Data center inspection auxiliary question and answer method, device and equipment and storage medium

Abstract

The invention relates to a data center inspection auxiliary question-answering method, device and equipment and a storage medium. Comprising the following steps: acquiring question information of an inspector, and extracting key information from the question information; matching the key information with non-text data in a preset non-text database, and determining target non-text data from the preset non-text database according to a non-text data matching result; matching the key information with question and answer data in a preset question and answer historical database, and determining target question and answer data from the preset question and answer historical database according to a question and answer data matching result; inputting the target non-text data and the target question and answer data into a preset cue word generation model to obtain a target cue word, and inputting the target cue word into a pre-trained target inspection large model to obtain reply information for the question information; and feeding back reply information to the inspection personnel. Therefore, the reply efficiency of the question information can be improved by utilizing machine learning, and inspection personnel can be helped to quickly complete inspection work.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technology, and in particular to a data center inspection auxiliary question-answering method, device, equipment and storage medium. Background Art

[0002] With the development of science and technology, related companies have established large-scale data centers in many places across the country. In order to ensure the long-term and stable operation of the data centers, professional inspection personnel are usually arranged to conduct regular inspections of the data centers.

[0003] In the related art, inspectors will conduct inspections in designated inspection areas according to inspection tasks. However, when faced with larger inspection areas or more tedious inspection tasks, inspectors need to constantly operate equipment and move in the inspection area. This process wastes a lot of time, resulting in low inspection efficiency. At the same time, the professional level of inspectors is difficult to grasp, resulting in poor accuracy of inspection results. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure provide a data center inspection auxiliary question and answer method, device, equipment and storage medium to solve the problems existing in the related art.

[0005] In a first aspect of an embodiment of the present disclosure, a data center inspection assisted question and answer method is provided, the method comprising: obtaining question information from an inspection personnel and extracting key information from the question information; matching the key information with non-text data in a preset non-text database to obtain a non-text data matching result, and determining target non-text data from the preset non-text database based on the non-text data matching result; matching the key information with question and answer data in a preset question and answer history database to obtain a question and answer data matching result, and determining target question and answer data from the preset question and answer history database based on the question and answer data matching result; inputting the target non-text data and target question and answer data into a preset prompt word generation model to obtain a target prompt word, and inputting the target prompt word into a pre-trained target inspection large model to obtain reply information for the question information; and feeding back the reply information to the inspection personnel.

[0006] A second aspect of an embodiment of the present disclosure provides a data center inspection auxiliary question and answer device, which is applied to the data center inspection auxiliary question and answer method as in the first aspect, and the device includes: an extraction module, which is used to obtain question information from inspection personnel and extract key information from the question information; a first matching module, which is used to match the key information with non-text data in a preset non-text database, obtain non-text data matching results, and determine target non-text data from the preset non-text database based on the non-text data matching results; a second matching module, which is used to match the key information with question and answer data in a preset question and answer history database, obtain question and answer data matching results, and determine target question and answer data from the preset question and answer history database based on the question and answer data matching results; an acquisition module, which is used to input the target non-text data and target question and answer data into a preset prompt word generation model, obtain target prompt words, and input the target prompt words into a pre-trained target inspection large model to obtain reply information for the question information; and a feedback module, which is used to feed back the reply information to the inspection personnel.

[0007] According to a third aspect of an embodiment of the present disclosure, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned data center inspection assisted question-and-answer method.

[0008] In a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the above-mentioned data center inspection assisted question-and-answer method are implemented.

[0009] According to a fifth aspect of the present disclosure, a computer program product is provided, and when the computer program / instructions are executed by a processor, the steps of the above-mentioned data center inspection assisted question-and-answer method are implemented.

[0010] At least one of the above-mentioned technical solutions adopted in the embodiments of the present disclosure can achieve the following beneficial effects: obtaining question information from patrol personnel and extracting key information from the question information; matching the key information with non-text data in a preset non-text database to obtain non-text data matching results, and determining target non-text data from the preset non-text database based on the non-text data matching results; matching the key information with question and answer data in a preset question and answer history database to obtain question and answer data matching results, and determining target question and answer data from the preset question and answer history database based on the question and answer data matching results; inputting the target non-text data and target question and answer data into a preset prompt word generation model to obtain target prompt words, and inputting the target prompt words into a pre-trained target inspection large model to obtain reply information for the question information; and feeding back the reply information to the patrol personnel.

[0011] Based on this, the disclosed embodiment can first extract key information from the question information, and then match the key information with the data in the preset non-text database and the preset question and answer history database, so that the target non-text data can be given for the key information, and the target question and answer data related to the key information in the historical process can also be given, and then the preset prompt word generation model is used to obtain the target prompt word for the question information according to the target question and answer data and the target non-text data, that is, the key information is further analyzed, and finally, the target inspection large model can be used to analyze the target prompt word, obtain the reply information for the question information, and feedback the reply information to the inspection personnel. It can be seen that the disclosed embodiment gives a reply to the questions raised by the inspection personnel without the need for the inspection personnel to move. In this process, the inspection personnel only need to ask questions, and then the machine model can be combined to analyze and answer the question information, not only for the question information itself, but also for the question and answer data related to the question information in the historical process, so as to ensure the continuity of the equipment inspection work and the traceability of historical problems; at the same time, the use of machine learning can also improve the reply efficiency while ensuring the professionalism and accuracy of the reply information, so as to help the inspection personnel complete the inspection work quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other purposes, features and advantages of the present disclosure will become more apparent by describing the embodiments of the present disclosure in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 A flowchart of a data center inspection assisted question-answering method provided by an exemplary embodiment of the present disclosure;

[0014] Figure 2 A flowchart of a question and answer history database processing process provided by an exemplary embodiment of the present disclosure;

[0015] Figure 3 A schematic diagram of the structure of an inspection auxiliary question-answering system provided by an exemplary embodiment of the present disclosure;

[0016] Figure 4 A structural schematic diagram of another inspection auxiliary question-answering system provided by an exemplary embodiment of the present disclosure;

[0017] Figure 5 A flowchart of another data center inspection auxiliary question-and-answer method provided by an exemplary embodiment of the present disclosure;

[0018] Figure 6A schematic diagram of the structure of a data center inspection auxiliary question-and-answer device provided by an exemplary embodiment of the present disclosure;

[0019] Figure 7 A schematic diagram of the structure of an electronic device provided for an exemplary embodiment of the present disclosure;

[0020] Figure 8 A schematic diagram of the structure of a computer system provided for an exemplary embodiment of the present disclosure;

[0021] Fig. 9 A schematic diagram of a computer program product provided for an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0023] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0024] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0025] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0026] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0027] Related companies have established large-scale data center infrastructure in many places across the country. In order to ensure the long-term and stable operation of the data center, daily inspections are essential. In addition, point-to-point regression inspections are also required for fault alarms.

[0028] In the prior art, in the conventional data center inspection method, it is first necessary to assign tasks to the inspection personnel according to the inspection task process, and the inspection personnel will inspect the designated computer room regularly according to the assigned tasks. It should be understood that the data center inspection is generally to inspect the equipment or components in each computer room. During the inspection process, the inspection personnel will check the operating data of each measuring point in the front-end page of the power environment monitoring system, and judge the operating status based on personal inspection and analysis; at the same time, the intelligent device panel in the computer room will read the data manually and fill it into the inspection form. At the same time, the hardware equipment in the data center is inspected by manual observation, and the hardware equipment is mainly tested by manual professionalism. It can be seen that the current inspection work of each computer room in the data center is mainly based on pure manual inspection. The inspection process is highly dependent on the professional level and personal inspection ability of the inspection personnel, and there is no professional guidance tool, resulting in low inspection efficiency, false detection and missed detection. In addition, due to the high mobility of the inspection personnel, the continuity of the inspection work is poor. It should be understood that the power environment monitoring system, referred to as the power environment system, is aimed at the equipment and environment of various stations (communication rooms, data centers, base stations, etc.), and realizes the functions of "telemetry, telesignaling, remote control, and remote adjustment" for the communication power supply, battery packs, UPS, generators, air conditioners and other intelligent and non-intelligent equipment in the station, as well as environmental quantities such as temperature and humidity, smoke, ground water, and access control, providing automated and intelligent management to achieve the purpose of unmanned and efficient management.

[0029] Based on this, the following problems exist in the conventional data center inspection method:

[0030] Problem 1: Lack of automatic real-time data analysis capabilities: Although Data Center Infrastructure Management (DCIM) collects a large amount of monitoring data corresponding to the operation of the computer room, due to the limitations of the inspection personnel in data analysis capabilities, these data often remain at the raw digital level and fail to be converted into easy-to-understand and intuitive information, resulting in the inability to effectively guide the inspection work. It should be understood that DCIM is a technology used to manage data center infrastructure. It includes monitoring, management and optimization of the data center's physical facilities, networks, storage, servers, power and cooling systems. DCIM can help data center administrators better understand the performance and status of the data center, thereby improving the reliability and efficiency of the data center.

[0031] Problem 2: Low usage rate of operation manual: Although the inspection personnel's operation manual describes effective operating procedures and methods, it is difficult for inspection personnel to effectively consult the contents of the operation manual, resulting in the manual not playing its due role.

[0032] Problem 3: High reliance on human expertise: The inspection process relies heavily on the professional knowledge and practical experience of inspectors, which is too subjective and prone to misjudgment.

[0033] Problem 4: Poor inspection continuity: When inspection personnel change, some historical problems in the computer room are difficult to solve, and the inspection quality and continuity face major challenges.

[0034] Problem 5: Lack of historical problem records and tracing: The current system is inadequate in recording historical equipment problems, making it difficult to efficiently trace current faults back to past records, affecting the depth of problem diagnosis and trend analysis.

[0035] Question 6: Low utilization rate of non-text data: For data centers, non-text data such as drawings is very important, but it is often not well utilized due to limitations in application methods and technologies.

[0036] In response to the above problems, the embodiments of the present disclosure provide a data center inspection assisted question-and-answer method, which can construct a large inspection model that has a certain understanding of the target data center based on the specifications of various professions during the construction process of the target data center, including circuits, pipelines, networks, cold sources, inspection processes, computer room cabinet layouts, etc. as training data; at the same time, the embodiments of the present disclosure can also use intelligent body technology to obtain the current dynamic environment system operation data, and use the large model data analysis capabilities to analyze the current dynamic environment system operation data, thereby directly providing more intuitive analysis results for inspection personnel.

[0037] The disclosed embodiment can also be connected to an external non-text database, that is, professional drawings can be used as an important reference for inspectors' questions to guide the inspection model to answer, so as to give more professional and accurate suggestions, dynamically expand and update the knowledge base, and improve the adaptability of the large model; at the same time, the disclosed embodiment can also build a point-to-point question and answer history database to ensure the continuity of the inspection work and the traceability of historical issues.

[0038] Figure 1 The following is a flow chart of a data center inspection auxiliary question-answering method provided by an exemplary embodiment of the present disclosure. Figure 1 As shown, the specific steps include:

[0039] S101, construct a non-text database based on the non-text data in the data generated during the construction process and the operation and maintenance process of the data center, and construct a training set based on the text data in the data, and then use the training set to train the preset question-answering model to obtain a large target inspection model.

[0040] In some embodiments, data generated during the construction process and operation and maintenance of the target data center can be collected, and a non-text database can be constructed based on the non-text data in the data, and a training set can be constructed based on the text data in the data. Among them, the non-text database may include non-text description information of each professional map and professional map, such as picture name, detailed description and picture type. It should be understood that the non-text database of the embodiment of the present disclosure can be a Milvus vector library. Among them, the goal of the Milvus vector database is to store, index and manage large-scale embedded vectors generated by deep neural networks and other machine learning models. As a database specially designed to process queries on input vectors, it is capable of processing vector indexes on a trillion scale.

[0041] Specifically, the disclosed embodiment needs to build a large inspection model for the target data center. First, it is necessary to collect text data during the construction process and the operation and maintenance process of the target data center. The text data during the construction process of the target data center may include instruction documents such as computer room circuits, pipelines, networks, cold sources, inspection processes, and computer room cabinet layouts, and may also include other text files. The text data during the operation and maintenance of the target data center includes text data generated when the equipment in the data center is operating normally or during maintenance. The text data may also be content manually recorded by relevant inspection personnel.

[0042] In practical applications, when faced with text data, the text data can be segmented into different blocks according to paragraphs, and the meaningless parts, such as the main directory and headlines, can be filtered out. Then, the segmented text blocks can be segmented into multiple sentences according to punctuation marks, and the multiple sentences can be input into the zero-shot learning model for training. It should be understood that the zero-shot learning model can be full-task zero-shot learning-mT5: mT5 is a multilingual version of the T5 (Text-to-Text Transfer Transformer) model, which is designed to handle a variety of natural language processing tasks, including translation, summarization, question and answer, etc., through a unified text-to-text framework. mT5 uses a large amount of multilingual data during training, enabling it to understand and generate text in multiple languages.

[0043] For example, when the text data corresponding to the target data center contains the text content "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building. Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively", at this time, the text content can be segmented and assembled into full-task zero-sample learning-mT5 input text according to punctuation marks. Specifically, it can be segmented and assembled into two sentences, "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building" and "Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively". After segmentation and assembly, these two texts are input into full-task zero-sample learning-mT5, and questions such as "What parts does the data center consist of?" and "How many mainframe buildings are there?" can be obtained. Based on this, the question "What parts does the data center consist of?" and its corresponding answer "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building.", and the question "How many mainframe buildings are there?" and its corresponding answer "Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively".

[0044] In practical applications, the embodiments of the present disclosure can process the text data generated during the construction and operation and maintenance of the target data center in the above manner to obtain multiple questions and corresponding answer information. Based on this, a training set can be constructed based on multiple questions and corresponding answer information, and the preset question-answering model can be trained using the training set. At the same time, a target inspection large model is obtained through model fine-tuning. After fine-tuning, the target inspection large model can respond to the target data center questions. It should be understood that the preset question-answering model here can be any question-answering model in the prior art that can achieve the output of answers to input questions.

[0045] In practical applications, when faced with non-text data, such as illustrations and corresponding illustration annotations in the manuals corresponding to the equipment deployed in the target data center. The disclosed embodiment also creates a set of non-text knowledge bases and their comprehensive recall algorithms, extracts key information from the question information initiated by the inspection personnel, and recalls the corresponding non-text data based on multiple feature matching. At the same time, a prompt word model corresponding to the target inspection large model is constructed, so that when interacting with the target inspection large model, the target prompt word corresponding to the question information can be determined, and then the target inspection large model is guided to generate specific outputs through the target prompt word, thereby helping the inspection personnel to obtain the required information more efficiently. It should be understood that Prompt is an interactive method based on natural language processing, which realizes communication between inspection personnel and machines through machine parsing of natural language.

[0046] Specifically, non-text data can be obtained during the construction and operation and maintenance of the target data center, such as circuit diagrams, piping diagrams, flow charts, tables and other non-text data. In addition, various images related to data centers, such as engineering drawings (CAD files), architectural drawings (BIM files) and electronics industry, can be collected and organized to build a non-text database.

[0047] For the above non-text data, it is necessary to uniformly convert all the collected non-text data into JPG format. Among them, for images in PNG format, it is relatively convenient to convert PNG format into JPG format. You can use the built-in image processing library (Python Imaging Library, PIL), use the Image.open function to open the PNG format image, and store it in the variable img, and then use the img.save method to save the image in JPG format. Use matplotlib and Rectangle in matplotlib.patches to read the CAD file and convert it into a JPG format image; for the relatively few eddx format flowcharts, you can use manual screenshots or use existing tools to convert them into JPG format images. It should be understood that the Image.ope function is used to open and identify image files; the variable img is a structure variable related to the image; the img.save function is used to save the image to a file; Matplotlib is a widely used drawing library in Python that can be used to create various types of static, dynamic or interactive charts and visualizations; the matplotlib.patches module is used to provide drawing methods for various shapes, including rectangles, circles, ellipses, polygons, etc.; the main function of the Rectangle function is to add a rectangular border to the specified area, so that the target area of ​​the image to be processed is visualized, thereby facilitating processing.

[0048] Finally, according to the fixed format, as shown in Table 1, a piece of non-text data can be used to describe various types of images corresponding to a device. Therefore, a piece of non-text data may include multiple types of data. Based on this, annotations can be written for each piece of non-text data, and the fields corresponding to each piece of non-text data can be vectorized using the embedding model (Embeddings), and then the vectorized non-text data can be stored in the non-text database.

[0049] Specifically, the annotations in Table 1 are used to number different non-text data. At the same time, the file type, file name, location, keywords, version and non-text description information of the non-text data are added to the corresponding numbered table, where the file type is the format type when the non-text data is saved; the file name is used to describe the device or professional diagram corresponding to the non-text data; the location is used to specify the device corresponding to the file name; the keyword is used to assist in retrieval and fast positioning. Here, a device may correspond to one or more keywords. Therefore, when corresponding to multiple keywords, multiple keywords can be grouped together, and a keyword group can be used as the retrieval condition for the non-text data; the version is used to describe the update of the non-text data. The newer the non-text data, the higher the version data corresponding to it; the non-text description information is used to describe the non-text data in detail. Based on this, a Milvus vector library for non-text data can be constructed. It should be understood that the Embedding model refers to the process of mapping high-dimensional data to low-dimensional space.

[0050] Table 1 Non-text knowledge annotation

[0051]

[0052] S102, obtaining question information, determining key information in the question information, and matching the key information with non-text data in a non-text database, and determining target non-text data according to the matching results.

[0053] In some embodiments, since the data includes text data and non-text data, the response to the question information may include text data or non-text data. Based on this, the disclosed embodiment creates a non-text database and its comprehensive recall algorithm, which recognizes the question information, matches multiple features to recall the corresponding non-text data, and constructs a prompt word model corresponding to the target inspection model.

[0054] In practical applications, the disclosed embodiments use two methods to compare the question information with the non-text data in the non-text database. For the first method, the universal natural language understanding model (RexUniNLU) can be used to extract the keyword groups in the question information, and the keywords in the keyword groups can be vectorized. Then, the vectorized keyword groups are matched with the keyword groups corresponding to the non-text data in the non-text database to obtain the top three candidate keyword groups a1, a2, and a3 with high similarity, and the similarities A1%, A2%, and A3% corresponding to each candidate keyword group are determined respectively. Among them, the similarity corresponding to the candidate keyword group a1 is A1%; the similarity corresponding to the candidate keyword group a2 is A2%; and the similarity corresponding to the candidate keyword group a3 is A3%. It should be understood that in practical applications, the selection of the number of keyword groups can be determined according to actual conditions.

[0055] Specifically, when extracting useful keywords from the question information, words without actual meaning such as auxiliary verbs and modal particles in the question information can be removed to obtain useful keywords. For example, when the question information is "What parts does a data center consist of", the keywords can be "data center", "part" and "compose", and then the keyword vectorization technology in the prior art is used to vectorize the keywords, and then match them one by one with the data in the non-text database. The keyword vectorization technology in the prior art can be Word2Vec, which is a group of related models used to generate word vectors, which are used for training to reconstruct linguistic word texts.

[0056] For the second method, the question information can be generated through the full-task zero-sample learning-mT5 model to generate text description information. Here, the text description information can reflect the real intention of the inspection personnel. At the same time, the text description information is vectorized by using the keyword vectorization technology in the prior art. Finally, the vectorized text description information is compared with the non-text description information corresponding to different non-text data in the non-text database, and n non-text description information with similarity>60% is selected. After the n non-text description information is arranged from large to small according to the similarity, the first three candidate non-text description information b1, b2, and b3 are taken, and their corresponding similarity percentages B1%, B2%, and B3 are obtained. Among them, the similarity corresponding to the candidate non-text description information b1 is B1%; the similarity corresponding to the candidate non-text description information b2 is B2%; and the similarity corresponding to the candidate non-text description information b3 is B3%. It should be understood that in actual applications, the selection of the number of non-text description information can be determined according to actual conditions.

[0057] In practical applications, after determining multiple candidate keyword groups and multiple candidate non-text description information respectively by two methods, it is also necessary to determine whether there is overlap between the content of the multiple candidate keyword groups and the multiple candidate non-text description information, that is, it is necessary to determine whether there is overlap in the text description content between a1, a2, a3 and b1, b2, b3. If not, the comprehensive recall rate is Recall=0, and the recall ends, that is, the recall of non-text data is not involved in the question information; when there are multiple overlaps, according to the recall rate calculation formula Recall=TP / (TP+FN), where TP represents the number of recalls, and the number of FN not recalled is used to obtain the coefficient 3 / n of B%, where n represents the total number of non-text data, and 3 represents the number of current candidate keyword groups and candidate non-text description information, and then the comprehensive recall rate is calculated respectively. For example, assuming that a1 and b2, a2 ​​and b1 overlap, the first comprehensive recall rate Recall-1 of a1 and b2, and the second comprehensive recall rate Recall-2 of a2 and b1 are respectively:

[0058] Recall-1=(A1%+(3 / n)*B2%) / 2

[0059] Recall-2=(A2%+(3 / n)*B1%) / 2

[0060] After determining the comprehensive recall rate between multiple candidate keyword groups and multiple candidate non-text description information, a piece of non-text data with the largest comprehensive recall rate is selected as the target non-text data, i.e., the final recall object, and input into the prompt word model. It should be understood that the recall rate is an indicator to measure the classification model's ability to identify positive class (class of interest) samples, indicating the proportion of samples that are correctly identified as positive by the model among all samples that are actually positive.

[0061] S103, constructing a question and answer history database, and matching key information in the question information with question and answer data in the question and answer history database, and determining target question and answer history data according to the matching result.

[0062] In some embodiments, in order to ensure the continuity and traceability of the inspection work, the disclosed embodiments also create a point-to-point question and answer storage and traceability question and answer recall method, the question information identifies the instantiated keywords, and stores the question and answer information point-to-point according to the instantiated keywords to form a question and answer history database, and at the same time recalls the target question and answer data according to the instantiated keywords and question semantics. The specific steps are as follows:

[0063] First, it is necessary to construct an instantiated keyword set. The instantiated keywords described in the embodiments of the present disclosure refer to texts that are directed to specific things, such as: Room 301, Building 1, Cabinet 4-2, Transmission Room, etc. Based on this type of data, specific equipment or components can be located. Based on the inherent characteristics of the inspection of the computer room, the focus is on equipment, sensors, circuits, pipelines, etc. These instantiated keywords are easy to obtain in the dictionary tables of various professional system databases. These data are uniformly sorted into data sets by profession, and these instantiated texts are vectorized using Word2vec and stored in the instantiated vector database.

[0064] The question-and-answer history database is a storage medium that uses the instantiated keywords in the instantiated vector database as the primary key to store the conversations between the inspection personnel and the large model. First, the question information can be extracted using the stuttering word segmentation. Here, the key information can be multiple word segmentation information after the question information is segmented, and each word segmentation is vectorized using Word2vec. Then, the vectorized multiple word segmentations are traversed and matched with the instantiated keywords in the question-and-answer history database for similarity, and the cosine similarity is used to solve the similarity between the word segmentations. Cosine similarity is a method to measure the size of the angle between two vectors in vector space. In two-dimensional space, vectors can be regarded as arrows starting from the origin, and cosine similarity is the cosine value of the angle between the two arrows. This value is between -1 and 1, and the larger the cosine value, the more similar the two vectors are, and the smaller the cosine value, the less similar the two vectors are. Set the similarity threshold>0.6, and the specific calculation formula is as follows:

[0065]

[0066] Where: A represents the vector corresponding to a word segmentation; B represents the vector corresponding to the instantiated keyword; A·B represents the dot product of vector A and vector B; ||A|| represents the modulus of vector A, that is, the length; ||B|| represents the modulus of vector B, that is, the length.

[0067] In actual applications, when the similarity of all word segments of the question information is <0.6, it means that the question information does not contain instantiated keywords and is not stored; when one or more similarities are >0.6, the word segment with the largest similarity is taken as the target keyword, the target keyword is used as the key (Key), and the question and answer is used as the value (Value). All the question and answer content of this round of dialogue is stored in the question and answer history database. The basis for judging the end of this round of dialogue is: the end of the dialogue or the emergence of new instantiated keywords. It should be understood that the key-value database is a simple and efficient data storage method that stores data in the form of key-value pairs. Each key is a unique identifier used to quickly retrieve and access the corresponding data value.

[0068] After matching the target keyword in the previous step, all the Q&A contents of this round of dialogue are stored in the Q&A history database, and the Q&A data is retrieved in the Q&A history database. The most recent multiple sets of Q&A data are retrieved based on the matched keywords. A set of Q&A data can include the Q&A data of a certain device or component at a certain time, and can also include the Q&A data of a certain device or component when it fails. There are two ways to process the feedback of multiple sets of Q&A data:

[0069] Method 1: Full return, that is, directly displaying multiple sets of question and answer data in the front-end interface. By consulting multiple sets of question and answer data, the inspection personnel can understand the historical failures of the corresponding equipment or components, analyze the current cause of the failure, and also draw on historical experience to propose solutions to the current problem, providing a direct reference for the inspection personnel to report or solve the problem.

[0070] Method 2: Perform semantic similarity analysis on multiple sets of question and answer data and the current question information one by one, use the convolutional neural network CNN algorithm to perform similarity analysis, extract features from the two-dimensional matrix of the two sentences obtained by Word2vec, use the two-dimensional matrix as input, and use the convolution layer and pooling layer to extract and classify features, so as to obtain the similarity results between the question information and multiple sets of question and answer data, and then sort the multiple sets of question and answer data according to the similarity results, select the set of question and answer data with the highest similarity as the target question and answer data, and input the set of target question and answer history data into the prompt word model. It should be understood that a set of question and answer history data may include one or more rounds of dialogue, so the target question and answer history data may contain only one round of dialogue or multiple rounds of dialogue, which can be set according to actual conditions.

[0071] Figure 2 The following is a flowchart of a question and answer history database processing process provided by an exemplary embodiment of the present disclosure. Figure 2 As shown, the specific steps include:

[0072] S201, when question information is obtained, segment the question information, and vectorize multiple segmented words to obtain multiple segmented word vectors.

[0073] Specifically, the question information can be segmented using Jieba and each segmented word can be vectorized using Word2vec.

[0074] S202, using cosine similarity, similarity is calculated between the multiple word segmentation vectors and the instantiated keywords in the instantiated vector database.

[0075] Specifically, each word segmentation vector is then matched with the instantiated keywords in the instantiated vector database by similarity traversal, and the similarity between word segmentations is solved using cosine similarity, which is a method for measuring the angle between two vectors in vector space. The specific solution process can be referred to above.

[0076] S203, determining whether the similarity of all word segments of the question information is less than 0.6.

[0077] Specifically, when the similarities of all the word segments of the question information are <0.6, it means that the question information does not contain the instantiated keyword. In this case, step S204 is executed; and when one or more similarities are >0.6, step S205 is executed.

[0078] S204: Do not store. That is, do not store the data generated in this question-and-answer process.

[0079] S205, storing all the question and answer contents of this round of dialogue into the question and answer history database; at the same time, determining the keyword with the greatest similarity as the target keyword, and determining multiple groups of question and answer data based on the target keyword.

[0080] Specifically, when one or more similarities are greater than 0.6, the one with the largest similarity is taken as the target keyword, the target keyword is used as the key, and the question and answer are used as the value. All the question and answer contents of this round of dialogue are stored in the question and answer history database. The termination of this round of dialogue is determined based on: the end of the dialogue or the appearance of a new instantiated keyword.

[0081] After matching the target keyword, all the questions and answers of this round of dialogue are stored in the question and answer history database, and the question and answer data is retrieved in the question and answer history database to retrieve the most recent multiple sets of question and answer data according to the matched keyword. Among them, a set of question and answer history data can include the question and answer data of a certain device or component at a certain time, and can also include the question and answer data of a certain device or component when it fails.

[0082] S206, answering the inspection personnel's questions based on the multiple sets of question and answer data.

[0083] Specifically, there are two ways to process feedback on multiple sets of historical question and answer data:

[0084] Method 1: Full return, that is, directly displaying multiple sets of question and answer data in the front-end interface. By consulting multiple sets of question and answer data, the inspection personnel can understand the historical failures of the corresponding equipment or components, analyze the current cause of the failure, and also draw on historical experience to propose solutions to the current problem, providing a direct reference for the inspection personnel to report or solve the problem.

[0085] Method 2: Perform semantic similarity analysis on multiple sets of question and answer data and the current question information one by one, use the convolutional neural network CNN algorithm to perform similarity analysis, extract features from the two-dimensional matrix of the two sentences obtained by Word2vec, use the two-dimensional matrix as input, and use the convolution layer and pooling layer to extract and classify features, so as to obtain the similarity results between the question information and multiple sets of question and answer data, and then sort the multiple sets of question and answer data according to the similarity results, select the set of question and answer data with the highest similarity as the target question and answer data, and input the set of target question and answer history data into the prompt word model. It should be understood that a set of question and answer history data may include one or more rounds of dialogue, so the target question and answer history data may contain only one round of dialogue or multiple rounds of dialogue, which can be set according to actual conditions.

[0086] S104, input the target question and answer data and the target non-text data into the prompt word model to obtain the target prompt word, and the target inspection model determines the reply information according to the target prompt word, where the reply information is the answer to the question information.

[0087] In some embodiments, the prompt word model can guide the target inspection model to generate specific outputs based on the text prompts input by the inspection personnel when interacting with the target inspection model, thereby helping the inspection personnel to obtain the required information more efficiently.

[0088] In some embodiments, the target inspection big model of the disclosed embodiments is also combined with data center infrastructure management (DCIM), and the two are connected as a whole through an intelligent agent. Relying on the intention recognition capability of the big model, the topological cascade identifies the corresponding tools in the intelligent agent, and calls the tools to obtain DCIM operation data, so as to realize the target inspection big model's recognition of the operating status of each device or component in each computer room of the target data center, and give responses to the question information based on the current status.

[0089] Figure 3 The structure diagram of a patrol inspection auxiliary question-answering system provided by an exemplary embodiment of the present disclosure is as follows. Figure 3As shown, the patrol inspection auxiliary question-answering system of the embodiment of the present disclosure is designed with three main bodies, namely patrol inspectors 301, intelligent agents 302 and DCIM 303. Based on this, when patrol inspectors 301 ask questions, intelligent agents 302 analyze the question information corresponding to the question, and obtain the operating status of each device or equipment in the computer room at the current moment from DCIM, so that answers can be given in combination with the actual operating conditions, and the specific answer information results are displayed to the patrol inspectors. It should be understood that the relevant processing of the question information in the embodiment of the present disclosure is all executed in intelligent agents 302, that is, the target patrol large model and the model algorithm designed for the processing process of the question information are all set in intelligent agents 302.

[0090] In actual applications, when establishing a connection between DCIM and an intelligent agent, a hierarchical topology of an operation data interface can be established, and corresponding equipment groups can be established according to the relationship between the equipment. For example, the equipment group can be "cold source-water-cooled air conditioner-liquid cooling cabinet". During the detection process, each sub-equipment has a corresponding detection tool. At this time, a hierarchical topological relationship between different sub-equipment can be established, and the detection tools corresponding to multiple sub-equipment in the group of equipment can be regarded as a tool group. Then, during inspection, the current operation status of the equipment group can be determined according to the functional description (function-description) parameter value corresponding to the tool group. Based on this, the interface professional topology cascade method can be used to accurately locate the required tool group according to the professional type involved in the semantics when the inspection personnel ask questions, thereby improving the matching accuracy of the target inspection large model, so that the target inspection large model can respond based on the operating environment background, making the response more targeted.

[0091] For example, to establish a hierarchical topology for the operation data interface, you can create tool groups according to major disciplines, minor disciplines, and sub-models, for example: "Cold source - water-cooled air conditioner - liquid cooling cabinet" / "Cold source - air-cooled air conditioner - ordinary cabinet", that is, in the tool group, the function-description parameter value is "Query the real-time input cooling capacity according to the computer room number, professional level: cold source - water-cooled air conditioner - liquid cooling cabinet" / "Query the real-time input cooling capacity according to the computer room number, professional level: cold source - air-cooled air conditioner - ordinary cabinet".

[0092] Figure 4 This is a schematic diagram of the structure of another inspection auxiliary question-answering system provided by an exemplary embodiment of the present disclosure. Figure 4As shown, when the patrol inspector 401 raises a question, the preset text processing model 402 will first process the question information to obtain key information, and then perform similarity matching on the key information with the data in the question and answer history database 404 and the non-text knowledge base 403, determine the target non-text data and the target question and answer data according to the matching results, and input the target non-text data and the target question and answer data into the prompt word model 405 to obtain the target prompt word, and the target patrol large model 406 determines the reply information according to the target prompt word and the current operation data of DCIM407, and sends the reply information to the patrol inspector 401. It should be understood that the specific model selection of the preset text processing model 402 will be selected according to the actual situation, and please refer to the above for the specific processing process of the question information.

[0093] The data center inspection auxiliary question-and-answer method provided in the embodiments of the present disclosure may be executed by a terminal or by a chip applied to the terminal.

[0094] Exemplarily, the above-mentioned terminals may include one or more of mobile phones, tablet computers, wearable devices, vehicle-mounted devices, laptop computers, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants (PDAs), and wearable devices based on augmented reality (AR) and / or virtual reality (VR) technology, and may also include but not be limited to remote control devices, wearable devices, street lamps, home appliances and other smart terminals, which are not specifically limited in the embodiments of the present disclosure.

[0095] Figure 5 A flowchart of another data center inspection assisted question-answering method provided by an exemplary embodiment of the present disclosure. Figure 5 As shown, the specific steps include:

[0096] S501, obtaining question information from inspection personnel, and extracting key information from the question information.

[0097] In some embodiments, the inspection personnel can initiate questions through the terminal, and when the server obtains the question information, it will extract the key information in the question information. Here, the key information can be a keyword group and text description information, that is, the question information can be subjected to keyword extraction to obtain multiple keywords, and the multiple keywords constitute the keyword group corresponding to the question information.

[0098] Specifically, assuming that the question information is "What parts does a data center consist of?", when extracting key information from the question information, auxiliary words and punctuation marks can be removed, that is, the key information can be "data center", "part" and "composition". Here, when extracting key information, the existing neural network model can be directly used for keyword extraction.

[0099] As for the text description information, the main purpose is to simplify the question information and obtain the semantics corresponding to the question information. For example, the text description information of the question information "What parts does a data center consist of?" can be "What parts does a data center consist of". Based on this, only the key information in the question information can be analyzed, thereby improving the processing efficiency of the question information.

[0100] S502, matching the key information with non-text data in a preset non-text database to obtain a non-text data matching result, and determining target non-text data from the preset non-text database according to the non-text data matching result.

[0101] Specifically, the disclosed embodiment can collect non-text data generated during the construction and operation and maintenance of the target data center to build a preset non-text database. The preset non-text database can include non-text description information of each professional map and professional map, such as picture name, detailed description and picture type.

[0102] Based on this, the key information can be directly matched with the non-text data in the preset non-text database to obtain non-text data matching the question information, which can be specifically professional pictures. Therefore, when providing feedback on the answer to the question information, it is not limited to text-based answers. Professional pictures related to the question information can be directly fed back to the inspection personnel, making it convenient for the inspection personnel to view and improve the inspection efficiency.

[0103] S503, matching the key information with the question and answer data in the preset question and answer history database to obtain a question and answer data matching result, and determining the target question and answer data from the preset question and answer history database according to the question and answer data matching result.

[0104] In some embodiments, the question and answer data generated by the inspectors in the historical process can be obtained, and a preset question and answer history database can be constructed based on the question and answer data. Based on this, when the inspectors have the same questions later, the historical question and answer data can be directly fed back to the inspectors, thereby helping the inspectors to conduct inspections quickly.

[0105] Specifically, it is necessary to match the key information corresponding to the question information with the question and answer data in the preset question and answer history database, and select the target question and answer data based on the matching results, and then feed back the question and answer data to the inspection personnel, so that the inspection personnel can be given suggestions quickly when the same question appears repeatedly.

[0106] S504, input the target non-text data and the target question and answer data into a preset prompt word generation model to obtain a target prompt word, and input the target prompt word into a pre-trained target inspection model to obtain response information for the question information.

[0107] In some embodiments, the disclosed embodiments also utilize a preset prompt word generation model to analyze and process target non-text data and target question and answer data, obtain target prompt words, and then use the target prompt words to guide the target inspection model to generate specific outputs.

[0108] Specifically, the preset prompt word generation model can be a prompt word model. Prompt can parse the target non-text data and target question and answer data through a machine to determine the final target prompt word, and then guide the target inspection model to generate output for the target data center through the target prompt word. The introduction of Prompt can help inspectors obtain the required information more efficiently while improving the output accuracy. It should be understood that Prompt is an interactive method based on natural language processing. It realizes communication between inspectors and machines through machine parsing of natural language.

[0109] It can be seen that the embodiments of the present disclosure can not only analyze the current question information, but also combine historical data, so that when inspecting the data center, a more accurate analysis can be made in combination with the historical usage of the data center.

[0110] S505, feeding back reply information to the inspection personnel.

[0111] In some embodiments, the reply information may be displayed in the front-end interface of the corresponding terminal.

[0112] Based on this, the disclosed embodiment can first extract key information from the question information, and then match the key information with the data in a pre-set preset non-text database and a preset question and answer history database. In this way, target non-text data can be given for the key information, and target question and answer data related to the key information in the historical process can also be given. Then, the preset prompt word generation model is used to obtain the target prompt words for the question information according to the target question and answer data and the target non-text data, that is, the key information is further analyzed. Finally, the target inspection large model can be used to analyze the target prompt words, obtain reply information for the question information, and feedback the reply information to the inspection personnel.

[0113] It can be seen that the embodiment of the present disclosure can give answers to the questions raised by the inspectors without the need for the inspectors to move. In this process, the inspectors only need to ask questions, and then the question information can be analyzed and answered in combination with the machine model. Not only the question information itself but also the question and answer data related to the question information in the historical process will be analyzed, thereby ensuring the continuity of the equipment inspection work and the traceability of historical issues. At the same time, the use of machine learning can also improve the reply efficiency while ensuring the professionalism and accuracy of the reply information, thereby helping the inspectors to quickly complete the inspection work.

[0114] In some embodiments, the key information includes a first keyword group and text description information, and the non-text data is multiple. The key information is matched with the non-text data in a preset non-text database to obtain a non-text data matching result, and the target non-text data is determined from the preset non-text database according to the non-text data matching result, which specifically includes the following steps:

[0115] Step 1, matching the first keyword group with the second keyword group included in the plurality of non-text data respectively, obtaining first non-text data matching results corresponding to the plurality of non-text data, and determining a plurality of candidate keyword groups from a preset non-text database according to the first non-text data matching results;

[0116] In some embodiments, the universal natural language understanding model (RexUniNLU) can be used to extract the first keyword group in the question information, and the keywords in the first keyword group are vectorized, and then the vectorized first keyword group is matched with the second keyword group corresponding to each non-text data in the non-text database, and the top three candidate keyword groups a1, a2, and a3 with high similarity are obtained from multiple second keyword groups, and the similarities A1%, A2%, and A3% corresponding to each candidate keyword group are determined respectively. Among them, the similarity corresponding to the candidate keyword group a1 is A1%; the similarity corresponding to the candidate keyword group a2 is A2%; and the similarity corresponding to the candidate keyword group a3 is A3%. It should be understood that in actual applications, the selection of the number of keyword groups can be determined according to actual conditions.

[0117] Specifically, when extracting useful keywords from the question information, words without actual meaning, such as auxiliary verbs and modal particles, can be removed from the question information to obtain useful keywords. For example, when the question information is "What parts does a data center consist of?", the keywords can be "data center", "parts" and "composition", and then the keywords are vectorized using keyword vectorization technology, and then matched one by one with the data in the non-text database. Among them, the keyword vectorization technology can be Word2Vec, which is a group of related models used to generate word vectors, which are used for training to reconstruct linguistic word texts.

[0118] Step 2, matching the text description information with the non-text description information included in the plurality of non-text data, obtaining second non-text data matching results corresponding to the plurality of non-text data, and determining a plurality of candidate non-text description information from a preset non-text database according to the second non-text data matching results;

[0119] In some embodiments, the question information can be generated through the full-task zero-sample learning-mT5 model to generate text description information. Here, the text description information can reflect the real intention of the inspection personnel. At the same time, the text description information is vectorized by using the keyword vectorization technology in the prior art. Finally, the vectorized text description information is compared with the non-text description information corresponding to different non-text data in the non-text database, and n non-text description information with similarity>60% is selected. After the n non-text description information is arranged from large to small according to the similarity, the first three candidate non-text description information b1, b2, and b3 are taken, and their corresponding similarity percentages B1%, B2%, and B3 are obtained. Among them, the similarity corresponding to the candidate non-text description information b1 is B1%; the similarity corresponding to the candidate non-text description information b2 is B2%; and the similarity corresponding to the candidate non-text description information b3 is B3%. It should be understood that in actual applications, the selection of the number of non-text description information can be determined according to actual conditions.

[0120] Step 3: when there are identical contents between multiple candidate keyword groups and multiple candidate non-text description information, determine the recall rate between the candidate keyword groups and candidate non-text description information with identical contents.

[0121] In some embodiments, after determining multiple candidate keyword groups and multiple candidate non-text description information respectively by two methods, it is still necessary to determine whether there is overlap in content between the multiple candidate keyword groups and the multiple candidate non-text description information, that is, it is necessary to determine whether there is overlapping text description content between a1, a2, a3 and b1, b2, b3. If not, the comprehensive recall rate is Recall=0, and the recall ends, that is, the question information does not involve the recall of non-text data.

[0122] When there are multiple overlaps, the recall rate calculation formula Recall = TP / (TP+FN), where TP represents the number of recalls and the number of FN not recalled is used to obtain the coefficient 3 / n of B%, where n represents the total number of non-text data and 3 represents the number of current candidate keyword groups and candidate non-text description information. Then, the average comprehensive recall rate is calculated respectively. For example, assuming that a1 and b2, a2 ​​and b1 overlap, the first comprehensive recall rate Recall-1 of a1 and b2, and the second comprehensive recall rate Recall-2 of a2 and b1 are respectively:

[0123] Recall-1=(A1%+(3 / n)*B2%) / 2

[0124] Recall-2=(A2%+(3 / n)*B1%) / 2

[0125] Step 4: determining a target keyword group and target text description information from candidate keyword groups and candidate non-text description information with the same content according to the recall rate, and determining target non-text data according to the target keyword group and target text description information.

[0126] Specifically, after determining the comprehensive recall rates between multiple candidate keyword groups and multiple candidate non-text description information, a piece of non-text data with the largest comprehensive recall rate is selected as the target non-text data, ie, the final recall object.

[0127] In some embodiments, the non-text database is determined by non-text data generated during the construction process and the operation and maintenance process of the target data center.

[0128] Specifically, non-text data can be obtained during the construction and operation and maintenance of the target data center, such as circuit diagrams, piping diagrams, flow charts, tables and other non-text data. In addition, various images related to data centers, such as engineering drawings (CAD files), architectural drawings (BIM files) and electronics industry, can be collected and organized to build a non-text database.

[0129] For the above non-text data, it is necessary to uniformly convert all the collected non-text data into JPG format. Among them, for images in PNG format, it is relatively convenient to convert PNG format into JPG format. You can use the built-in image processing library (Python Imaging Library, PIL), use the Image.open function to open the PNG format image, and store it in the variable img, and then use the img.save method to save the image in JPG format. Use matplotlib and Rectangle in matplotlib.patches to read the CAD file and convert it into a JPG format image; for the relatively few eddx format flowcharts, you can use manual screenshots or use existing tools to convert them into JPG format images. It should be understood that the Image.ope function is used to open and identify image files; the variable img is a structure variable related to the image; the img.save function is used to save the image to a file; Matplotlib is a widely used drawing library in Python that can be used to create various types of static, dynamic or interactive charts and visualizations; the matplotlib.patches module is used to provide drawing methods for various shapes, including rectangles, circles, ellipses, polygons, etc.; the main function of the Rectangle function is to add a rectangular border to the specified area, so that the target area of ​​the image to be processed is visualized, thereby facilitating processing.

[0130] Finally, according to the fixed format, as shown in Table 1, a piece of non-text data can be used to describe various types of images corresponding to a device. Therefore, a piece of non-text data may include multiple types of data. Based on this, annotations can be written for each piece of non-text data, and the fields corresponding to each piece of non-text data can be vectorized using the embedding model (Embeddings), and then the vectorized non-text data can be stored in the non-text database.

[0131] Specifically, the annotations in Table 1 are used to number different non-text data. At the same time, the file type, file name, location, keywords, version and non-text description information of the non-text data are added to the corresponding numbered table, where the file type is the format type when the non-text data is saved; the file name is used to describe the device or professional diagram corresponding to the non-text data; the location is used to specify the device corresponding to the file name; the keyword is used to assist in retrieval and fast positioning. Here, a device may correspond to one or more keywords. Therefore, when corresponding to multiple keywords, multiple keywords can be grouped together, and a keyword group can be used as the retrieval condition for the non-text data; the version is used to describe the update of the non-text data. The newer the non-text data, the higher the version data corresponding to it; the non-text description information is used to describe the non-text data in detail. Based on this, a Milvus vector library for non-text data can be constructed. It should be understood that the Embedding model refers to the process of mapping high-dimensional data to low-dimensional space.

[0132] Table 1 Non-text knowledge annotation

[0133]

[0134]

[0135] In some embodiments, there are multiple question and answer data, and the key information is matched with the question and answer data in a preset question and answer history database to obtain the question and answer data matching results, and the target question and answer data is determined from the preset question and answer history database based on the question and answer data matching results, including: matching the key information with multiple question and answer data respectively to obtain question and answer data matching results corresponding to the multiple question and answer data; when the question and answer data matching results corresponding to the multiple question and answer data are all greater than a preset threshold, determining the target question and answer data from the multiple question and answer data based on the question and answer data matching results corresponding to the multiple question and answer data.

[0136] In some embodiments, in order to ensure the continuity and traceability of the inspection work, the disclosed embodiments also create a point-to-point question and answer storage and traceability question and answer recall method, the question information identifies the instantiated keywords, and stores the question and answer information point-to-point according to the instantiated keywords to form a question and answer history database, and at the same time recalls the target question and answer data according to the instantiated keywords and question semantics. The specific steps are as follows:

[0137] First, it is necessary to construct an instantiated keyword set. The instantiated keywords described in the embodiments of the present disclosure refer to texts that are directed to specific things, such as: Room 301, Building 1, Cabinet 4-2, Transmission Room, etc. Based on this type of data, specific equipment or components can be located. Based on the inherent characteristics of the inspection of the computer room, the focus is on equipment, sensors, circuits, pipelines, etc. These instantiated keywords are easy to obtain in the dictionary tables of various professional system databases. These data are uniformly sorted into data sets by profession, and these instantiated texts are vectorized using Word2vec and stored in the instantiated vector database.

[0138] The question-and-answer history database is a storage medium that uses the instantiated keywords in the instantiated vector database as the primary key to store the conversations between the inspection personnel and the large model. First, the question information can be extracted using the stuttering word segmentation. Here, the key information can be multiple word segmentation information after the question information is segmented, and each word segmentation is vectorized using Word2vec. Then, the vectorized multiple word segmentations are traversed and matched with the instantiated keywords in the question-and-answer history database for similarity, and the cosine similarity is used to solve the similarity between the word segmentations. Cosine similarity is a method to measure the size of the angle between two vectors in vector space. In two-dimensional space, vectors can be regarded as arrows starting from the origin, and cosine similarity is the cosine value of the angle between the two arrows. This value is between -1 and 1, and the larger the cosine value, the more similar the two vectors are, and the smaller the cosine value, the less similar the two vectors are. Set the similarity threshold>0.6, and the specific calculation formula is as follows:

[0139]

[0140] Where: A represents the vector corresponding to a word segmentation; B represents the vector corresponding to the instantiated keyword; A·B represents the dot product of vector A and vector B; ||A|| represents the modulus of vector A, that is, the length; ||B|| represents the modulus of vector B, that is, the length.

[0141] In actual applications, when the similarity of all word segments of the question information is <0.6, it means that the question information does not contain instantiated keywords and is not stored; when one or more similarities are >0.6, the word segment with the largest similarity is taken as the target keyword, the target keyword is used as the key (Key), and the question and answer is used as the value (Value). All the question and answer content of this round of dialogue is stored in the question and answer history database. The basis for judging the end of this round of dialogue is: the end of the dialogue or the emergence of new instantiated keywords. It should be understood that the key-value database is a simple and efficient data storage method that stores data in the form of key-value pairs. Each key is a unique identifier used to quickly retrieve and access the corresponding data value.

[0142] After matching the target keyword in the previous step, all the Q&A contents of this round of dialogue are stored in the Q&A history database, and the Q&A data is retrieved in the Q&A history database. The most recent multiple sets of Q&A data are retrieved based on the matched keywords. A set of Q&A data can include the Q&A data of a certain device or component at a certain time, and can also include the Q&A data of a certain device or component when it fails. There are two ways to process the feedback of multiple sets of Q&A data:

[0143] Method 1: Full return, that is, directly displaying multiple sets of question and answer data in the front-end interface. By consulting multiple sets of question and answer data, the inspection personnel can understand the historical failures of the corresponding equipment or components, analyze the current cause of the failure, and also draw on historical experience to propose solutions to the current problem, providing a direct reference for the inspection personnel to report or solve the problem.

[0144] Method 2: Perform semantic similarity analysis on multiple sets of question and answer data and the current question information one by one, use the convolutional neural network CNN algorithm to perform similarity analysis, extract features from the two-dimensional matrix of the two sentences obtained by Word2vec, use the two-dimensional matrix as input, and use the convolution layer and pooling layer to extract and classify features, so as to obtain the similarity results between the question information and multiple sets of question and answer data, and then sort the multiple sets of question and answer data according to the similarity results, select the set of question and answer data with the highest similarity as the target question and answer data, and input the set of target question and answer history data into the prompt word model. It should be understood that a set of question and answer history data may include one or more rounds of dialogue, so the target question and answer history data may contain only one round of dialogue or multiple rounds of dialogue, which can be set according to actual conditions.

[0145] In some embodiments, text data generated during the construction and operation and maintenance of the target data center can also be obtained; the text data can be split according to preset text splitting rules to obtain multiple sentences; based on the sentence contents corresponding to the multiple sentences, the question sentences corresponding to the multiple sentences can be respectively determined; a training set can be constructed using the multiple sentences and the corresponding question sentences, and a preset question-answering model can be trained using the training set to obtain a large target inspection model.

[0146] Specifically, the disclosed embodiment needs to build a large inspection model for the target data center. First, it is necessary to collect text data during the construction process and the operation and maintenance process of the target data center. The text data during the construction process of the target data center may include instruction documents such as computer room circuits, pipelines, networks, cold sources, inspection processes, and computer room cabinet layouts, and may also include other text files. The text data during the operation and maintenance of the target data center includes text data generated when the equipment in the data center is operating normally or during maintenance. The text data may also be content manually recorded by relevant inspection personnel.

[0147] In practical applications, when faced with text data, the text data can be segmented into different blocks according to paragraphs, and the meaningless parts, such as the main directory and headlines, can be filtered out. Then, the segmented text blocks can be segmented into multiple sentences according to punctuation marks, and the multiple sentences can be input into the zero-shot learning model for training. It should be understood that the zero-shot learning model can be full-task zero-shot learning-mT5: mT5 is a multilingual version of the T5 (Text-to-Text Transfer Transformer) model, which is designed to handle a variety of natural language processing tasks, including translation, summarization, question and answer, etc., through a unified text-to-text framework. mT5 uses a large amount of multilingual data during training, enabling it to understand and generate text in multiple languages.

[0148] For example, when the text data corresponding to the target data center contains the text content "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building. Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively", at this time, the text content can be segmented and assembled into full-task zero-sample learning-mT5 input text according to punctuation marks. Specifically, it can be segmented and assembled into two sentences, "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building" and "Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively". After segmentation and assembly, these two texts are input into full-task zero-sample learning-mT5, and questions such as "What parts does the data center consist of?" and "How many mainframe buildings are there?" can be obtained. Based on this, the question "What parts does the data center consist of?" and its corresponding answer "Our data center includes a mainframe building, a power machine building, an electric machine building, and a comprehensive office building.", and the question "How many mainframe buildings are there?" and its corresponding answer "Among them, there are three mainframe buildings, and the computer rooms are distributed in the mainframe buildings respectively".

[0149] In practical applications, the embodiments of the present disclosure can process the text data generated during the construction and operation and maintenance of the target data center in the above manner to obtain multiple questions and corresponding answer information. Based on this, a training set can be constructed based on multiple questions and corresponding answer information, and the preset question-answering model can be trained using the training set. At the same time, a target inspection large model is obtained through model fine-tuning. After fine-tuning, the target inspection large model can respond to the target data center questions. It should be understood that the preset question-answering model here can be any question-answering model in the prior art that can achieve the output of answers to input questions.

[0150] In some embodiments, the present disclosure can also obtain the current operating data corresponding to each device in the target data center; input the target prompt words and the current operating data into a pre-trained target inspection model, obtain reply information for the question information, and feed back the reply information to the inspection personnel.

[0151] In some embodiments, the target inspection big model of the disclosed embodiments is also combined with data center infrastructure management (DCIM), and the two are connected as a whole through an intelligent agent. Relying on the intention recognition capability of the big model, the topological cascade identifies the corresponding tools in the intelligent agent, and calls the tools to obtain DCIM operation data, so as to realize the target inspection big model's recognition of the operating status of each device or component in each computer room of the target data center, and give responses to the question information based on the current status.

[0152] Figure 3 The following is a schematic diagram of the structure of an inspection auxiliary question-answering system provided by an exemplary embodiment of the present disclosure. Figure 3 As shown, the patrol inspection auxiliary question-answering system of the embodiment of the present disclosure is designed with three main bodies, namely patrol inspectors 301, intelligent agents 302 and DCIM 303. Based on this, when patrol inspectors 301 ask questions, intelligent agents 302 analyze the question information corresponding to the question, and obtain the operating status of each device or equipment in the computer room at the current moment from DCIM, so that answers can be given in combination with the actual operating conditions, and the specific answer information results are displayed to the patrol inspectors. It should be understood that the relevant processing of the question information in the embodiment of the present disclosure is all executed in intelligent agents 302, that is, the target patrol large model and the model algorithm designed for the processing process of the question information are all set in intelligent agents 302.

[0153] In actual applications, when establishing a connection between DCIM and an intelligent agent, a hierarchical topology of an operation data interface can be established, and corresponding equipment groups can be established according to the relationship between the equipment. For example, the equipment group can be "cold source-water-cooled air conditioner-liquid cooling cabinet". During the detection process, each sub-equipment has a corresponding detection tool. At this time, a hierarchical topological relationship between different sub-equipment can be established, and the detection tools corresponding to multiple sub-equipment in the group of equipment can be regarded as a tool group. Then, during inspection, the current operation status of the equipment group can be determined according to the functional description (function-description) parameter value corresponding to the tool group. Based on this, the interface professional topology cascade method can be used to accurately locate the required tool group according to the professional type involved in the semantics when the inspection personnel ask questions, thereby improving the matching accuracy of the target inspection large model, so that the target inspection large model can respond based on the operating environment background, making the response more targeted.

[0154] For example, to establish a hierarchical topology for the operation data interface, you can create tool groups according to major disciplines, minor disciplines, and sub-models, for example: "Cold source - water-cooled air conditioner - liquid cooling cabinet" / "Cold source - air-cooled air conditioner - ordinary cabinet", that is, in the tool group, the function-description parameter value is "Query the real-time input cooling capacity according to the computer room number, professional level: cold source - water-cooled air conditioner - liquid cooling cabinet" / "Query the real-time input cooling capacity according to the computer room number, professional level: cold source - air-cooled air conditioner - ordinary cabinet".

[0155] Based on this, the disclosed embodiment uses unique text materials in the professional field of data centers as training materials to build a large inspection model that has a preliminary understanding of data center inspections; uses keyword vectorization and summary extraction technology to optimize and integrate external dynamic non-text knowledge bases, and ensures the effective recall of non-text knowledge by maximizing the comprehensive average recall rate algorithm; then uses intelligent agent technology to capture the operating parameters of the dynamic DCIM system in real time, so that the large inspection model can provide feedback in the context of considering the current system operating status, thereby creating a more intelligent question-answering method. In addition, we have constructed a question-answering history database, through word segmentation of question information, and matching based on the cosine similarity between the word segmentation and the preset instantiation keyword library, so as to construct a key-value database indexed by instantiation keywords, aiming to accelerate the information retrieval process and improve query accuracy; applies similarity matching algorithms to perform text similarity calculations and sorting algorithms, and efficiently filters out question-answering data generated in historical processes that are closely related to the current question from the huge question-answering database. It can be seen that the disclosed embodiment can make scientific answers based on the current data center operating status, non-text materials, question and answer historical data and other information. It not only ensures the objectivity and accuracy of the answers to questions during the inspection process, but also solves the problems of insufficient continuity and poor traceability of the inspection work.

[0156] The above mainly introduces the solution provided by the embodiment of the present disclosure from the perspective of the server. It is understandable that, in order to implement the above functions, the server includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0157] The embodiments of the present disclosure may divide the server into functional units according to the above method examples. For example, each functional module may be divided corresponding to each function, or two or more functions may be integrated into one management module. The above integrated modules may be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiments of the present disclosure is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0158] In the case of dividing each functional module according to each function, an exemplary embodiment of the present disclosure provides a data center inspection auxiliary question and answer device, which can be a server or a chip applied to a server. Figure 6 The following is a schematic diagram of a data center inspection auxiliary question-answering device provided by an exemplary embodiment of the present disclosure. Figure 6 As shown, the data center inspection auxiliary question-answering device 600 includes:

[0159] The acquisition module 601, an extraction module, is used to acquire question information from the inspection personnel and extract key information from the question information;

[0160] A first matching module 602 is used to match the key information with non-text data in a preset non-text database to obtain a non-text data matching result, and determine target non-text data from the preset non-text database according to the non-text data matching result;

[0161] A second matching module 603 is used to match the key information with the question and answer data in the preset question and answer history database to obtain a question and answer data matching result, and determine the target question and answer data from the preset question and answer history database according to the question and answer data matching result;

[0162] An acquisition module 604 is used to input the target non-text data and the target question and answer data into a preset prompt word generation model to obtain a target prompt word, and input the target prompt word into a pre-trained target inspection large model to obtain answer information for the question information;

[0163] The feedback module 605 is used to feed back the reply information to the inspection personnel.

[0164] In an optional manner, the key information includes a first keyword group and text description information, and the non-text data is multiple. The first matching module 602 is also used to match the first keyword group with the second keyword group included in the multiple non-text data, obtain the first non-text data matching results corresponding to the multiple non-text data, and determine multiple candidate keyword groups from the preset non-text database according to the first non-text data matching results; match the text description information with the non-text description information included in the multiple non-text data, obtain the second non-text data matching results corresponding to the multiple non-text data, and determine multiple candidate non-text description information from the preset non-text database according to the second non-text data matching results; in the case that the same content exists between the multiple candidate keyword groups and the multiple candidate non-text description information, determine the recall rate between the candidate keyword groups and the candidate non-text description information with the same content; determine the target keyword group and the target text description information from the candidate keyword group and the candidate non-text description information with the same content according to the recall rate, and determine the target non-text data according to the target keyword group and the target text description information.

[0165] In an optional manner, the non-text database is determined by non-text data generated during the construction process and operation and maintenance process of the target data center.

[0166] In an optional manner, there are multiple question and answer data, and the second matching module 603 is also used to match the key information with the multiple question and answer data respectively to obtain question and answer data matching results corresponding to the multiple question and answer data; when the question and answer data matching results corresponding to the multiple question and answer data are all greater than a preset threshold, the target question and answer data is determined from the multiple question and answer data based on the question and answer data matching results corresponding to the multiple question and answer data.

[0167] In an optional manner, the data center inspection auxiliary question and answer device 600 also includes a construction module 606, which is used to obtain text data generated during the construction process and operation and maintenance process of the target data center; split the text data according to preset text splitting rules to obtain multiple sentences; determine the question sentences corresponding to the multiple sentences according to the sentence contents corresponding to the multiple sentences; use the multiple sentences and the corresponding question sentences to construct a training set, and use the training set to train the preset question and answer model to obtain the target inspection large model.

[0168] In an optional manner, the data center inspection auxiliary question and answer device 600 also includes a reply module 607, and the reply module 607 is also used to obtain the current operating data corresponding to each device in the target data center; input the target prompt word and the current operating data into the pre-trained target inspection large model, obtain reply information for the question information, and feed back the reply information to the inspection personnel.

[0169] An embodiment of the present disclosure also provides an electronic device, comprising: at least one processor; a memory for storing instructions executable by at least one processor; wherein the at least one processor is used to execute instructions to implement the steps of the above method disclosed in the embodiment of the present disclosure.

[0170] Figure 7 The structure diagram of an electronic device provided by an exemplary embodiment of the present disclosure is shown in FIG. Figure 7 As shown, the electronic device 700 includes at least one processor 701 and a memory 702 coupled to the processor 701 , and the processor 701 can execute corresponding steps in the above method disclosed in the embodiment of the present disclosure.

[0171] The processor 701 may also be referred to as a central processing unit (CPU), which may be an integrated circuit chip having signal processing capabilities. Each step in the method disclosed in the embodiment of the present disclosure may be completed by an integrated logic circuit of hardware in the processor 701 or by instructions in the form of software. The processor 701 may be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present disclosure may be directly embodied as being executed by a hardware decoding processor, or may be executed by a combination of hardware and software modules in a decoding processor. The software module may be located in a memory 702, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, or other mature storage media in the art. The processor 701 reads the information in the memory 702 and completes the steps of the method in combination with its hardware.

[0172] In addition, when various operations / processes according to the present disclosure are implemented by software and / or firmware, they can be transmitted from a storage medium or a network to a computer system having a dedicated hardware structure, for example, Figure 8The computer system 800 shown installs the programs constituting the software, and when the various programs are installed, the computer system can execute various functions, including the functions described above. Figure 8 A schematic diagram of the structure of a computer system provided by an exemplary embodiment of the present disclosure.

[0173] Computer system 800 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0174] like Figure 8 As shown, the computer system 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0175] A plurality of components in the computer system 800 are connected to the I / O interface 805, including: an input unit 806, an output unit 807, a storage unit 808, and a communication unit 809. The input unit 806 can be any type of device capable of inputting information to the computer system 800, and the input unit 806 can receive input digital or character information, and generate key signal inputs related to the inspection personnel settings and / or function control of the electronic device. The output unit 807 can be any type of device capable of presenting information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 808 can include but is not limited to a disk, an optical disk. The communication unit 809 allows the computer system 800 to exchange information / data with other devices over a network such as the Internet, and can include but is not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, for example, a Bluetooth™ device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0176] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above. For example, in some embodiments, the above methods disclosed in the embodiments of the present disclosure may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 802 and / or a communication unit 809. In some embodiments, the computing unit 801 may be configured to perform the above methods disclosed in the embodiments of the present disclosure by any other appropriate means (e.g., by means of firmware).

[0177] The embodiment of the present disclosure also provides a computer-readable storage medium, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the above method disclosed in the embodiment of the present disclosure.

[0178] The computer-readable storage medium in the disclosed embodiments may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. More specifically, the computer-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0179] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0180] Fig. 9 Schematic diagram of a computer program product provided by an exemplary embodiment of the present disclosure. Fig. 9 As shown, the computer program product 900 includes a computer program 901, wherein the computer program 901 implements the above method disclosed in the embodiment of the present disclosure when executed by a processor.

[0181] In an embodiment of the present disclosure, computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code may be executed entirely on the patrol personnel computer, partially on the patrol personnel computer, as a separate software package, partially on the patrol personnel computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the patrol personnel computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer.

[0182] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0183] The modules, components or units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the names of the modules, components or units do not, in some cases, limit the modules, components or units themselves.

[0184] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0185] The above descriptions are only some embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, a technical solution formed by replacing the above features with the technical features with similar functions disclosed in the present disclosure (but not limited to).

[0186] Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A data center inspection auxiliary question-answering method, characterized in that: include: Obtaining question information from inspection personnel and extracting key information from the question information; Matching the key information with non-text data in a preset non-text database to obtain a non-text data matching result, and determining target non-text data from the preset non-text database according to the non-text data matching result; Matching the key information with question and answer data in a preset question and answer history database to obtain a question and answer data matching result, and determining target question and answer data from the preset question and answer history database according to the question and answer data matching result; Input the target non-text data and the target question-answer data into a preset prompt word generation model to obtain a target prompt word, and input the target prompt word into a pre-trained target inspection large model to obtain answer information for the question information; The reply information is fed back to the inspection personnel.

2. The method according to claim 1, characterized in that The key information includes a first keyword group and text description information, the non-text data is multiple, and the key information is matched with the non-text data in a preset non-text database to obtain a non-text data matching result, and the target non-text data is determined from the preset non-text database according to the non-text data matching result, including: Matching the first keyword group with the second keyword group included in the plurality of non-text data respectively, obtaining first non-text data matching results corresponding to the plurality of non-text data, and determining a plurality of candidate keyword groups from the preset non-text database according to the first non-text data matching results; Matching the text description information with the non-text description information included in the plurality of non-text data respectively to obtain second non-text data matching results corresponding to the plurality of non-text data, and determining a plurality of candidate non-text description information from the preset non-text database according to the second non-text data matching results; In the case where the same content exists between a plurality of the candidate keyword groups and a plurality of the candidate non-text description information, determining the recall rate between the candidate keyword groups and the candidate non-text description information having the same content; A target keyword group and target text description information are determined from the candidate keyword group and the candidate non-text description information having the same content according to the recall rate, and the target non-text data are determined according to the target keyword group and the target text description information.

3. The method according to claim 2, characterized in that The non-text database is determined by non-text data generated during the construction and operation and maintenance of the target data center.

4. The method according to claim 1, characterized in that There are multiple question and answer data, and matching the key information with the question and answer data in a preset question and answer history database to obtain a question and answer data matching result, and determining target question and answer data from the preset question and answer history database according to the question and answer data matching result, includes: Matching the key information with the plurality of question and answer data respectively to obtain question and answer data matching results corresponding to the plurality of question and answer data; When the question and answer data matching results corresponding to the multiple question and answer data are all greater than a preset threshold, the target question and answer data is determined from the multiple question and answer data according to the question and answer data matching results corresponding to the multiple question and answer data.

5. The method according to claim 1, characterized in that The method further comprises: Acquire text data generated during the construction and operation and maintenance of the target data center; Splitting the text data according to a preset text splitting rule to obtain multiple sentences; Determining question statements corresponding to the plurality of statements respectively according to the statement contents corresponding to the plurality of statements; A training set is constructed using a plurality of the statements and the corresponding question statements, and a preset question-answering model is trained using the training set to obtain the target inspection large model.

6. The method according to claim 1, characterized in that The method further comprises: Obtain the current operating data corresponding to each device in the target data center; The target prompt word and the current operation data are input into the pre-trained target inspection model to obtain reply information for the question information, and the reply information is fed back to the inspection personnel.

7. A data center inspection auxiliary question-answering device, characterized in that: include: An extraction module, used to obtain question information from inspection personnel and extract key information from the question information; A first matching module, used for matching the key information with non-text data in a preset non-text database to obtain a non-text data matching result, and determining target non-text data from the preset non-text database according to the non-text data matching result; A second matching module is used to match the key information with the question and answer data in the preset question and answer history database to obtain a question and answer data matching result, and determine the target question and answer data from the preset question and answer history database according to the question and answer data matching result; An acquisition module, used for inputting the target non-text data and the target question and answer data into a preset prompt word generation model to obtain a target prompt word, and inputting the target prompt word into a pre-trained target inspection large model to obtain answer information for the question information; A feedback module is used to feed back the reply information to the inspection personnel.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images