Operation and maintenance work order automatic classification method and system based on large model technology

Abstract

The invention provides an operation and maintenance work order automatic classification method and system based on a large model technology, and relates to the field of data processing. The method comprises the following steps: obtaining operation and maintenance demand description information; processing the operation and maintenance demand description information, and extracting a demand abstract; the processing comprises performing abstract extraction through a prompt word engineering method based on a preset prompt word; inputting the demand abstract into a pre-deployed work order classification model, and outputting to obtain a work order category corresponding to the operation and maintenance demand description information; the work order categories comprise a first category and a second category; the first category comprises a technical support demand and a fault repair demand; the second category comprises a desktop service demand, a conference guarantee demand, a visiting guarantee demand, a question answering demand and a collaborative support demand; and filling the work order category and the demand abstract into a work order template, creating an operation and maintenance work order, and completing automatic classification of the operation and maintenance work order.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to an operation and maintenance work order automatic classification method and system based on large model technology. BACKGROUND

[0002] In the road network guarantee center-operation and maintenance work order management system, after collecting the user operation and maintenance guarantee demand through telephone or text, work order classification is a key step to ensure that the user demand is quickly and accurately allocated to the corresponding processing team. Work order classification usually relies on manual classification after listening or reading the operation and maintenance demand description information, which not only consumes time and effort, but also is easily influenced by personal experience, leading to inconsistent classification.

[0003] In recent years, with the development of natural language processing technology, some automatic classification systems based on keyword rules or machine learning models are used for work order classification, but these systems are often limited by the generalization ability of pre-set keyword rules or models, and have low flexibility; and for complex and variable operation and maintenance demand description information, the language understanding ability is insufficient, and the processing effect is not good, especially when facing unstructured text, misclassification is easy to occur. In addition, these systems usually need a large amount of labeled data for training, and the maintenance cost is high.

[0004] In the prior art, such as Chinese invention patent with publication number CN119090219A, a large model-based operation and maintenance work order rapid creation method and system is disclosed, the method specifically as follows: interface packaging: taking the problem description, business classification, problem level, work order source and work order title of historical operation and maintenance work order as the fine-tuning parameters of the large model, packaging the large model inference interface through model pre-training; creating work order: inputting the work order problem description on the page, returning the work order title, business classification, work order source and problem level necessary parameter information of work order creation through the large model inference interface and directly filling them into the front-end creation page input box, and allowing the work order creator to adjust the recommended work order parameters, and the created work order is used as the parameter for subsequent large model fine-tuning. However, the invention patent does not disclose how to use the extracted demand summary to participate in the fine-tuning process of the classification model. SUMMARY

[0005] To solve the technical problems existing in the prior art, the purpose of the present application is to provide an operation and maintenance work order automatic classification method and system based on large model technology, which can more accurately automatically extract work order demand, improve work order classification accuracy, and enhance the generalization ability of the model. When training the work order classification model, the dependence on labeled data is greatly reduced, and an efficient parameter fine-tuning method is adopted, so that the model training cost is lower and the efficiency is higher.

[0006] To achieve the above-mentioned purposes, the application provides an operation and maintenance work order automatic classification method based on large model technology, and the steps are as follows:

[0007] Obtain operation and maintenance requirement description information;

[0008] Process the operation and maintenance requirement description information and extract a requirement summary;

[0009] The processing includes extracting the summary based on a preset prompt word through a prompt word engineering method;

[0010] Input the requirement summary into a pre-deployed work order classification model to output a work order category corresponding to the operation and maintenance requirement description information;

[0011] The work order category includes a first category and a second category;

[0012] The first category includes technical support requirements and fault repair requirements; and the second category includes desktop business requirements, meeting guarantee requirements, visit guarantee requirements, question and answer requirements, and collaborative support requirements;

[0013] Fill the work order category and the requirement summary into a work order template to create an operation and maintenance work order and complete the automatic classification of the operation and maintenance work order.

[0014] According to one technical solution of the application, the process of realizing the work order classification model includes:

[0015] Obtain a classification data set;

[0016] The classification data set includes a requirement summary obtained through historical operation and maintenance requirement description information and a work order category corresponding to the requirement summary;

[0017] Fine-tune a pre-trained general model using the classification data set and update the weight of the pre-trained general model to obtain the work order classification model;

[0018] Deploy the work order classification model to a local environment.

[0019] According to one technical solution of the application, fine-tune a pre-trained general model using the classification data set and update the weight of the pre-trained general model, and the process is as follows:

[0020] Fine-tune the pre-trained general model based on the LoRA method using the classification data set to obtain the weight of a low-rank adapter;

[0021] Combine the weight of the low-rank adapter with the original weight of the pre-trained general model to obtain the updated weight;

[0022] updating the weights of the pre-trained general model using the updated weights.

[0023] According to one of the technical solutions of the present application, based on the preset prompt word, the abstract extraction is carried out through the prompt word engineering method, and the specific process includes:

[0024] The operation and maintenance requirement description information and the preset prompt word are input into the pre-deployed abstract extraction model at the same time, and the output requirement abstract is obtained.

[0025] According to one of the technical solutions of the present application, the general model and the abstract extraction model are both open source large language models.

[0026] The present application also provides an operation and maintenance work order automatic classification system based on large model technology, comprising:

[0027] An operation and maintenance requirement description information acquisition module is used to obtain operation and maintenance requirement description information.

[0028] An abstract extraction model is used to process the operation and maintenance requirement description information and extract a requirement abstract.

[0029] The processing includes abstract extraction through the prompt word engineering method based on the preset prompt word.

[0030] A work order classification model is used to output a work order category corresponding to the operation and maintenance requirement description information according to the input requirement abstract.

[0031] The work order category includes a first category and a second category.

[0032] The first category includes technical support requirements and fault repair requirements, and the second category includes desktop business requirements, meeting guarantee requirements, visit guarantee requirements, question and answer requirements and collaborative support requirements.

[0033] A work order creation module is used to fill the work order category and the requirement abstract into a work order template and create an operation and maintenance work order.

[0034] According to one of the technical solutions of the present application, the operation and maintenance work order automatic classification system based on large model technology further comprises a module for implementing the work order classification model.

[0035] The module for implementing the work order classification model includes:

[0036] A data collection module is used to obtain a classification data set.

[0037] The classification data set includes a requirement abstract obtained through historical operation and maintenance requirement description information and a work order category corresponding to the requirement abstract.

[0038] a fine-tuning module, configured to fine-tune a pre-trained general model using the classification data set and update weights of the pre-trained general model to obtain the work order classification model;

[0039] a deployment module, configured to deploy the work order classification model to a local environment.

[0040] According to one of the technical solutions of the present application, the fine-tuning module is further configured to:

[0041] fine-tune the pre-trained general model based on a LoRA method using the classification data set to obtain weights of a low-rank adapter;

[0042] merge the weights of the low-rank adapter with original weights of the pre-trained general model to obtain updated weights;

[0043] update the weights of the pre-trained general model using the updated weights.

[0044] According to one of the technical solutions of the present application, the summary extraction model is further configured to input the operation and maintenance requirement description information and the preset prompt word into a pre-deployed summary extraction model to obtain an output requirement summary.

[0045] According to one of the technical solutions of the present application, the general model and the summary extraction model are both open source large language models.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] 1. Through the powerful text understanding and generation capability of the large model, combined with the customized prompt word engineering and fine-tuning technology, the work order requirement can be automatically extracted more accurately, thereby avoiding the operation of manually preparing keyword rules.

[0048] 2. The dependence on labeled data is greatly reduced, and at the same time, an efficient parameter fine-tuning method is adopted, only part of the model parameters are trained, so that the model training cost is lower and the efficiency is higher.

[0049] 3. The natural language understanding and reasoning ability of the large model is utilized, even if it faces complex and unstructured user requirements, the user requirement summary can be accurately extracted through the prompt word engineering technology, thereby improving the work order classification accuracy and enhancing the generalization ability of the model.

[0050] 4. The automatic classification significantly shortens the pre-processing time of work order processing, so that the operation and maintenance team can respond and solve problems faster, and the overall operation and maintenance efficiency and service quality are improved. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description only show some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0052] Figure 1 A flowchart schematically showing an automatic operation order classification method based on large model technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0053] The description of the embodiments of the present application should be combined with the corresponding drawings, which should be regarded as a part of the complete description. In the drawings, the shape or thickness of the embodiments can be exaggerated and simplified or convenient for illustration. Moreover, the parts of the structures in the drawings will be described separately, and it should be noted that the elements not shown or not described in the drawings are in the form known by those skilled in the art.

[0054] The description of the embodiments herein, any reference to direction or position, is only for the convenience of description, and cannot be understood as any limitation on the scope of protection of the present application. The following description of the preferred embodiments will involve combinations of features, which can exist independently or in combination, and the present application is not particularly limited to the preferred embodiments. The scope of the present application is defined by the claims.

[0055] As shown in Figure 1 , an automatic operation order classification method based on large model technology of the present application, the steps are as follows:

[0056] S1, obtaining operation demand description information;

[0057] S2, processing the operation demand description information and extracting the demand abstract;

[0058] The processing includes extracting the abstract based on the preset prompt words through the prompt word engineering method;

[0059] S3, inputting the demand abstract into the pre-deployed order classification model, and outputting the order category corresponding to the operation demand description information;

[0060] The order category includes a first category and a second category;

[0061] The first category includes technical support demand and fault repair demand; the second category includes desktop business demand, meeting guarantee demand, visit guarantee demand, question and answer demand and collaborative support demand;

[0062] S4, fill the ticket category and demand summary into the ticket template to create an operation and maintenance ticket, and complete the automatic classification of the operation and maintenance ticket.

[0063] In this embodiment, the operation and maintenance ticket automatic classification method based on large model technology includes a training stage and an inference stage. The present embodiment is the inference stage of ticket classification based on a pre-deployed ticket classification model, mainly including:

[0064] S1, obtaining operation and maintenance demand description information;

[0065] Collect the operation and maintenance demand description information of user operation and maintenance guarantee. The operation and maintenance demand description information is single-round or multi-round dialogue text.

[0066] S2, processing the operation and maintenance demand description information and extracting the demand summary;

[0067] Through the large model prompt word engineering technology, the operation and maintenance demand description information is summarized and extracted, the core demand content is retained, and the demand summary text is formed.

[0068] S3, inputting the demand summary into the pre-deployed ticket classification model to output the ticket category corresponding to the operation and maintenance demand description information;

[0069] The demand summary text is input into the ticket classification model (text classification large model) generated in the training stage for inference, and the ticket category is output;

[0070] The ticket category includes first-level classification including technical support, fault repair, and other categories; the second-level classification of technical support includes desktop business, meeting guarantee, visit guarantee, question and answer, and cooperative support.

[0071] S4, filling the ticket category and demand summary into the ticket template to create an operation and maintenance ticket;

[0072] The ticket category, demand summary and other business required fields are automatically filled into the ticket system.

[0073] This step can directly complete the creation of the operation and maintenance ticket, or the following manual review step can be added.

[0074] S5, manually review the pre-filled ticket fields, click submit after accurate and correct, and complete the creation of the operation and maintenance ticket.

[0075] In the operation and maintenance ticket automatic classification method based on large model technology, the process of realizing the ticket classification model includes:

[0076] Obtain the classification data set;

[0077] The classification data set includes a demand summary obtained from historical operation and maintenance demand description information, and a work order category corresponding to the demand summary.

[0078] The pre-trained general model is fine-tuned using the classification data set, and the weights of the pre-trained general model are updated to obtain a work order classification model.

[0079] The work order classification model is deployed to the local environment.

[0080] In this embodiment, the embodiment is a stage of training and deploying the work order classification model, mainly including:

[0081] First, obtain the classification data set;

[0082] The classification data set includes a demand summary obtained from historical operation and maintenance demand description information, and a work order category corresponding to the demand summary:

[0083] Specifically, first, collect the operation and maintenance demand description information (historical operation and maintenance demand description information) of user operation and maintenance guarantee, which is single-round or multi-round dialogue text (such as: Hello, January 1, 2025 14:00-17:00, a delegation of a certain company will visit the dispatch center hall, the number of people is about 10, please provide conference support. ), and classify and label according to the category of the operation and maintenance demand description information.

[0084] For example, the first-level classification includes technical support, network security, fault repair, and others; among them, the second-level classification of technical support includes desktop business, conference support, visit support, question and answer, and cooperation support, forming a (operation and maintenance demand description information, category) data set.

[0085] However, because the second-level classification can be uniquely attributed to the first-level classification, the work order category in the data set constructed here is the second-level classification, which is used for subsequent fine-tuning training of the work order classification model.

[0086] Then, the demand summary is extracted:

[0087] The operation and maintenance demand description information is summarized and extracted by the large model prompt word engineering technology, the core demand content is retained, the demand summary is extracted, and then a classification data set including the demand summary and the work order category is formed, which is used for subsequent fine-tuning training of the work order classification model.

[0088] Among them, the demand summary is used to construct the data set, rather than the operation and maintenance demand description information, because the operation and maintenance demand description information contains a large amount of useless redundant information, and after a step of summary extraction, the core information of the demand can be retained, and it is also helpful to the more accurate training and reasoning of the subsequent work order classification model.

[0089] The above summary summary prompt word writing can be as follows:

[0090] """Instruction:

[0091] You are a text summarization expert. Please read the following user input carefully and summarize it:\n

[0092] {text}\n

[0093] Requirements:\n

[0094] 1. Please retain the core information of the given text and do not delete the core information;\n

[0095] 2. Please only summarize the given text and do not add other irrelevant information;\n

[0096] 3. The number of words should be within 30. \n

[0097] Response: """

[0098] Second, we fine-tune the pre-trained general model using the classification dataset and update its weights to obtain a work order classification model.

[0099] Specifically, a large classification model is trained.

[0100] First, the classification dataset is constructed into a JSON format that can be recognized by fine-tuning. Conversations represent a pair of (demand summary, category). For example, the demand summary is "10108 Computer cannot connect to the Internet, request processing." and the category is "Desktop business".

[0101] Then, select a general large language model, set fine-tuning parameters, and start efficient parameter fine-tuning. The core parameters include the original weight address of the general large language model, the address of the JSON data, the address of the fine-tuned adapter weights, the number of training rounds, the training batch size, the test batch size, and the learning rate. The specific training parameter settings are as follows:

[0102] python finetune.py\

[0103] --model_name_or_path"The original weight address of the general large language model"\

[0104] --data_path "json format data address"\

[0105] --bf16 True\

[0106] --output_dir "fine-tuned adapter weight address"\

[0107] --num_train_epochs 5

[0108] --per_device_train_batch_size 2

[0109] --per_device_eval_batch_size 1

[0110] --gradient_accumulation_steps 8

[0111] --evaluation_strategy "no"

[0112] --save_strategy "steps"

[0113] --save_steps 1000

[0114] --save_total_limit 10

[0115] --learning_rate 3e-4

[0116] --weight_decay 0.1

[0117] --adam_beta1 0.95

[0118] --warmup_ratio 0.01

[0119] --lr_scheduler_type "cosine"

[0120] --logging_steps 1

[0121] --report_to "none"

[0122] --model_max_length 512

[0123] --lazy_preprocess True

[0124] --gradient_checkpointing

[0125] --use_lora

[0126] Finally, the low-rank adapter weights are merged with the original weights of the general large language model, the from_pretrained(), merge_and_unload(), and save_pretrained() functions of the model=AutoPeftModelForCausalLM are used to load the low-rank adapter weights, merge the original weights and the low-rank adapter, and save the weights of the obtained work order classification model to a directory. The from_pretrained() and save_pretrained() functions of AutoTokenizer are used to copy the tokenizer to the directory of the work order classification model weights. At this time, the weights of the work order classification model have the ability to classify demand abstract texts, forming a text classification professional large model.

[0127] In the operation and maintenance work order automatic classification method based on large model technology, the pre-trained general model is fine-tuned using the classification data set, and the weights of the pre-trained general model are updated, as follows:

[0128] The pre-trained general model is fine-tuned based on the LoRA method using the classification data set to obtain the weights of the low-rank adapter.

[0129] The weights of the low-rank adapter are merged with the original weights of the pre-trained general model to obtain updated weights.

[0130] The weights of the pre-trained general model are updated using the updated weights.

[0131] In the operation and maintenance work order automatic classification method based on large model technology, based on the preset prompt words, the prompt word engineering method is used for abstract extraction, and the specific process includes:

[0132] The operation and maintenance demand description information and the preset prompt words are input into the pre-deployed abstract extraction model to obtain the output demand abstract.

[0133] In the inference stage of the present embodiment, the operation and maintenance demand description information is obtained as single-turn or multi-turn dialogue text through the user demand input interface. Then, the operation and maintenance demand description information is abstracted by the large model prompt word engineering technology, the core demand content is retained, and the demand abstract is extracted. Then, the demand abstract is input into the work order classification model to obtain the work order category.

[0134] In the operation and maintenance work order automatic classification method based on large model technology, the general model and the abstract extraction model are both open source large language models.

[0135] In the present embodiment, the general model and the abstract extraction model can be selected from the qwen2.5 series or the qwen3 series 8B and less small parameter models. Such models have fewer parameters, fast training and inference, and are easy to deploy, which is suitable for environments with limited resources.

[0136] The model verifies the effect of the work order classification as follows:

[0137] Based on 5000 pieces of classification data (training set: test set = 4:1, a total of 5 categories, 1000 pieces of data per category), the general large model with a parameter of 7B is fine-tuned on the training set, and after fine-tuning training is completed, the evaluation is performed on the test set, and the classification accuracy can reach 100%.

[0138] The operation and maintenance work order automatic classification system based on large model technology comprises:

[0139] An operation and maintenance demand description information acquisition module is configured to obtain operation and maintenance demand description information.

[0140] An abstract extraction model is configured to process the operation and maintenance demand description information and extract a demand abstract.

[0141] The processing includes abstract extraction based on a preset prompt word through a prompt word engineering method.

[0142] A work order classification model is configured to output a work order category corresponding to the operation and maintenance demand description information according to an input demand abstract.

[0143] The work order category comprises a first category and a second category.

[0144] The first category comprises technical support demand and fault repair demand, and the second category comprises desktop business demand, meeting guarantee demand, visit guarantee demand, question and answer demand, and collaborative support demand.

[0145] A work order creation module is configured to fill the work order category and the demand abstract into a work order template to create an operation and maintenance work order.

[0146] The operation and maintenance work order automatic classification system based on large model technology further comprises a module for implementing the work order classification model.

[0147] The module for implementing the work order classification model comprises:

[0148] A data collection module is configured to obtain a classification data set.

[0149] The classification data set comprises a demand abstract obtained from historical operation and maintenance demand description information and a work order category corresponding to the demand abstract.

[0150] A fine-tuning module is configured to fine-tune a pre-trained general model using the classification data set and update the weight of the pre-trained general model to obtain the work order classification model.

[0151] A deployment module is configured to deploy the work order classification model to a local environment.

[0152] The fine-tuning module is also used for:

[0153] The low-rank adapter weight is obtained by fine-tuning the pre-trained general model based on the LoRA method using the classification data set.

[0154] The low-rank adapter weight is obtained by fine-tuning the pre-trained general model based on the LoRA method using the classification data set.

[0155] The low-rank adapter weight is obtained by fine-tuning the pre-trained general model based on the LoRA method using the classification data set.

[0156] The abstract extraction model is also used for inputting the operation and maintenance requirement description information and the preset prompt word into the pre-deployed abstract extraction model at the same time to obtain the output requirement abstract.

[0157] In the operation and maintenance order automatic classification system based on the large model technology, the general model and the abstract extraction model are open source large language models.

[0158] The operation and maintenance order automatic classification method and system based on the large model technology of the application have the following method steps: obtaining operation and maintenance requirement description information; processing the operation and maintenance requirement description information to extract a requirement abstract; the processing includes abstract extraction by a prompt word engineering method based on a preset prompt word; inputting the requirement abstract into a pre-deployed order classification model to output an order category corresponding to the operation and maintenance requirement description information; the order category includes a first category and a second category; the first category includes technical support requirements and fault repair requirements; the second category includes desktop business requirements, meeting guarantee requirements, visit guarantee requirements, question and answer requirements and collaborative support requirements; filling the order category and the requirement abstract into an order template to create an operation and maintenance order and complete the automatic classification of the operation and maintenance order.

[0159] In addition, it should be noted that the application can be provided as a method, device or computer program product. Therefore, the embodiments of the application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the application can adopt the form of a computer program product implemented on one or more computer usable storage media containing computer usable program code.

[0160] The embodiments of the present application are described with reference to the flowchart illustrations and / or block diagrams of the methods, terminal devices (systems) and computer program products according to the embodiments of the present application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing terminal devices to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal devices, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0161] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal devices to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the function specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0162] It should also be noted that, in the present document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a list of elements does not only include those elements, but also other elements not explicitly listed, or other elements inherent to such process, method, article or terminal device. Without more limitations, an element defined by the statement "including a..." does not exclude the presence of additional identical elements in the process, method, article or terminal device including the said element.

[0163] Finally, it should be noted that the above description is of preferred embodiments of the application, and that although preferred embodiments of the application have been described, modifications and adaptations taken within the spirit of the application as described are to be considered as within the scope of the application. Therefore, the following claims are intended to cover all such modifications and adaptations that come within the scope of the embodiments of the application.

Claims

1. A method for automatic classification of operation and maintenance work orders based on large model technology, characterized in that: Here are the steps: Obtain operation and maintenance requirement description information; Processing the operation and maintenance requirement description information to extract the requirement summary; The processing includes extracting a summary based on preset prompt words through a prompt word engineering method; Input the requirement summary into a pre-deployed work order classification model, and output the work order category corresponding to the operation and maintenance requirement description information; The work order categories include a first category and a second category; The first category includes technical support needs and fault repair needs; the second category includes desktop business needs, conference support needs, visit support needs, question answering needs and collaborative support needs; Fill the work order category and requirement summary into the work order template, create an operation and maintenance work order, and complete the automatic classification of the operation and maintenance work order.

2. The automatic classification method for operation and maintenance work orders based on large model technology according to claim 1 is characterized in that: The process of implementing the work order classification model includes: Get a classification dataset; The classification data set includes a demand summary obtained through historical operation and maintenance demand description information, and a work order category corresponding to the demand summary; Fine-tuning the pre-trained general model using the classification data set and updating the weight of the pre-trained general model to obtain the work order classification model; Deploy the work order classification model to the local environment.

3. The automatic classification method for operation and maintenance work orders based on large model technology according to claim 2 is characterized in that: Fine-tune the pre-trained general model using the classification dataset and update the weights of the pre-trained general model, the process is as follows; Using the classification dataset, fine-tuning the pre-trained general model based on the LoRA method to obtain the weights of the low-rank adapter; Merging the weights of the low-rank adapter with the original weights of the pre-trained general model to obtain updated weights; The weights of the pre-trained general model are updated using the updated weights.

4. The automatic classification method for operation and maintenance work orders based on large model technology according to claim 2 or 3 is characterized in that: Based on the preset prompt words, the summary is extracted through the prompt word engineering method. The specific process includes: The operation and maintenance requirement description information and the preset prompt words are simultaneously input into the pre-deployed summary extraction model to obtain the output requirement summary.

5. The automatic classification method for operation and maintenance work orders based on large model technology according to claim 4 is characterized in that: Both the general model and the summary extraction model are open source large language models.

6. An automatic classification system for operation and maintenance work orders based on large model technology, characterized by: include: Operation and maintenance requirement description information collection module, used to obtain operation and maintenance requirement description information; A summary extraction model is used to process the operation and maintenance requirement description information and extract the requirement summary; The processing includes extracting a summary based on preset prompt words through a prompt word engineering method; A work order classification model is used to output a work order category corresponding to the operation and maintenance requirement description information based on the input requirement summary; The work order categories include a first category and a second category; The first category includes technical support needs and fault repair needs; the second category includes desktop business needs, conference support needs, visit support needs, question answering needs and collaborative support needs; The work order creation module is used to fill the work order category and requirement summary into the work order template to create an operation and maintenance work order.

7. The automatic classification system for operation and maintenance work orders based on large model technology according to claim 6 is characterized in that: Also included is a module for implementing the work order classification model; The module for implementing the work order classification model includes: Data collection module, used to obtain classification data sets; The classification data set includes a demand summary obtained through historical operation and maintenance demand description information, and a work order category corresponding to the demand summary; A fine-tuning module, configured to fine-tune the pre-trained general model using the classification data set and update the weights of the pre-trained general model to obtain the work order classification model; A deployment module is used to deploy the work order classification model to a local environment.

8. The automatic classification system for operation and maintenance work orders based on large model technology according to claim 7 is characterized in that: The fine-tuning module is further used to: Using the classification dataset, fine-tuning the pre-trained general model based on the LoRA method to obtain the weights of the low-rank adapter; The weights of the low-rank adapter are then merged with the original weights of the pre-trained general model to obtain updated weights; The updated weights are then used to update the weights of the pre-trained general model.

9. The automatic classification system for operation and maintenance work orders based on large model technology according to claim 7 or 8 is characterized in that: The summary extraction model is also used to input the operation and maintenance requirement description information and preset prompt words into the pre-deployed summary extraction model at the same time to obtain the output requirement summary.

10. The automatic classification system for operation and maintenance work orders based on large model technology according to claim 9 is characterized in that: Both the general model and the summary extraction model are open source large language models.

Images