Data center operation and maintenance suggestion generation method based on multi-modal input technology

Abstract

The invention discloses a data center operation and maintenance suggestion generation method based on a multi-modal input technology, and the method comprises the steps: carrying out the interaction with a user, and receiving the input; the drawing information and the sensor installation position are converted into semantic information capable of being understood by a large language model based on the multi-modal technology, and if the information is insufficient, recording is carried out; cleaning the data collected by the sensor and generating an abstract; constructing a first cue word, inputting the first cue word into a large language model, and generating temperature distribution and air supply condition analysis; inquiring adjustable and controllable equipment in the room and a control range of the adjustable and controllable equipment; in combination with the analysis and control information, constructing a second cue word, inputting the second cue word into the large language model again, and generating an operation and maintenance optimization suggestion; and finally, respectively sorting the room information with the successfully generated suggestions and the information of the failed rooms into structured files and feeding back the structured files to the user. According to the method, drawings, sensors and equipment control information can be fused, intelligent analysis and suggestion generation are performed through a large language model, and efficient, intelligent and low-threshold data center operation and maintenance support is realized.

Description

Technical Field

[0001] This invention belongs to the field of automatic control of building heating, ventilation and air conditioning systems, and particularly relates to a method for generating data center operation and maintenance suggestions based on multimodal input technology. Background Technology

[0002] With the rapid development of information technology, data centers, as core information infrastructure, are constantly expanding in scale and becoming increasingly reliant on environmental control systems, especially heating, ventilation, and air conditioning (HVAC) systems. Typical data center scenarios often consist of numerous large rooms with high ceilings, used to deploy high-density server racks. To ensure the stable operation of server equipment, strict temperature and humidity control must be maintained.

[0003] In such large-space environments, multiple sets of air supply equipment are typically deployed, such as ceiling fan coil units, floor air supply systems, and cold / hot aisle partition units. These air supply devices are flexibly distributed, numerous, and have complex control parameters; their control strategies directly affect cooling efficiency and energy consumption levels.

[0004] However, server deployments in data centers are typically dynamic and uneven. Servers in some areas may be under high load, generating more heat, while other areas may be idle or under low load, resulting in a significant non-uniform distribution of heat sources. This inconsistent spatial heat load makes it difficult for traditional control strategies based on the assumption of uniform airflow to achieve ideal regulation effects. This can lead to localized overheating and insufficient cooling, as well as overall energy waste.

[0005] Therefore, how to accurately analyze and intelligently control the air supply system of a data center based on the actual spatial heat source distribution and environmental conditions has become one of the key issues in the operation and maintenance of green data centers. Especially in complex scenarios with multiple air supply devices, uneven heat sources, and adjustable devices, leveraging advanced artificial intelligence technology to optimize strategies and generate control suggestions has significant engineering value and application prospects.

[0006] In recent years, with the continuous advancement of artificial intelligence technology, large language models have shown significant breakthroughs in the field of natural language processing. Representative models such as GPT (Generative Pretrained Transformer) and DeepSeek, trained on massive amounts of text and multi-source data, possess powerful contextual understanding and language generation capabilities. These models can not only parse complex natural language instructions but also handle multimodal inputs including charts, structured data, and even image content.

[0007] Compared to traditional rule engines or model algorithms, large language models possess stronger knowledge transfer and generalization capabilities, enabling semantic reasoning, anomaly detection, and strategy suggestion generation in scenarios lacking explicit rule definitions. Particularly in problems involving complex contexts, heterogeneous data fusion, and dynamic scene analysis, large language models can achieve human-level expert-level judgment and decision support through comprehensive modeling of natural language, structured parameters, and graphical information.

[0008] Therefore, a data center air supply strategy analysis system based on a large language model can not only more accurately understand the natural language commands of operations and maintenance personnel, but also combine drawing information, real-time sensor data, and equipment control boundaries to intelligently analyze temperature distribution and energy consumption, and provide actionable air supply optimization suggestions. This provides a new technical path for achieving intelligent and low-carbon operation and maintenance of data centers. Summary of the Invention

[0009] The purpose of this invention is to provide a data center operation and maintenance suggestion generation method based on multimodal input technology to address the needs of existing data center HVAC systems.

[0010] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution: a method for generating data center operation and maintenance suggestions based on multimodal input technology, comprising the following steps: S1. Interact with the user, accept user instructions and documents, including data center air conditioning system design drawings and sensor files; the sensor files include sensor installation locations and sensor detection data. S2. Based on multimodal technology, the data center air conditioning system design drawings and sensor installation locations corresponding to the target room (i.e., the room where the user needs to provide operation and maintenance suggestions) are converted into information that the large language model can understand; if the required information is insufficient, the room and the insufficient information are recorded. S3. Convert the real-time data detected by the sensors corresponding to the target room into information that the large language model can understand, including cleaning the data and generating semantic summaries of data trends; if the required information is insufficient, record the room and the insufficient information, and proceed to step S7. S4. Construct a first prompt word based on the information transformed in steps S2 and S3, input the first prompt word into the large language model, and generate an analysis of the temperature distribution and air supply of the data center air conditioning system. S5. The query system allows for the control of equipment and control range; S6. Based on the analysis of temperature distribution and air supply of the data center air conditioning system, as well as the control equipment and control range, construct a second prompt word, input the second prompt word into the large language model, and generate operation and maintenance suggestions; S7. Summarize the room information for which maintenance suggestions were successfully generated into a first file, and summarize the room information for which maintenance suggestions were not successfully generated into a second file. Feedback both files to the user.

[0011] Furthermore, the multimodal input technology includes matching the user-inputted file with a predefined semantic description using a text-image matching model to extract room structure, air supply equipment layout, and sensor location information; if the data center air conditioning system design drawings and sensor location information of the target room are missing or the matching score is lower than a preset threshold, then the required information is considered insufficient.

[0012] Furthermore, the sensor data includes room environmental monitoring parameters; the environmental monitoring parameters include temperature, humidity, wind speed, and CO2 concentration.

[0013] Furthermore, the semantic summary includes the parameter fluctuation trend and deviation analysis from the set value within each time period.

[0014] Furthermore, both the first and second prompt words are multi-turn dialogue contexts constructed using natural language, containing device parameters, spatial structure information, and user-expected goals.

[0015] Furthermore, the large language model is a large-scale pre-trained language model that supports multimodal input, and the model has the ability to extract related information from natural language, image descriptions and structured data.

[0016] Furthermore, the two feedback documents provided to the user are presented in the form of structured reports, including the identification number of each room, suggested content or reasons for failure, and suggested update time.

[0017] Furthermore, based on the feedback in the second document, the user modifies or supplements the relevant information and re-executes steps S1 to S7 until all target rooms have successfully generated operation and maintenance suggestions.

[0018] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-described method for generating data center operation and maintenance suggestions based on multimodal input technology.

[0019] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for generating data center operation and maintenance suggestions based on multimodal input technology.

[0020] Compared with the prior art, the beneficial effects of the present invention are: First, this invention utilizes a large language model to perform semantic analysis and interpretable reasoning on the air supply situation of data centers. It can combine the natural language commands of operation and maintenance personnel, equipment operating status and environmental data to generate analysis results with logical structure, thereby improving operation and maintenance efficiency, reducing reliance on personnel's professional knowledge and enhancing the level of intelligence. Second, through multimodal input technology, this invention can organically integrate various types of information—including drawings and images, structured table data, sensor time series data, spatial topology information, and text descriptions—and provide them to the large language model for processing. This enables the model to have a more comprehensive contextual basis when understanding the relationship between spatial structure, equipment distribution, and actual working conditions, thereby generating more accurate and actionable operation and maintenance suggestions. Third, this invention supports generating targeted control strategy suggestions based on the correspondence between the equipment control range and the heat load distribution, which can optimize the overall energy efficiency of the air supply system while avoiding local overcooling or overheating, thus helping to reduce system energy consumption and extend equipment life. Fourth, in the event of incomplete information or missing data, this invention has the ability to record, identify and trace back, accurately mark the rooms that cannot generate suggestions and the reasons therefor, and provide feedback to the user in a structured form to assist in subsequent completion or manual intervention, thereby improving the robustness and traceability of the system. Fifth, the prompt word construction mechanism proposed in this invention can convert drawing structure, sensor trends and control range information into natural language prompts, adapt to the native input method of large language models, avoid complex rule templates or interface development, and has good versatility and scalability. In summary, this invention provides an efficient, intelligent, and interpretable method for generating operation and maintenance suggestions for data center HVAC systems by integrating multimodal input and large language model reasoning capabilities, and has broad engineering application prospects and promotional value. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a data center operation and maintenance suggestion generation method based on multimodal input technology, as provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of an electronic device structure provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0024] This invention aims to propose an innovative method for generating intelligent operation and maintenance suggestions for data center air conditioning systems. This method combines the powerful understanding and generation capabilities of large language models to achieve rapid generation of multimodal data processing summaries and operation and maintenance suggestions, enabling precise control of data center air conditioning systems. This intelligent generation mechanism significantly improves the system's energy efficiency and flexibility, thereby promoting the development of data center air conditioning systems towards intelligence and energy conservation.

[0025] This invention aims to rapidly generate operation and maintenance suggestions for data center air conditioning systems through multimodal input technology and large language models, thereby improving the system's automation level and operational efficiency while ensuring the energy efficiency of the control strategy. Figure 1 As shown, in a preferred embodiment of the present invention, the data center operation and maintenance suggestion generation method based on multimodal input technology includes the following steps S1 to S7. The specific implementation process of each step is described in detail below.

[0026] S1. Interact with the user, accept user instructions and files, including data center air conditioning system design drawings and sensor files; the sensor files include specific information such as sensor installation location, sensor detectable data types, range, and communication methods; It should be noted that in step S1 and subsequent steps of using a large language model in this invention, the method of this invention is not limited to using a specific large language model. It can adapt to and integrate multiple large language models, including but not limited to open-source and commercial models, to improve the adaptability and accuracy of the method. In this embodiment, in step S1 and subsequent steps of using a large language model, the deepseek-v3 model is used as the large language model, and the transmission process of questions and responses is realized by calling the API of the corresponding model. deepseek-v3 is a general-purpose language model that, through training on a large amount of text data, can understand and generate natural language, primarily in Chinese and English usage environments. The deepseek-v3 model can perform deep analysis of user questions, identifying key variables and information needs in the questions. Specifically, by analyzing the semantic and syntactic structure of user questions, the deepseek-v3 model can determine the topic and intent of the questions, and thus identify key variables related to the questions. At the same time, the deepseek-v3 model can also understand the user's information needs based on the context of the questions, thereby providing more accurate and relevant answers.

[0027] It should be noted that in step S1 of the present invention, the large language model includes an input module, which is capable of receiving various forms of file input and has the ability to convert file information into text.

[0028] In this embodiment, the input module interacts with the user through an interface, receiving control commands and processing their formatting in a timely manner. Specifically, the control commands refer to instructions involving the control of one or more HVAC subsystems, such as instructions for controlling multiple devices within a single room. The input module can handle various types of input, including text, images, and file input.

[0029] It should be noted that in step S1 of this invention, the input module can receive various forms of file input and has the ability to convert file information into text. The input module not only receives various types of files (such as data center air conditioning system design drawings, sensor files, etc.), but also can process heterogeneous data within these files. Whether it's an image, table, or text file, the input module can interact with the user through a preset interface, thereby ensuring that the received file can be effectively processed. During this process, the input module identifies the file type and adopts an appropriate parsing method based on the specific structure of the file.

[0030] In this embodiment, the input module interacts with the user through an interface, receiving control commands and processing their formatting in a timely manner. Specifically, the control commands refer to instructions involving the control of one or more data center air conditioning subsystems, such as instructions for controlling multiple devices within a single room. The input module can handle various types of input, including text and file input.

[0031] It should be noted that in step S1 of this invention, the input module is not only responsible for receiving user control commands, but also for parsing these commands and determining their feasibility for execution. The system uses a large language model to perform semantic understanding of the user-input commands and determine whether they are valid control commands. For example, for some mathematical problems or inputs unrelated to data center operations (such as "1+1=2?"), the system will determine that they do not conform to the specifications of control commands and will immediately provide feedback to the user, thereby avoiding the execution of invalid commands, saving system resources, and improving overall efficiency.

[0032] It should be noted that in step S1 of this invention, the system provides real-time feedback for each step. If the input module cannot correctly parse a file or the control command does not conform to the preset specifications, the system will immediately provide feedback to the user regarding the problem through the interface. This feedback mechanism not only ensures the efficient operation of the system but also helps users correct erroneous input in a timely manner, improving the user experience.

[0033] S2. For the target room (i.e. the room for which operation and maintenance suggestions are required), based on multimodal technology, the data center air conditioning system design drawings and sensor (including temperature sensor) installation locations are converted into information that the large language model can understand; if the required information is insufficient, the room and the insufficient information are recorded, and the process proceeds to step S7. It should be noted that in step S2 of this invention, the CLIP (Contrastive Language-Image Pretraining) model is used to analyze the drawing images. First, the original CAD drawings or PDF images are standardized, including image format conversion, resolution adjustment, and drawing segmentation. Then, each drawing image block is input into the CLIP image encoder module. Simultaneously, several predefined semantic description texts, such as "There is an air vent in the room," "A temperature sensor is installed in this area," and "No return air equipment," are prepared and input into the CLIP text encoder module. By calculating the cosine similarity between the image encoding and the text encoding, the matching items with higher scores are identified, thus determining the location and number of air vents and the sensor installation areas in each room of the drawing. The recognition results are then converted into structured natural language descriptions. This descriptive information serves as a key input for constructing subsequent prompts, assisting the large language model in subsequent analysis.

[0034] It should be noted that in step S2 of this invention, if it is found during the analysis that a room lacks complete drawing information and sensor installation location information, or the CLIP matching score is lower than the set threshold (e.g., the threshold is set to 0.24), the room will be marked as "insufficient information", and its room number and missing items (e.g., missing sensor location, air supply equipment, return air equipment, or one type or one item is not identified) will be recorded, providing a basis for subsequent maintenance personnel to supplement information or for manual intervention.

[0035] S3. Convert the real-time data detected by the sensors (including temperature sensors) into information that the large language model can understand, including cleaning the data and generating semantic summaries of the data trends; if the required information is insufficient (if the data detected by the sensors is insufficient within a preset time range, it is considered that the information is insufficient; for example, if the temperature data detected by the temperature sensor is insufficient in the last hour), record the room and the insufficient information, and proceed to step S7. It should be noted that in step S3 of the present invention, the collected raw sensor data is cleaned, including timestamp alignment, missing value filling, outlier removal, and unit standardization.

[0036] It should be noted that in step S3 of the present invention, the method for generating semantic summaries of data trends is as follows: statistical analysis and window segmentation calculation are performed on sensor data to extract representative data change patterns, such as temperature rise / fall trends, daytime peak values, fluctuation ranges, etc., and semantic description content is generated. This semantic summary text will be used as input together with the drawing recognition results to form complete prompt words, which will be used by the subsequent large language model for comprehensive analysis of temperature distribution and air supply.

[0037] S4. Construct the first prompt word, input the first prompt word into the large language model, and generate an analysis of the temperature distribution and air supply of the data center air conditioning system; It should be noted that in step S4 of this invention, the construction of the first prompt word includes spatial structure information fusion: organizing the information identified in step S2, such as the location, number, direction, air supply type (top supply, floor supply), presence of return air vents, and sensor distribution location, in a natural language manner; environmental state summary embedding: introducing the semantic sensor trend summary generated in step S3, enabling the language model to perceive the current working condition; analysis task instruction setting (such as analyzing whether the temperature and humidity distribution is uniform, and if not, which device caused it): clarifying the prompt word target through guided natural language description, prompting the language model to perform "analytical" output.

[0038] S5. The query system can control the equipment and the control range; It should be noted that in step S5 of this invention, it is necessary to query all adjustable devices in the data center air conditioning system of the target room, as well as the corresponding control parameters and adjustment ranges of each device, to support the generation of subsequent operation and maintenance suggestions. The query process includes: device location: matching all air conditioning devices whose physical locations fall within the range of the room based on the room number or spatial coordinates in the device database; control capability identification: reading the adjustable parameters of each device, such as air volume, supply air temperature, damper angle, and operating status; adjustment range extraction: for each control parameter, reading its supported adjustment range and unit. The query results are standardized and organized into structured data or natural language descriptions, which serve as input for the subsequent construction of prompt words.

[0039] S6. Based on the analysis of temperature distribution and air supply of the data center air conditioning system, as well as the control equipment and control range, construct a second prompt word, input the second prompt word into the large language model, and generate operation and maintenance suggestions; It should be noted that in step S6 of this invention, constructing the second prompt includes: embedding analysis results: introducing the analysis content of the large language model on the current operating status of the room air supply system, including descriptions of existing problems and speculations on possible causes; embedding equipment and control capability context: providing detailed information on currently adjustable equipment, including equipment number, control parameters, current set values, and adjustment ranges; defining task target instructions: setting target-oriented semantic instructions, clarifying what type of suggestion content the large language model is expected to generate. For example: Please tell me whether the current temperature and humidity distribution is uniform, whether it is caused by a certain device, and among the currently adjustable devices, which device can I adjust to which specific value to improve the current situation? S7. Summarize the room information for which maintenance suggestions were successfully generated into a first file, and summarize the room information for which maintenance suggestions were not successfully generated into a second file. Provide both files to the user.

[0040] It should be noted that in step S7 of this invention, for all scenarios where operation and maintenance suggestions were successfully generated, the system will organize and summarize them into a first file. This file will include for each scenario where operation and maintenance suggestions were successfully generated, the degree of inquiries and air supply situation analysis, and the final operation and maintenance suggestions. This file is not only easy to archive, but can also serve as a reference for operation and maintenance personnel to take subsequent measures.

[0041] It should be noted that in step S7 of this invention, for scenarios where control code was not successfully generated, the relevant information needs to be recorded in a second file. This file should contain detailed information about the ungenerated scenario, including the scenario number and missing key information. This file will help users understand why maintenance suggestions could not be generated for a specific scenario, and modify or supplement relevant information based on feedback, thereby improving the generation effect.

[0042] To better demonstrate the specific implementation and technical effects of the present invention, the data center operation and maintenance suggestion generation method based on multimodal input technology shown in steps S1 to S7 of the above preferred implementation is applied to a specific example.

[0043] Example Room 101 of a data center has 8 sets of data racks, each with 20 spaces for storing servers. Only the first and fifth sets of data racks are in normal operation.

[0044] The user's control command was: "Provide maintenance suggestions for the air conditioning system in room 101." and the corresponding file was uploaded.

[0045] Step 1. Determine that the user's command is related to air conditioning operation and maintenance, and that the controlled object is room 101.

[0046] Step 2. Analyze the system design documents to confirm the location of each unit, and that each unit corresponds to one air supply terminal. Confirm that temperature sensors 1-8 correspond to data cabinets 1-8.

[0047] Step 3. The system analyzes and summarizes the data from room 101, outputting the following text: "Temperature sensor 1 maintained a temperature of approximately 28°C for 1 hour."; "Temperature sensor 2 maintained a temperature of approximately 24°C for 1 hour."; ..., "Temperature sensor 5 maintained a temperature of approximately 28°C for 1 hour." Step 4. Generate corresponding prompts (including the output text from Step 3, the set target temperature, and maintaining the temperature distribution as much as possible), and analyze them using a large language model to obtain the following results: "The temperature of the first and fifth data cabinets is higher, around 28℃, while the temperature of the remaining data cabinets is lower, around 24℃. The uneven temperature distribution may be due to excessive load on the first and fifth data cabinets, or excessively high air supply temperature and insufficient air supply volume; the low temperature of the remaining data cabinets may be due to insufficient load, or excessively low air supply temperature and excessively high air supply volume." The target temperature in this embodiment is 26°C.

[0048] Step 5. Query the system's adjustable devices and control range to obtain information on adjustable devices and data ranges, for example: Fan coil unit FCU_101_01: Current air volume is 350 m³ / h 3 / h, adjustable range is 200–400 m 3 / h; Water valve WAV_101_01, current opening degree 60%, adjustable range 0–100%. Water valve WAV_101_02, current opening degree 60%, adjustable range 0–100%. Step 6. Based on the conclusions of Step 4 and the information from Step 5, construct prompt words and output suggestions from the large language model: "It is recommended to maintain the airflow of fan coil unit FCU_101_01 at 350 m³ / h." 3 The water valve opening is adjusted to control the distribution of cooling capacity, with the h value remaining constant.

[0049] Increase the opening of water valves WAV_101_01 and WAV_101_05 from the current 60% to 80% to enhance the cold water flow and improve cooling capacity, thereby effectively reducing the temperature around the first and fifth data cabinets.

[0050] It is also recommended that in the 2-4 and 6-8 idle server rack areas, the air valve opening be appropriately closed or reduced to below 30% to avoid ineffective air supply and energy waste.

[0051] If the system has zoned air supply capabilities, it is recommended to dynamically start and stop the air supply terminals based on the rack's operating status to optimize regional energy efficiency. Step 7. Generate the corresponding operation and maintenance information file, for example: Successfully generated operation and maintenance suggestion file: Basic Information 1. Room structure and operational status The room contains 8 data racks, each with space for 20 servers. Current running status: The first and fifth racks are operating normally. The remaining 6 racks are either empty or not powered on. Air supply terminal configuration: Each rack area is equipped with one air supply terminal, for a total of 8 terminal devices. Sensor configuration: Temperature sensor numbers: T1–T8, corresponding to racks 1–8 respectively. 2. Temperature and Operational Data Analysis Conclusion: The temperatures of the first and fifth data cabinets were relatively high, around 28℃, while the temperatures of the remaining data cabinets were relatively low, around 24℃. This uneven temperature distribution may be due to excessive load on the first and fifth data cabinets, or excessively high supply air temperature with insufficient airflow. The excessively low temperatures of the remaining data cabinets may be due to insufficient load, or excessively low supply air temperature with excessive airflow. 3. Operation and Maintenance Recommendations "It is recommended to maintain the airflow of fan coil unit FCU_101_01 at 350 m³ / h."3 The water valve opening is adjusted to control the distribution of cooling capacity, with the h value remaining constant.

[0052] Increase the opening of water valves WAV_101_01 and WAV_101_05 from the current 60% to 80% to enhance the cold water flow and improve cooling capacity, thereby effectively reducing the temperature around the first and fifth data cabinets.

[0053] It is also recommended that in the 2-4 and 6-8 idle server rack areas, the air valve opening be appropriately closed or reduced to below 30% to avoid ineffective air supply and energy waste.

[0054] If the system has zoned air supply capabilities, it is recommended to dynamically start and stop the air supply terminals based on the rack's operating status to optimize regional energy efficiency. Unable to generate maintenance suggestion file: Basic Information Error type: Drawing information missing Explanation of the reason for failure: The air conditioning system design drawings for room 102 have not been uploaded. Unable to identify the location and configuration of the air supply terminal; Recommended action: Please confirm whether the drawing file contains the layout of this room; Please upload the design drawings (DWG / PDF format) for room 102 and resubmit the request.

[0055] In summary, this invention provides a data center operation and maintenance suggestion generation method based on multimodal input technology. It can comprehensively process multi-source heterogeneous information, including HVAC system design drawings, sensor monitoring data, and equipment control parameters. By introducing a text-image alignment model and a large language model, it performs in-depth understanding and reasoning analysis of the data center room environment status and air supply system configuration, thereby automatically generating operable and energy-saving operation and maintenance optimization suggestions. This method not only significantly improves operation and maintenance efficiency and intelligence, and reduces reliance on professional knowledge, but also enhances system stability and traceability through information missing detection and group feedback mechanisms. It is suitable for large-scale, distributed, and highly complex HVAC system management scenarios in actual data centers. As shown in the embodiments, this invention has good feasibility and engineering promotion value, and can provide effective technical support for building a green, efficient, and intelligent data center operation and maintenance platform.

[0056] Figure 2 This is a schematic diagram of an electronic device structure provided in an embodiment of the present invention. Please refer to... Figure 2The electronic device provided in this embodiment includes a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement a data center operation and maintenance suggestion generation method based on multimodal input technology according to the present invention.

[0057] It should be noted that, in addition to Figure 2 In addition to the memory and processor shown, electronic devices may include other hardware depending on their actual functions, which will not be elaborated further.

[0058] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for generating data center operation and maintenance suggestions based on multimodal input technology.

[0059] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for generating data center operation and maintenance suggestions based on multimodal input technology, characterized in that, Includes the following steps: S1. Interact with the user, accept user instructions and documents, including data center air conditioning system design drawings and sensor files; the sensor files include sensor installation locations and sensor detection data. S2. Convert the data center air conditioning system design drawings and sensor installation locations corresponding to the target room into information that the large language model can understand; if the required information is insufficient, record the room and the insufficient information, and proceed to step S7. S3. Convert the real-time data detected by the sensors corresponding to the target room into information that the large language model can understand, including cleaning the data and generating semantic summaries of data trends; if the required information is insufficient, record the room and the insufficient information, and proceed to step S7. S4. Construct a first prompt word based on the information transformed in steps S2 and S3, input the first prompt word into the large language model, and generate an analysis of the temperature distribution and air supply of the data center air conditioning system. S5. The query system allows for the control of equipment and control range; S6. Based on the analysis of temperature distribution and air supply of the data center air conditioning system, as well as the control equipment and control range, construct a second prompt word, input the second prompt word into the large language model, and generate operation and maintenance suggestions; S7. Summarize the room information for which maintenance suggestions were successfully generated into a first file, and summarize the room information for which maintenance suggestions were not successfully generated into a second file. Feedback both files to the user.

2. The method according to claim 1, characterized in that, The multimodal input technology includes matching user-input files with predefined semantic descriptions using a text-image matching model to extract room structure, air supply equipment layout, and sensor location information; if the data center air conditioning system design drawings and sensor location information of the target room are missing or the matching score is lower than a preset threshold, then the required information is considered insufficient.

3. The method according to claim 1, characterized in that, The sensor data includes room environmental monitoring parameters, including temperature, humidity, wind speed, and CO2 concentration.

4. The method according to claim 1, characterized in that, The semantic summary includes the parameter fluctuation trend and deviation analysis from the set value within each time period.

5. The method according to claim 1, characterized in that, Both the first and second prompt words are multi-turn dialogue contexts constructed using natural language, which include device parameters, spatial structure information, and user-expected goals.

6. The method according to claim 1, characterized in that, The large language model is a large-scale pre-trained language model that supports multimodal inputs. The large language model has the ability to extract related information from natural language, image descriptions and structured data.

7. The method according to claim 1, characterized in that, The two feedback documents given to the user are presented in the form of structured reports, including the identification number of each room, suggested content or reasons for failure, and suggested update time.

8. The method according to claim 1, characterized in that, Based on the feedback in the second document, users modify or supplement the relevant information and re-execute steps S1 to S7 until all target rooms have successfully generated operation and maintenance suggestions.

9. An electronic device comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the data center operation and maintenance suggestion generation method based on multimodal input technology as described in any one of claims 1-8.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements a data center operation and maintenance suggestion generation method based on multimodal input technology as described in any one of claims 1-8.

Images