intrusions. This greatly enhances the system's proactive defense capabilities and operational stability in complex network environments, ensuring the safety and reliability of control commands.

[0028] This invention not only stores the original environmental data, but also combines regular data backup and recovery mechanisms to ensure long-term data security and traceability, thereby improving the maintainability and lifecycle management capabilities of the entire system. Attached Figure Description

[0029] Figure 1 : A schematic diagram of the process of this invention. Detailed Implementation

[0030] The present invention will be further described in detail below through specific embodiments, but it should not be construed as limiting the scope of the subject matter of the present invention to the following embodiments. All technologies implemented based on the above content of the present invention fall within the scope of the present invention.

[0031] The following is a brief explanation of some of the terms related to this invention: Large-scale artificial intelligence models refer to machine learning models with extremely large parameters and complex computational structures. They are capable of processing massive amounts of data and completing various complex tasks, such as natural language processing and image recognition.

[0032] Large-scale model data extraction refers to the process of extracting and generating valuable information or data from massive datasets using large-scale deep learning models. This process relies on the powerful computing capabilities of large models and their ability to identify complex data patterns in order to achieve efficient and accurate data extraction and analysis.

[0033] Edge intelligence refers to a technology that applies artificial intelligence and data analytics to edge devices, enabling these devices to have a certain level of intelligence so that they can process, analyze, and make decisions locally without relying entirely on the cloud or a central server.

[0034] SS-RSRP refers to the received signal power of the synchronization signal and the reference signal of the physical broadcast channel block received by the user equipment from the base station; this signal is necessary for the user equipment to initially access the network and obtain system information.

[0035] S-SINR refers to the ratio of signal strength to interference noise within a symbol, that is, the ratio of reference signal power to interference noise power within the measurement bandwidth; where signal power mainly refers to the received power of the synchronization signal, while interference noise power includes the sum of non-serving cell interference, adjacent channel interference, and ambient noise power.

[0036] like Figure 1 As shown, this invention provides a method for intelligent processing and storage of data center environmental data combined with large-scale model fine-tuning, including the following steps: Step 1: Use the intelligent grid segmentation method to divide the environmental data of the computer room collected through the network transmission protocol into grids; dynamically adjust the grid size of each grid based on the grid usage density.

[0037] The intelligent grid segmentation method is based on the characteristics of six protocols: SNMP (device status acquisition), Wi-Fi (high-speed data transmission), ZigBee (low-power environment monitoring), Bluetooth (near-field device interaction), UWB (high-precision positioning), and RFID (asset identification), to construct an adaptive segmentation strategy.

[0038] The formula for calculating the mesh density is:

[0039] in, This indicates the mesh density; the more causal relationships there are, the denser the mesh. This indicates the density of user groups; the more users in the same user group within a grid, the denser the grid. This indicates the density of transmission protocols within a grid originating from the same channel; the more transmission protocols within a grid originating from the same channel, the denser the grid. This represents the concurrent return density; the higher the concurrent return density, the denser the mesh. u W t and W c These represent the weights used for user group density, transmission protocol density from the same channel, and concurrent backhaul density, respectively. These weights can be set according to the actual situation.

[0040] High-density areas are divided into larger grids, and low-density areas into smaller grids. This allows for finer grid division, ensuring more accurate and meaningful data within each grid. The grid size is dynamically adjusted based on density, and the adjusted grid size is calculated using the following formula:

[0041] in, This is the adjusted grid size; This is the initial grid size; It is the network usage density for each grid; It is the average network usage density of the entire backhaul data grid.

[0042] The data density of each protocol is calculated by integrating the CompositeDensity formula, which incorporates user groups, channel affiliation, and concurrent backhaul weights, to clarify the spatial distribution differences of data from different protocols. Then, the grid is dynamically adjusted according to the NewGridSize formula: the grid is enlarged for high-density data areas such as UWB to reduce redundant calculations, and the grid is reduced for low-density areas such as RFID to ensure acquisition accuracy, thus achieving a match between protocol characteristics and grid granularity.

[0043] Step 2: Calculate the overall score of each grid cell and the overall score of the area to obtain the difference between the overall score of each grid cell and the overall score of the area; evaluate the overall grid performance based on the difference between the overall score of the grid cells and the overall score of the area.

[0044] Based on various indicators of dynamic environmental data within the grid, such as temperature, humidity, and voltage, a weighted scoring method is used to calculate the comprehensive score of each grid. The weights can be adjusted according to actual needs to reflect the impact of different indicators on overall performance. The comprehensive score of each grid is calculated using the following formula:

[0045] in, It is the overall score of the i-th grid. and The first Mean values ​​of SS-RSRP and SS-SINR for each grid cell These are the weights for SS-RSRP and SS-SINR, respectively, which can be set according to the actual situation; For each grid consisting of multiple cells, its area comprehensive score is calculated, which is the average of the comprehensive scores of all cells within the grid, to reflect the overall performance level of the grid. The formula for calculating the area comprehensive score is as follows:

[0046] in, It is the overall score of the area. It is the overall score of the i-th grid cell within the grid, where n is the number of grid cells in the area. This represents the geographic location weight of the i-th grid cell, which can be set based on geographic location and regional importance.

[0047] The difference between the grid's overall score and the area's overall score = - .

[0048] In the intelligent processing of environmental data in the computer room, the calculation of the difference in the comprehensive score of the grid within the grid is the core of performance evaluation. This difference is directly used to determine the comprehensive performance of the grid and provides a basis for subsequent optimization.

[0049] If the difference between the grid's overall score and the area's overall score is negative, the grid's overall performance is considered to need improvement; if the difference is zero, the grid's overall performance is considered to be at a balanced level; if the difference is positive, the grid's wireless data integrity and overall performance are considered to be good.

[0050] Preferably, the geographic location weight is expressed as follows: ; in, It is the geographic location weight of the i-th grid cell. It is the first The distance from the center of one grid cell to the center of another grid cell can be calculated using latitude and longitude coordinates. It represents the regional importance of the i-th grid, which can be user-defined and is used to indicate the importance of a specific region to the grid cells in the overall grid.

[0051] Finally, the computer device forwards the gridded data to the edge intelligent nodes, which improves the efficiency of multi-protocol data aggregation and reduces the computing pressure on the cloud, providing a spatially correlated data foundation for accurate data retrieval from large models, while enhancing the targeting and effectiveness of data collection and analysis of environmental data in the data center. After the large model data retrieval process is completed, the conclusions of the completed large model are fine-tuned. Through the combination of these technologies, this patent achieves efficient generation of data retrieval question sets.

[0052] Step 3: Store the data after mesh generation and mesh comprehensive performance analysis into the database.

[0053] Step 4: Train the large model using the database, and perform a first-stage fine-tuning of the large model based on ridge regression overfitting loss control, as well as a second-stage fine-tuning that incorporates a perceptual analysis safety mechanism; store the entire process data of the large model training, and perform regular data backup and recovery operations.

[0054] The one-stage fine-tuning based on ridge regression overfitting loss control is as follows: The data in the database is categorized by structure, dimension, dimension value, and common metrics. Data with the same metrics are grouped into fine-tuned clusters. Ridge regression overfitting loss is then calculated, and its loss function is:

[0055] The formula for preventing overfitting is:

[0056] Where X represents the input; y represents the predicted output; || represents the regularization operation; I represents the identity matrix; and θ is the fitting hyperparameter. is a weight constant; 'a' is the weight of the identity matrix; This means finding the value of θ given a fixed value for a.

[0057] The two-phase fine-tuning, which integrates perception and analysis security mechanisms, is as follows: Building upon the initial learning of simple data in the first stage of the large model, training on complex data is then implemented. To prevent forgetting from the first stage, a small amount of random knowledge is retained to reduce the forgetting problem. Furthermore, the challenging data from the first stage has been reinforced, and errors in the first stage have been corrected.

[0058] Parse the associated path file to read the associated data and path of the current data; By deploying the fanotify component on the large model computing node, log file operation events corresponding to the large model are monitored in real time. If a file operation change event occurs, the file operation event is sent to the perception component. The perception component aggregates the events and aggregates the log file operations corresponding to each parameter of the large model computing. The perception component sends the aggregated event operations to the decision component. The decision component determines whether the large model has been invaded by a virus based on the event information, including the number of file operations that occurred within a period of time, and whether the file operation status corresponding to each query condition is consistent within the same period of time.

[0059] If the decision component determines that the large model log file has been compromised by a virus, the decision component will send a signal to the execution component, which will then isolate or reset the large model operation.

[0060] The two-stage fine-tuning and perception analysis path form a collaborative mechanism to ensure model optimization and operational safety. The first-stage fine-tuning clusters the scene data based on local region features, including structure, dimension, and dimensionality. After generating fine-tuned clusters, overfitting is controlled using a ridge regression overfitting loss formula, incorporating input X, output y, regularization, and the identity matrix I, thus completing simple data learning. The second stage builds upon this foundation to learn complex data, retaining a small amount of random knowledge to prevent forgetting, while reinforcing challenging data from the first stage and correcting erroneous data, thereby improving the model's adaptability to data from the data center's environmental dynamics.

[0061] The system stores all data from the entire training process of the large-scale model, including the data itself, related data, anomaly detection data, profiling analysis output tables, and historical records. It also performs regular data backup and recovery operations to prevent data loss or corruption. Regular data backups ensure data security and availability to handle various unforeseen circumstances.

[0062] Regular data backup and recovery operations are performed, with critical data being backed up periodically to ensure rapid data recovery in case of system failure or unexpected events, thus guaranteeing system continuity and data integrity. The data storage module is closely integrated with grid partitioning and data fine-tuning, forming a crucial link in achieving a closed-loop intelligent processing system for data center environmental data. It not only provides reliable data support and ensures the traceability of the entire intelligent processing flow, but also supports the system's retrospective analysis of historical data through historical record functions. This transforms data center environmental management from passive response to proactive prediction and optimization, significantly improving the intelligence level and decision-making accuracy of data center energy consumption management. In another embodiment of the present invention, a data center environmental data combination large model fine-tuning intelligent processing and storage system is also provided, comprising: The data acquisition and grid division module is used to divide the environmental data of the computer room collected through network transmission protocols into grids using an intelligent grid segmentation method; and dynamically adjusts the grid size of each grid based on the grid usage density. The grid performance analysis and evaluation module is used to calculate the difference between the comprehensive score of each grid and the comprehensive score of the area by calculating the comprehensive score of each grid and the comprehensive score of the area; and to evaluate the overall grid performance based on the difference between the comprehensive score of the grid and the comprehensive score of the area. The large model fine-tuning training module is used to train large models using a database, and to perform one-stage fine-tuning of the large models based on ridge regression overfitting loss control, as well as two-stage fine-tuning that incorporates a perceptual analysis security mechanism. The data caching module is used to store the data that has undergone gridding and performance analysis to the database; it stores the entire process data in the training of large models and performs regular data backup and recovery operations.

[0063] In another embodiment of the present invention, an electronic device is proposed, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described intelligent processing and storage method for fine-tuning a large model of combined environmental data from a computer room.

[0064] In another embodiment of the present invention, a computer-readable storage medium is provided, storing a computer program that enables a computer to execute the above-described intelligent processing and storage method for combining environmental data of a computer room with a large model for fine-tuning.

[0065] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent substitutions, and improvements made by those skilled in the art to the above embodiments without departing from the scope of the technical solution of the present invention, based on the technical essence of the present invention, shall still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for intelligent processing and storage of computer room environmental data combined with large-scale model fine-tuning, characterized in that, Includes the following steps: Step 1: Use the intelligent grid segmentation method to divide the environmental data of the computer room collected through network transmission protocols into grids; dynamically adjust the grid size of each grid based on the grid usage density; Step 2: Calculate the overall score of each grid cell and the overall score of the area, and obtain the difference between the overall score of each grid cell and the overall score of the area; evaluate the overall grid performance based on the difference between the overall score of the grid cells and the overall score of the area. Step 3: Store the data after mesh generation and mesh synthesis performance analysis into the database; Step 4: Train the large model using the database, and perform a first-stage fine-tuning of the large model based on ridge regression overfitting loss control, and a second-stage fine-tuning that incorporates a perceptual analysis security mechanism. Store all data from the entire training process of a large model and perform regular data backup and recovery operations.

2. The intelligent processing and storage method for combining environmental data from a computer room with large-scale model fine-tuning as described in claim 1, characterized in that: In step 1, the network transmission protocols include: SNMP, Wi-Fi, ZigBee, Bluetooth, UWB, and RFID; The formula for calculating the density of the grid is: in, Indicates the grid density used; Indicates user group density; This indicates that the transmission protocols within the mesh originate from the same channel density; The value represents the concurrent backhaul density; Wu, Wt, and Wc represent the weights used for user group density, transmission protocol density from the same channel, and concurrent backhaul density, respectively.

3. The intelligent processing and storage method for combining environmental data from a computer room with large-scale model fine-tuning as described in claim 1, characterized in that: In step 1, the grid size of each grid is dynamically adjusted based on the grid density, and the adjusted grid size is calculated using the following formula: in, This is the adjusted grid size; This is the initial grid size; It is the network usage density for each grid; It is the average network usage density of the entire backhaul data grid.

4. The intelligent processing and storage method for combining environmental data from a computer room with a large model for fine-tuning, as described in claim 1, is characterized in that: In step 2, the formula used to calculate the overall score for each grid cell is: in, It is the overall score of the i-th grid. and The first Mean values ​​of SS-RSRP and SS-SINR for each grid cell These are the weights of SS-RSRP and SS-SINR, respectively.

5. The intelligent processing and storage method for combining environmental data from a computer room with a large model for fine-tuning, as described in claim 4, is characterized in that: In step 2, the formula used to calculate the comprehensive score of the area is: in, It is the overall score of the area. It is the overall score of the i-th grid cell within the grid, where n is the number of grid cells in the area. This represents the geographic location weight of the i-th grid cell; The difference between the grid's overall score and the area's overall score = - .

6. The intelligent processing and storage method for combining environmental data from a computer room with large-scale model fine-tuning as described in claim 1, characterized in that: In step 4, the one-stage fine-tuning based on ridge regression overfitting loss control specifically includes: The data in the database is clustered according to local regional characteristics to generate fine-tuned clusters. The overfitting loss of ridge regression is calculated, and its loss function is: Where X represents the input; y represents the predicted output; || represents the regularization operation; I represents the identity matrix; and θ is the fitting hyperparameter. is a weight constant; 'a' is the weight of the identity matrix; This means finding the value of θ given a fixed value for a.

7. The intelligent processing and storage method for combining environmental data from a computer room with large-scale model fine-tuning as described in claim 1, characterized in that: In step 4, the two-stage fine-tuning that integrates the perception and analysis security mechanism specifically includes: Parse the associated path file to read the associated data and path of the current data; The fanotify component, deployed on the large model computing node, monitors the log file operation events corresponding to the large model in real time, and the perception component aggregates and processes the events. The decision component determines whether the large model has been compromised by a virus based on the aggregated event information, the number of file operations, and the consistency of query conditions. If the system is determined to have been compromised by a virus, the execution component will isolate or reset the large model's operations.

8. A data center environmental data integration and large-scale model fine-tuning intelligent processing and storage system, characterized in that, include: The data acquisition and grid division module is used to perform grid division on the environmental data of the computer room acquired through network transmission protocols using an intelligent grid division method; The grid size of each grid is dynamically adjusted based on the grid density. The grid performance analysis and evaluation module is used to calculate the difference between the comprehensive score of each grid and the comprehensive score of the area by calculating the comprehensive score of each grid and the comprehensive score of the area; and to evaluate the overall grid performance based on the difference between the comprehensive score of the grid and the comprehensive score of the area. The large model fine-tuning training module is used to train large models using a database, and to perform one-stage fine-tuning of the large models based on ridge regression overfitting loss control, as well as two-stage fine-tuning that incorporates a perceptual analysis security mechanism. The data caching module is used to store the data that has undergone gridding and performance analysis to the database; it stores the entire process data in the training of large models and performs regular data backup and recovery operations.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent processing and storage method for fine-tuning a large model of combined environmental data in a computer room, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program enables the computer to execute the intelligent processing and storage method for combining large-scale models of computer room dynamic and environmental data as described in any one of claims 1 to 7.

Images