Intelligent order building method and system based on operation and maintenance industry large model

Abstract

The invention relates to the technical field of large models, in particular to an intelligent order building method and system based on an operation and maintenance industry large model, and the method comprises the following steps: collecting a work order list of a work order system; cleaning, integrating and standardizing the collected work order list to form a data set required by fine tuning of the model; selecting a pre-training model, performing fine tuning training on the pre-training model by using the data set, and outputting a fine-tuned model; executing a task of predicting work order classification by using the fine-tuned model; operation and maintenance groups and operation and maintenance personnel in the region are automatically identified through a model classification result; calling a large model for reasoning and returning a work order title; realizing a system integration process; the method and the system have the beneficial effects that the method and the system for intelligent order establishment based on the operation and maintenance industry large model, provided by the invention, realize automatic order establishment of links such as fault repair and equipment inspection in the operation and maintenance industry by applying big data and artificial intelligence technologies, and greatly improve the working efficiency.

Description

Technical Field

[0001] The present invention relates to the field of big model technology, and specifically to a method and system for intelligent order creation based on a big model of the operation and maintenance industry. Background Art

[0002] With the rapid development of information technology, the role of the operation and maintenance industry has become increasingly prominent. Operation and maintenance engineers are responsible for monitoring and managing IT infrastructure to ensure the stability, security and efficiency of the system. However, as the complexity of IT systems continues to increase, the workload of operation and maintenance is also increasing, and the requirements for operation and maintenance personnel are also getting higher and higher. Therefore, how to use big data and artificial intelligence technology to achieve automated and intelligent establishment of operation and maintenance work orders has become an urgent problem to be solved in the current operation and maintenance industry.

[0003] The traditional way of establishing operation and maintenance work orders mainly relies on manual work, which requires operation and maintenance engineers to manually collect fault information, analyze the cause of the problem, develop solutions, and establish work orders. This method has the following problems: First, it is inefficient, especially when facing a large number of operation and maintenance problems, it takes a lot of time and manpower; second, it is prone to errors, and omissions and errors are inevitable during manual processing; third, there is a lack of standardization, and different engineers may have different processing methods and standards, resulting in uneven service quality.

[0004] In order to solve the above problems, we began to explore the use of big data and artificial intelligence technology to realize the automation and intelligent establishment of operation and maintenance work orders. At present, some researchers and companies have achieved certain results in this regard. For example, some companies use machine learning algorithms to mine and analyze operation and maintenance data, and realize the automatic identification and processing of operation and maintenance problems. However, these methods generally have the following shortcomings: First, the model accuracy is low, especially for the identification and processing of complex operation and maintenance scenarios; second, the model generalization ability is poor, and it is difficult to adapt to operation and maintenance scenarios of different types and scales; third, there is a lack of in-depth exploration of the characteristics of the operation and maintenance industry and targeted modeling. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for intelligent order creation based on a large model of the operation and maintenance industry to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for intelligent order creation based on a large model of the operation and maintenance industry, the method comprising the following steps:

[0007] Step 1: Collect the work order list of the work order system;

[0008] Step 2: Clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning;

[0009] Step 3: Select a pre-trained model, fine-tune it using the data set, and output the fine-tuned model;

[0010] Step 4: Use the fine-tuned model to perform the task of predicting the work order classification;

[0011] Step 5: Automatically identify the operation and maintenance group and personnel in the area through the model classification results;

[0012] Step 6: Call the large model for inference and return the work order title;

[0013] Step 7: Implement the system integration process.

[0014] Preferably, the specific operations of step 1 include:

[0015] Realize interaction with the work order system, develop a RESTful API interface, and obtain data from the work order system. The collected data is a list of work orders in the past three years, including work order title, work order problem description, work order category name, operation and maintenance project team, and operation and maintenance engineer name. The API interface returns json as the data exchange format.

[0016] Preferably, the specific operations of step 2 include:

[0017] Delete duplicate work orders. There may be duplicate records in the data set, which may lead to deviations in the analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order is ensured. Process irrelevant characters. The data set contains irrelevant characters such as line breaks and tabs. These characters have no practical significance for data analysis. Deleting these irrelevant characters improves the neatness and usability of the data set. Complete classification information. The work order classification name in the data set is not obtained because of errors in the data collection process or problems with the data source itself. In this case, complete the incomplete classification information to ensure the classification accuracy and consistency of the data set.

[0018] Preferably, step 3 specifically includes:

[0019] 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed;

[0020] 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, which is 442.

[0021] 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6;

[0022] 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints;

[0023] 3.5 Divide the data set organized in step 2, using 80% of the sample data as the training set and 20% of the sample data as the test set, taking into account the time distribution of the samples. Generally speaking, use the earlier patent samples as the training set and the newer patent samples as the test set;

[0024] 3.6 Start the model training script and perform supervised fine-tuning training;

[0025] Step 4 specifically includes:

[0026] 4.1. Use the training model path model_path output in step 3.4 as the model prediction path, execute the prediction script, and obtain the category of the work order classification by inputting the work order problem;

[0027] Step 5 specifically includes:

[0028] 5.1. Associate the classification categories with the operation and maintenance project team and operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.1.

[0029] 5. According to claim 1, a method for intelligent order creation based on a large model of the operation and maintenance industry, characterized in that: step 6 specifically includes:

[0030] 6.1. Construct prompt text as "Based on the input text, summarize the existing information in concise language to generate a title. The title summary should be brief and semantically coherent, highlighting the theme of the article. Do not expand or explain the content. Please answer in Chinese. Enter text: + work_order_description", and concatenate the request json, where the json request format is '{"work_order_description":"Work order problem description"}'. Enter the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title;

[0031] Step 7 specifically includes:

[0032] 7.1 During the system integration process, a user interaction interface is designed to ensure that users can easily use the results provided by the model to make decisions and perform operations. The API interface is designed using Flask. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

[0033] An intelligent order creation system based on a large model of the operation and maintenance industry, the system consists of a work order collection module, a work order preprocessing module, a model fine-tuning module, a task execution module, a recognition module, a reasoning module and an integration module;

[0034] The work order collection module is used to collect the work order list of the work order system;

[0035] The work order preprocessing module is used to clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning;

[0036] The model fine-tuning module is used to select a pre-trained model, fine-tune it using a dataset, and output the fine-tuned model;

[0037] The task execution module is used to perform the task of predicting the classification of work orders using the fine-tuned model;

[0038] The identification module is used to automatically identify the operation and maintenance team and personnel in the area based on the model classification results;

[0039] The inference module is used to call the large model for inference and return the work order title;

[0040] Integration module, used to implement the system integration process.

[0041] Preferably, the work order collection module implements interaction with the work order system, develops a RESTful API interface, and obtains data from the work order system. The collected data is a list of work orders for the past three years, including work order title, work order problem description, work order category name, operation and maintenance project team, and operation and maintenance engineer name. The API interface returns json as the data exchange format.

[0042] Preferably, the work order preprocessing module deletes duplicate work orders. Duplicate records may exist in the data set, which may lead to deviations in the analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order is ensured; irrelevant character processing: the data set contains irrelevant characters such as line breaks and tabs, which have no practical significance for data analysis. Deleting these irrelevant characters improves the neatness and usability of the data set; and classification information completion: the work order classification name in the data set is not obtained because of errors in the data collection process or problems with the data source itself. In this case, the incomplete classification information is completed to ensure the classification accuracy and consistency of the data set.

[0043] Preferably, the model fine-tuning module specifically includes:

[0044] 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed;

[0045] 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, which is 442.

[0046] 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6;

[0047] 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints;

[0048] 3.5 Divide the data set organized in step 2, using 80% of the sample data as the training set and 20% of the sample data as the test set, taking into account the time distribution of the samples. Generally speaking, use the earlier patent samples as the training set and the newer patent samples as the test set;

[0049] 3.6 Start the model training script and perform supervised fine-tuning training;

[0050] The task execution module uses the training model path model_path output in step 3.4 as the model prediction path, executes the prediction script, and obtains the category of the work order classification by inputting the work order problem;

[0051] Identification module, including:

[0052] 5.1. Associate the classification categories with the operation and maintenance project team and operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.1.

[0053] Preferably, the reasoning module specifically includes:

[0054] 6.1. Construct prompt text as "Based on the input text, summarize the existing information in concise language to generate a title. The title summary should be brief and semantically coherent, highlighting the theme of the article. Do not expand or explain the content. Please answer in Chinese. Enter text: + work_order_description", and concatenate the request json, where the json request format is '{"work_order_description":"Work order problem description"}'. Enter the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title;

[0055] Integrated modules, including:

[0056] 7.1 During the system integration process, a user interaction interface is designed to ensure that users can easily use the results provided by the model to make decisions and perform operations. The API interface is designed using Flask. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The method and system for intelligent order creation based on the big model of the operation and maintenance industry proposed in the present invention realizes the automatic order creation for fault repair, equipment inspection and other links in the operation and maintenance industry by using big data and artificial intelligence technology, greatly improving work efficiency. In the past, manual processing of such work often required a lot of time, and was prone to problems such as missed orders and wrong orders. The present invention improves the processing speed through intelligent identification and automatic order creation, thereby improving the overall operation and maintenance efficiency.

[0059] Through deep learning of massive data, a rich knowledge base and decision-making model have been formed, which can provide comprehensive and accurate data support and decision-making reference for operation and maintenance managers. Managers can better grasp the operation and maintenance status, formulate more scientific and reasonable operation and maintenance strategies and plans, and improve the operation and maintenance management level and decision-making efficiency.

[0060] Based on the big model of the operation and maintenance industry, it can accurately identify and classify various faults and equipment problems, avoiding the subjectivity and uncertainty of manual judgment. Through intelligent analysis, the present invention can automatically match the corresponding solution and work order type according to the fault phenomenon, ensuring the accuracy of order creation and the pertinence of processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] For example, see Figure 1 The present invention provides a technical solution: a method for intelligent order creation based on a large model of the operation and maintenance industry, the method comprising the following steps:

[0064] Step 1: Collect the work order list from the work order system.

[0065] Step 2: Clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning.

[0066] Step 3: Select a pre-trained model, fine-tune it using the dataset, and output the fine-tuned model.

[0067] Step 4: Use the fine-tuned model to perform the task of predicting work order classification.

[0068] Step 5: Automatically identify the operation and maintenance group and personnel in the area through the model classification results.

[0069] Step 6: Call the large model for inference and return the work order title.

[0070] Step 7: Implement the system integration process.

[0071] Further, step 1 specifically includes the following:

[0072] 1.1 Implement interaction with the work order system, develop a RESTful API interface, and obtain data from the work order system. The collected data is a list of work orders in the past three years, including work order titles, work order problem descriptions, work order category names, operation and maintenance project teams, and operation and maintenance engineer names. The API interface returns json as the data exchange format.

[0073] Among them, the work order title is a summary of the work order content. The work order content can be quickly understood through the work order title, and the operation and maintenance personnel can understand the subject and issues involved in the work order in a short time. The work order titles include: failure to change specifications, routine inspection of a cloud center, application for batch modification of virtual machine passwords, firewall problem consultation, etc.

[0074] The work order problem description includes the requirements raised by the operation and maintenance personnel, system change requests, problem reports and solution processes.

[0075] The work order classification information classification name is a more detailed classification of the work order, such as according to different problem types, service types or urgency. For example, there are 442 classification categories such as instance creation request, function requirements, security inspection, tenant-side service request, New Year's Day security, equipment inspection, etc.

[0076] The operation and maintenance project team is responsible for handling work orders of different categories. When work orders are submitted, they will be directly assigned to the corresponding project team according to the type of work order. At the same time, when submitting a work order, you can also specify the name of the operation and maintenance engineer on duty to ensure that the work order can be handled promptly and effectively.

[0077] Furthermore, step 2 specifically includes:

[0078] 2.1 Deleting duplicate work orders: There may be duplicate records in the data set, which may lead to biased analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order can be ensured.

[0079] 2.2 Processing of irrelevant characters: The data set contains irrelevant characters such as line breaks and tabs, which have no practical significance for data analysis and may even affect the correctness of the analysis results. Removing these irrelevant characters will help improve the cleanliness and usability of the data set.

[0080] 2.3 Completion of classification information: The work order classification name in the data set is not obtained, which may be due to errors in the data collection process or problems with the data source itself. In this case, the incomplete classification information is completed to ensure the classification accuracy and consistency of the data set.

[0081] Furthermore, step 3 specifically includes:

[0082] 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed.

[0083] 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, and the parameter is 442.

[0084] 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6.

[0085] 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints.

[0086] 3.5 Divide the data set organized in step 2, and use 80% of the sample data as the training set and 20% of the sample data as the test set. At the same time, consider the time distribution of the samples. Generally speaking, use the patent samples from the earlier period as the training set and the newer patent samples as the test set. This can not only better evaluate the generalization performance of the model on new patents, but also reflect the dynamic changes in technological development.

[0087] 3.6 Start the model training script and perform supervised fine-tuning training.

[0088] Furthermore, step 4 specifically includes:

[0089] 4.1. Use the training model path model_path output in step 3.4 as the model prediction path, execute the prediction script, and obtain the category of the work order classification by inputting the work order problem.

[0090] Furthermore, step 5 specifically includes:

[0091] 5.1. Associating the classification categories with the operation and maintenance project team and the operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.1. For example, if a work order is classified as "center demand", then this work order will be assigned to the operation and maintenance project team "operation global center group", and the system will identify the operation and maintenance management personnel responsible for the project.

[0092] Furthermore, step 6 specifically includes:

[0093] 6.1. Construct the prompt text as "Based on the input text, use concise language to summarize the existing information and generate a title. The title summary should be brief and semantically coherent, highlight the theme of the article, and do not expand or explain the content. Please answer in Chinese. Enter text: + work_order_description", and concatenate the request json, where the json request format is '{"work_order_description":"Work order problem description"}'. Enter the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title.

[0094] Furthermore, step 7 specifically includes:

[0095] 7.1 During the system integration process, design a user interaction interface to ensure that users can easily use the results provided by the model to make decisions and perform operations. Use Flask to design the API interface. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

[0096] Embodiment 2

[0097] On the basis of the first embodiment, an intelligent order creation system based on the large model of the operation and maintenance industry is proposed, the system consists of a work order collection module, a work order preprocessing module, a model fine-tuning module, a task execution module, an identification module, a reasoning module and an integration module;

[0098] The work order collection module is used to collect the work order list of the work order system; realize the interaction with the work order system, develop a RESTful API interface, and obtain the data of the work order system. The collected data is the work order list of the past three years, including the work order title, work order problem description, work order category name, operation and maintenance project team, and operation and maintenance engineer name. The API interface returns json as the data exchange format.

[0099] The work order preprocessing module is used to clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning; delete duplicate work orders. There may be duplicate records in the data set, which may lead to deviations in the analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order is ensured; irrelevant character processing. The data set contains irrelevant characters such as line breaks and tabs. These characters have no practical significance for data analysis. Deleting these irrelevant characters improves the neatness and usability of the data set; completing classification information. The work order classification name in the data set is not obtained because of errors in the data collection process or problems with the data source itself. In this case, the incomplete classification information is completed to ensure the classification accuracy and consistency of the data set.

[0100] The model fine-tuning module is used to select a pre-trained model, fine-tune it using a data set, and output the fine-tuned model; specifically, it includes:

[0101] 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed;

[0102] 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, which is 442.

[0103] 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6;

[0104] 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints;

[0105] 3.5 Divide the data set organized in step 2, using 80% of the sample data as the training set and 20% of the sample data as the test set, taking into account the time distribution of the samples. Generally speaking, use the earlier patent samples as the training set and the newer patent samples as the test set;

[0106] 3.6 Start the model training script and perform supervised fine-tuning training.

[0107] The task execution module is used to use the fine-tuned model to perform the task of predicting the work order classification. The training model path model_path output in step 3.4 is used as the model prediction path, and the prediction script is executed. By inputting the work order problem, the category of the work order classification is obtained.

[0108] The identification module is used to automatically identify the operation and maintenance group and personnel in the area through the model classification results; specifically, it includes:

[0109] 5.1. Associate the classification categories with the operation and maintenance project team and operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.1.

[0110] The inference module is used to call the large model inference to return the work order title; specifically includes:

[0111] 6.1. Construct the prompt text as "Based on the input text, use refined language to summarize the existing information and generate a title. The title summary should be brief, semantically coherent, and highlight the theme of the article. Do not expand or explain the content. Please answer in Chinese. Enter the text: + work_order_description", and concatenate the request json, where the json request format is '{"work_order_description":"Work order problem description"}'. Input the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title.

[0112] Integration module, used to implement the system integration process; specifically includes:

[0113] 7.1 During the system integration process, a user interaction interface is designed to ensure that users can easily use the results provided by the model to make decisions and perform operations. The API interface is designed using Flask. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

[0114] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent order creation based on a large model of the operation and maintenance industry, characterized by: The method comprises the following steps: Step 1: Collect the work order list of the work order system; Step 2: Clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning; Step 3: Select a pre-trained model, fine-tune it using the data set, and output the fine-tuned model; Step 4: Use the fine-tuned model to perform the task of predicting the work order classification; Step 5: Automatically identify the operation and maintenance group and personnel of the area through the model classification results; Step 6: Call the large model for inference and return the work order title; Step 7: Implement the system integration process.

2. According to claim 1, a method for intelligent order creation based on a large model of the operation and maintenance industry is characterized by: The specific operations of step 1 include: Realize interaction with the work order system, develop a RESTful API interface, and obtain data from the work order system. The collected data is a list of work orders in the past three years, including work order title, work order problem description, work order category name, operation and maintenance project team, and operation and maintenance engineer name. The API interface returns json as the data exchange format.

3. According to claim 1, a method for intelligent order creation based on a large model of the operation and maintenance industry, characterized in that: The specific operations of step 2 include: Delete duplicate work orders. There are duplicate records in the data set, which lead to deviations in the analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order is ensured; irrelevant character processing. The data set contains irrelevant characters such as line breaks and tabs. These characters have no practical significance for data analysis. Deleting these irrelevant characters improves the neatness and usability of the data set; completing classification information. The work order classification name in the data set is not obtained because of errors in the data collection process or problems with the data source itself. In this case, complete the incomplete classification information to ensure the classification accuracy and consistency of the data set.

4. According to claim 1, a method for intelligent order creation based on a large model of the operation and maintenance industry is characterized by: Step 3 specifically includes: 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed; 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, which is 442. 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6; 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints; 3.5 Divide the data set organized in step 2, using 80% of the sample data as the training set and 20% of the sample data as the test set. Considering the time distribution of the samples, generally speaking, the earlier patent samples are used as the training set and the newer patent samples are used as the test set; 3.6 Start the model training script and perform supervised fine-tuning training; Step 4 specifically includes: 4.

1. Use the training model path model_path output in step 3.4 as the model prediction path, execute the prediction script, and obtain the category of the work order classification by inputting the work order problem; Step 5 specifically includes: 5.

1. Associate the classification categories with the operation and maintenance project team and operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.

1.

5. According to claim 1, a method for intelligent order creation based on a large model of the operation and maintenance industry is characterized by: Step 6 specifically includes: 6.

1. Construct the prompt text as "Based on the input text, summarize the existing information in concise language to generate a title. The title summary should be brief and semantically coherent, highlighting the theme of the article. Do not expand or explain the content. Please answer in Chinese. Enter the text: + work_order_description", and concatenate the request json. The json request format is '{"work_order_description":"Work order problem description"}'. Input the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title; Step 7 specifically includes: 7.1 During the system integration process, a user interaction interface is designed to ensure that users can easily use the results provided by the model to make decisions and perform operations. The API interface is designed using Flask. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

6. An intelligent order creation system based on an operation and maintenance industry big model according to the intelligent order creation method based on an operation and maintenance industry big model according to any one of claims 1 to 5, characterized in that: The system consists of a work order collection module, a work order preprocessing module, a model fine-tuning module, a task execution module, a recognition module, a reasoning module and an integration module; The work order collection module is used to collect the work order list of the work order system; The work order preprocessing module is used to clean, integrate and standardize the collected work order list to form the data set required for model fine-tuning; The model fine-tuning module is used to select a pre-trained model, fine-tune it using a dataset, and output the fine-tuned model; The task execution module is used to perform the task of predicting the classification of work orders using the fine-tuned model; The identification module is used to automatically identify the operation and maintenance team and personnel in the area based on the model classification results; The inference module is used to call the large model for inference and return the work order title; Integration module, used to implement the system integration process.

7. The intelligent order creation system based on the big model of the operation and maintenance industry according to claim 6 is characterized by: The work order collection module realizes interaction with the work order system, develops a RESTful API interface, and obtains data from the work order system. The collected data is a list of work orders in the past three years, including work order titles, work order problem descriptions, work order category names, operation and maintenance project teams, and operation and maintenance engineer names. The API interface returns json as the data exchange format.

8. The intelligent order creation system based on the operation and maintenance industry big model according to claim 6 is characterized by: The work order preprocessing module deletes duplicate work orders. Duplicate records may exist in the data set, which may lead to deviations in the analysis results. By identifying and deleting these duplicate work orders, the uniqueness of each work order is ensured; irrelevant character processing: the data set contains irrelevant characters such as line breaks and tabs, which have no practical significance for data analysis. Deleting these irrelevant characters improves the neatness and usability of the data set; classification information completion: the work order classification name in the data set is not obtained because of errors in the data collection process or problems with the data source itself. In this case, the incomplete classification information is completed to ensure the classification accuracy and consistency of the data set.

9. The intelligent order creation system based on the big model of the operation and maintenance industry according to claim 6 is characterized by: The model fine-tuning module specifically includes: 3.1 Use the BERT algorithm for training, set the pre-trained model train_model to BERT-base-chinese, and set the model path model_path, which is where the model parameters and structure are saved after the model training is completed; 3.2 Set the model parameters. batch_size is the number of samples used for gradient update in one training process, which is set to 32. number_of_epochs is the number of iterations for model training, which is set to 5. num_labels is the category of work order classification, which is 442. 3.3 Use AdamW optimizer to optimize the parameters of the BERT model, design a fine-tuning strategy, set the initial learning rate to 5e-5, the weight decay coefficient to 0.01, the decay rates β1 and β2 of the first-order and second-order moment estimates to 0.9 and 0.999, and the small numerical stability constant ε, usually set to 1e-6; 3.4 Set up model checkpoints and use TensorFlow's tf.train.Checkpoint API to manage model checkpoints; 3.5 Divide the data set organized in step 2, using 80% of the sample data as the training set and 20% of the sample data as the test set. Considering the time distribution of the samples, generally speaking, the earlier patent samples are used as the training set and the newer patent samples are used as the test set; 3.6 Start the model training script and perform supervised fine-tuning training; The task execution module uses the training model path model_path output in step 3.4 as the model prediction path, executes the prediction script, and obtains the category of the work order classification by inputting the work order problem; Identification module, including: 5.

1. Associate the classification categories with the operation and maintenance project team and operation and maintenance management personnel. Determine the corresponding project team and the list of operation and maintenance management personnel through the work order classification categories obtained in step 4.

1.

10. The intelligent order creation system based on the operation and maintenance industry big model according to claim 6 is characterized by: The reasoning module specifically includes: 6.

1. Construct the prompt text as "Based on the input text, summarize the existing information in concise language to generate a title. The title summary should be brief and semantically coherent, highlighting the theme of the article. Do not expand or explain the content. Please answer in Chinese. Enter the text: + work_order_description", and concatenate the request json. The json request format is '{"work_order_description":"Work order problem description"}'. Input the above concatenated content into the large model Baichuan2-13B-Chat, and wait for the large model inference to complete and then return the work order title; Integrated modules, including: 7.1 During the system integration process, a user interaction interface is designed to ensure that users can easily use the results provided by the model to make decisions and perform operations. The API interface is designed using Flask. Users only need to enter the work order problem description, and the system will automatically call the API interface to generate the work order classification and work order title, and automatically assign the work order to the corresponding operation and maintenance personnel.

Images